Adopt AI Study

Final study report

Written by
EUROPEAN COMMISSION
Directorate-General for Communications Networks, Content and Technology (CNECT)
Unit: Artificial Intelligence Policy Development and Coordination (CNECT.A.2)
Contact: Martin Ulbrich
E-mail: martin.ulbrich@ec.europa.eu
European Commission
B-1049 Brussels
 EUROPEAN COMMISSION

Adopt AI Study
Final study report

Directorate-General for Communications Networks, Content and Technology (CNECT)

Artificial Intelligence Policy Development and Coordination (CNECT.A.2)
 Adopt AI Study

Europe Direct is a service to help you find answers

to your questions about the European Union.
Freephone number (*):

00 800 6 7 8 9 10 11
(*) The information given is free, as are most calls (though some operators, phone boxes or hotels
may charge you).

LEGAL NOTICE
This document has been prepared for the European Commission however it reflects the views only of the
authors, and the Commission cannot be held responsible for any use which may be made of the information
contained therein.
More information on the European Union is available on the Internet (http://www.europa.eu).
Luxembourg: Publications Office of the European Union, 2024
KK-02-24-570-EN-N
ISBN 978-92-68-15940-8
DOI 10.2759/22251
© European Union, 2024
Reproduction is authorised provided the source is acknowledged.
 Adopt AI Study

Table of contents

Executive summary ........................................................................................... 1

Introduction ...................................................................................................... 5
1 Chapter 1: State of play/ Scoping report ......................................................... 6
1.1 Introduction.......................................................................................... 6
1.2 State of play of the public procurement on AI ........................................... 6
1.2.1 Artificial Intelligence initiatives at the European Union level .................. 6
1.2.2 Artificial Intelligence initiatives at the national level ............................10
1.2.3 Overview of AI projects in the public sector .......................................14
1.3 Barriers to and benefits of the adoption of AI in the public and private sectors
18
1.3.1 Challenges ....................................................................................18
1.3.1.1 Procurement process challenges .......................................................18
1.3.1.2 Data challenges .............................................................................19
1.3.1.3 AI technology challenges .................................................................20
1.3.1.4 Organisational capacity challenges ...................................................21
1.3.1.5 Sectoral and AI technologies-related challenges .................................22
1.3.1.6 AI technological considerations of the challenges ...............................24
1.3.2 Benefits ........................................................................................24
1.3.2.1 Tools and ways to support the achievement of the benefits .................26
1.4 Economic analysis ................................................................................27
1.4.1 An introduction to AI adoption .........................................................29
1.4.2 Budget analysis of public procurement of AI ......................................32
1.4.3 Market analysis on sector-level dynamics and AI technologies uptake ...35
1.4.4 Uptake of AI technologies by AI typology ..........................................43
1.5 Final selection of public sector policy areas ..............................................51
2 Chapter 2: Qualitative and Quantitative analysis of key sectors ........................56
2.1 Introduction.........................................................................................56
2.2 eGovernment .......................................................................................58
2.2.1 Challenges and AI solutions for e-government ...................................60
2.2.2 Digital value chain of e-government .................................................66
2.2.3 Main drivers and barriers.................................................................68
2.2.4 Case Study: Mercè - citizen science for better urban life (Spain) ..........73
2.2.5 Economic Analysis ..........................................................................75
2.3 Mobility ...............................................................................................77
2.3.1 Challenges and AI solutions/applications in the mobility sector ............80
2.3.2 Value chain analysis .......................................................................86
2.3.3 Main drivers and barriers.................................................................94
2.3.4 Case Study: The Mobility Data Space (Germany) ...............................99
2.3.5 Economic analysis ........................................................................ 102
2.4 Health ............................................................................................... 103
2.4.1 Challenges and AI solutions/applications in the health sector ............. 105
2.4.2 Value chain analysis ..................................................................... 110
2.4.3 Main drivers and barriers............................................................... 116
2.4.4 Case study: Developing the next generation of bionic prostheses for low-
limb amputees (Belgium) ........................................................................... 120
2.4.5 Economic analysis ........................................................................ 122
2.5 Education .......................................................................................... 123
 Adopt AI Study

2.5.1 Challenges and corresponding AI solutions/applications in the education

sector 126
2.5.2 Education value chain analysis ....................................................... 135
2.5.3 Main drivers and barriers............................................................... 140
2.5.4 Case study: Educating the world on AI: The elements of AI MOOC
(Finland) 145
2.5.5 Economic analysis ........................................................................ 147
2.6 Conclusions ....................................................................................... 149
3 Chapter 3: Consultation of stakeholders’ analytical, comparative report ........... 153
3.1 Introduction....................................................................................... 153
3.2 Health workshop ................................................................................ 153
3.2.1 Exercise 1: Challenges affecting the uptake of AI by the public sector . 153
3.2.2 Exercise 2: Measures to increase the adoption of AI ......................... 156
3.2.3 Exercise 3: Examples of cooperation between the public sector and AI
providers and success stories ..................................................................... 158
3.3 E-government workshop ..................................................................... 159
3.3.1 Exercise 1: Challenges affecting the uptake of AI by the public sector . 159
3.3.2 Exercise 2: Measures to increase the adoption of AI ......................... 162
3.3.3 Exercise 3: Examples of cooperation between the public sector and AI
providers and success stories ..................................................................... 163
3.4 Mobility (transport) workshop .............................................................. 164
3.4.1 Exercise 1: Challenges affecting the uptake of AI by the public sector . 164
3.4.2 Exercise 2: Measures to increase the adoption of AI ......................... 165
3.4.3 Exercise 3: Examples of cooperation between the public sector and AI
providers and success stories ..................................................................... 166
3.5 Education workshop ............................................................................ 166
3.5.1 Exercise 1: Challenges affecting the uptake of AI by the public sector . 167
3.5.2 Exercise 2: Measures to increase the adoption of AI ......................... 169
3.5.3 Exercise 3: Examples of cooperation between the public sector and AI
providers and success stories ..................................................................... 170
3.6 Policy workshop ................................................................................. 171
3.6.1 Exercise 1: Identifying solutions to challenges affecting the uptake of AI
by the public sector ................................................................................... 171
3.6.2 Exercise 2: Exploring solutions and opportunities for overcoming barriers
176
3.7 Online survey .................................................................................... 179
3.7.1 Sectoral comparison of challenges in the online survey ..................... 181
3.7.2 Comparison of policy recommendations in the online survey .............. 190
3.8 Conclusion ......................................................................................... 194
4 Chapter 4: Policy recommendations on the uptake of AI in the public sector ..... 195
4.1 Introduction....................................................................................... 195
4.2 Comparative assessment of policy recommendations .............................. 195
5 Conclusion ............................................................................................... 230
6 Annexes .................................................................................................. 239
6.1 Chapter 1 Annex ................................................................................ 239
6.1.1 Methodology ................................................................................ 239
6.1.2 Public sector areas ....................................................................... 240
 Adopt AI Study

6.1.3 Mapping of public sector areas versus economic sectors .................... 242
6.1.4 AI technologies ............................................................................ 243
6.1.5 List of initiatives ........................................................................... 244
6.1.6 List of projects ............................................................................. 247
6.2 Chapter 2 Annex ................................................................................ 260
6.2.1 Definitions ................................................................................... 260
6.2.2 Methodology ................................................................................ 263
6.2.3 Data sources and collection ........................................................... 266
6.2.4 Policy overview ............................................................................ 267
6.3 Chapter 3 & 4 Annexes ....................................................................... 284
6.3.1 Methodology ................................................................................ 284
6.3.2 Workshop materials ...................................................................... 285
6.4 References ........................................................................................ 287
7 HOW TO OBTAIN EU PUBLICATIONS............................................................ 292
7.1 Free publications: ............................................................................... 292
7.2 Priced publications: ............................................................................ 292
7.3 Priced subscriptions: ........................................................................... 292

Table of Figures

Figure 1 Timeline of EU AI initiatives .................................................................... 7

Figure 2 Actions of the Coordinated Plan on AI (2021) relevant to public procurement 9
Figure 3 Types of national initiatives that include AI ..............................................10
Figure 4 Types of national initiatives that focus on AI ............................................11
Figure 5 Classification of public sector priority areas for AI ....................................12
Figure 6 AI procurement national approaches ......................................................13
Figure 7 Administrative levels of AI projects.........................................................14
Figure 8 Classification of public sector areas of AI projects.....................................16
Figure 9 AI projects' typologies ..........................................................................17
Figure 10 Macro categories of challenges .............................................................18
Figure 11 External obstacles to the adoption of AI ................................................23
Figure 12 Internal obstacles to the adoption of AI .................................................23
Figure 13 Barriers to the adoption of AI by technology ..........................................24
Figure 14 Macro categories of benefits ................................................................25
Figure 15 Likelihood of digital transformation success, 2021 ..................................28
Figure 16 Enterprises using AI technologies by Member State, 2021 .......................30
Figure 17 Forecasted job landscape, 2020 ...........................................................31
Figure 18 Human capital dimension by Member State, 2021 ..................................32
Figure 19 Average spending on AI projects classified by public sector policy area .....33
Figure 20 Average spending on AI projects classified by AI typology .......................34
Figure 21 AI adoption per industry, 2021 .............................................................36
Figure 22 Digital intensity per sector, 2021 (very high digital intensity index) ..........38
 Adopt AI Study

Figure 23 Digital intensity per sector, 2021 (high digital intensity index) .................38
Figure 24 Digital intensity score per sector, 2021 .................................................39
Figure 25 R&D business expenditure, 2019 ..........................................................41
Figure 26 R&D investment per sector, EU, 2020 ...................................................42
Figure 27 Correspondence of AI typologies from AI Watch to Market-based
classification ....................................................................................................43
Figure 28 Approach for selection of sectors based on AI technologies analysis ..........44
Figure 29 Forecasted market size by AI technologies, 2020-2026 ...........................44
Figure 30 Most widely used AI-powered solutions, 2019 ........................................45
Figure 31 Hype Cycle for Artificial Intelligence, 2021 .............................................46
Figure 32 Forecasted Predictive Analytics market share, 2019-2027........................48
Figure 33 Natural Language Processing market share, 2019 ...................................49
Figure 34 Forecasted Speech Recognition market share, 2018-2026 .......................49
Figure 35 Forecasted US Voice Recognition market share, 2014-2025 .....................50
Figure 36 Machine Learning market share, 2019 ...................................................50
Figure 37 Criteria for selection of the key sectors .................................................52
Figure 38 Four key sectors .................................................................................53
Figure 39 Policy breakdown per mode taken from the mapping exercise ..................59
Figure 40 E-government and AI policies in the EU-27 ............................................60
Figure 41 Summary of challenges and AI solutions for digital public services ............61
Figure 42: The policy stages heuristic .................................................................63
Figure 43: Procurement process of AI .................................................................68
Figure 44 Strengths, Weaknesses, Opportunities and Threats of the uptake and public
procurement of AI-technologies in the e-government sector ..................................72
Figure 45 Member states typology ......................................................................76
Figure 46 Mobility and AI – breakdown by policy mode ..........................................79
Figure 47 Mobility and AI – breakdown by Member State .......................................80
Figure 48 Summary of challenges and AI solutions for mobility ..............................81
Figure 49 Transportation and mobility – value chain .............................................86
Figure 50 Summary of conclusions for value chain analysis. ...................................87
Figure 51 Strengths, Weaknesses, Opportunities and Threats of the uptake and public
procurement of AI-technologies in the mobility and transport sector .......................99
Figure 52 Mobility & AI: Member state classification (Malta and Cyprus are categorized
as “planners”) ................................................................................................ 102
Figure 53 Health and AI – breakdown by policy type ........................................... 104
Figure 54 Health and AI policies – breakdown by Member State ........................... 105
Figure 55 Summary of challenges and AI solutions for health ............................... 105
Figure 56 Health sector value chain .................................................................. 111
Figure 57 Summary of key intervention areas for public authorities and public
procurement along the health value chain. ........................................................ 116
 Adopt AI Study

Figure 58 Strengths, Weaknesses, Opportunities and Threats of the uptake and public
procurement of AI-technologies in the healthcare sector...................................... 120
Figure 59 Health & AI – Member state classification ............................................ 122
Figure 60 Regression table - Health .................................................................. 123
Figure 61: Policies per education level ............................................................... 125
Figure 62: Breakdown per policy type ............................................................... 125
Figure 63 Summary of challenges and AI solutions for education .......................... 126
Figure 64 Education value chain ....................................................................... 135
Figure 65 Comparison of venture capitals in Education Technology for China, the US
and Europe .................................................................................................... 138
Figure 66 Drivers and barriers to the uptake of AI in education ............................ 139
Figure 67 Strengths, Weaknesses, Opportunities and Threats of the uptake and public
procurement of AI-technologies in the education sector ....................................... 145
Figure 68 Respondents’ affiliation ..................................................................... 179
Figure 69 Respondents’ country of origin ........................................................... 180
Figure 70 Respondents’ sectors ........................................................................ 181
Figure 71 Task 1 Methodology .......................................................................... 240
Figure 72 Public sector policy areas .................................................................. 241
Figure 73 Full overview of public sector policy areas ........................................... 241
Figure 74 Mapping of public sector areas vis-à-vis NACE codes ............................. 242
Figure 75 AI general classification ..................................................................... 243
Figure 76 AI Watch AI typologies ...................................................................... 244
Figure 77 The 4 sectors studied in the chapter ................................................... 260
Figure 78 Summary of the structural factors affecting the adoption of AI in public
services taken from an AI Watch study.............................................................. 263

Table of Tables

Table 1 European Union initiatives on AI ............................................................... 7

Table 2 Sector importance vis-a-vis studied indicators ..........................................55
Table 3 PESTEL analysis of the uptake of AI technologies in public procurement in the
e-government sector .........................................................................................69
Table 4 Regression table – E-government ............................................................77
Table 5 PESTEL analysis of the uptake of AI technologies in public procurement in the
mobility and transport sector .............................................................................95
Table 6 PESTEL analysis of the uptake of AI technologies in public procurement in the
Healthcare sector ........................................................................................... 117
Table 7 PESTEL analysis of the uptake of AI technologies in public procurement in the
education sector ............................................................................................. 141
Table 8 Comparison of procurement process challenges between sectors. .............. 183
Table 9 Comparison of data challenges between sectors. ..................................... 185
 Adopt AI Study

Table 10 Comparison of AI technology challenges between sectors. ...................... 187

Table 11 Comparison of organisational capacity challenges. ................................. 189
Table 12 Comparison of recommendations ......................................................... 192
Table 13 Ranking of policy recommendations ..................................................... 229
Table 14 Typologies of AI public sector activity from the literature ........................ 263
 Adopt AI

Executive summary
Artificial Intelligence (AI) is developing fast and the benefits of adopting AI are widely
recognized. AI applications can contribute to better public services, e.g. by improving
citizen-government interaction, enabling smarter analytical capabilities or improving
efficiency across public-sector domains and supporting democratic processes. The use
of AI systems can bring benefits across all key public-sector activities. Through early
adoption of AI, the public sector can be the first mover in adopting AI that is secure,
trustworthy and sustainable. Public procurement is one of the key measures that have
a strong potential to facilitate AI adoption and help stimulate demand and offer
trustworthy and secure AI technologies in Europe.
AI applications can contribute to better public services for example by improving citizen-
government interaction, enabling smarter analytical capabilities, delivering shorter
feedback loops or improving efficiency. However, the available evidence suggests that
there is still considerable untapped potential for the uptake and public procurement of
AI. Therefore, one of the focus areas for the European Commission is to develop and
support European actions to accelerate the process of AI uptake and strategic,
sustainable and ethical public procurement of AI systems by the public sector in the EU
and thus maximise the impact and benefits of human-centric and trustworthy AI.
Against this background, DG CNECT commissioned a study on how best to support public
procurement of AI systems in the European Union and help to transform public
procurement processes themselves. The main purpose and objective of the
commissioned study are to support the European Commission with evidence and
operational recommendations on how public procurement of AI systems can be
facilitated.
The study includes the State of play/scoping report (presented in chapter 1), the
Assessment of the key sectors (presented in chapter 2), Consultation and assessment
of the results of the consultation of stakeholders (presented in chapter 3) and the
Assessment of policy options and recommendations (presented in chapter 4).
Chapter 1: State of play/ Scoping report
The study opens with a scoping section, which aims to identify the state of play on the
uptake and public procurement of AI technologies by the public sector in the EU and to
select four priority key sectors that are most ready for large-scale deployment of AI
technologies. This is achieved by analysing available literature, evidence, and data and
is presented through four main sections. The first section (State of play of the public
procurement of AI) provides an overview of the political and legislative initiatives on AI
at the European level as well as a focus on the AI strategies put in place at the national
level throughout the EU. The second section (Challenges and benefits in public
procurement of AI) identifies the main challenges and benefits faced by Member States’
governments when considering the public procurement of AI technologies. The third
section (Economic analysis) analyses the economic potential of further investments in
AI based on existing quantitative data on public procurement, market trends and
maturity of AI technologies. Finally, the selection of the four key sectors that are most
ready for large-scale deployment of AI technologies is provided in the fourth and final
section and is based on the information gathered and analysis conducted throughout
the three other sections of the scoping activities.
More specifically, the final selection of the sectors is based on three research questions,
further detailed in the methodological note, on identifying (1) the main gaps and
obstacles that hinder public procurement and uptake of AI, as well as (2) the benefits
and added value that can be generated by a wider uptake and procurement of AI
systems in the EU and the main sectors/areas, and (3) AI technologies with high
economic potential (e.g. technological readiness, high growth performance) where
accelerated public procurement could bring the highest EU added value. The approach
taken to tackle the research questions consists of four main steps. Firstly, providing a

1
 Adopt AI

state of play focused on public procurement by mapping and analysing, both

qualitatively and quantitatively, national initiatives on AI. Secondly, conducting a
challenges and benefits analysis in public procurement by analysing and mapping
identified challenges and benefits for the public procurement of AI and complementing
it with findings of the private sector. Thirdly, performing an economic analysis by
quantitatively mapping current investments in AI projects and future national
projections as well as analysing innovation indicators in selected economic sectors and
market indicators of AI typologies. Lastly, selecting the four priority areas based on the
above steps and triangulating the findings based on the specific criteria of identified
trends in governmental initiatives, projects, and the market, maturity of AI
technologies, and the economic potential of the sectors that will create European added
value for the public procurement of AI.
The scoping concludes that the four key sectors that are ready for the large-scale
deployment of AI are: Health, Mobility, E-Government and Education. This is based on
the findings of the qualitative and quantitative analyses of governmental AI initiatives
and the economic analysis. The former provides, on the one hand, an overview of
governments’ priority policy areas for the adoption of AI, and on the other hand, the
current state of play of adopted AI policy areas and technologies. The economic analysis
evaluates the digital maturity levels of sectors by looking at the following indicators: AI
adoption levels, digital intensity, and R&D intensity of different sectors. Along the same
line, the economic potential of AI technologies was also assessed based on four key
indicators: market growth, adoption rates, labour productivity impact, and employment
generation.
Chapter 2: Qualitative and quantitative analysis of key sectors
This chapter analyses the dynamics of the adoption and use of Artificial Intelligence (AI)
in four key public sectors in the European Union: eGovernment, mobility, healthcare
and education. Leading the way in terms of AI maturity, these central sectors are
already undergoing significant change as a result of the introduction of this paradigm-
changing technology. Using the insights from these policy areas, acquired through a
multi-method research design, this chapter aims to increase the understanding of the
uptake of AI by public organisations. This is paramount if the European Union is to reach
its goal of becoming a global leader in the field of AI and other policy goals such as the
twin-transition.
In eGovernment and general public services, AI´s potential to considerably improve
operational efficiency, and public services and allow more open government has seen
many administrations begin to fund, develop and use AI solutions. In addition to a
burgeoning GovTech space, chatbots and virtual assistants are now commonplace in
many member states at both the national and regional levels. While much of the
innovation thus far has focused mostly on upgrading established governmental
practices, a continued investment could see AI radically redesign existing vertical and
hierarchical-oriented administrative structures for the benefit of everyone. This is
contingent on well-informed public sector actors with expertise in AI adoption and the
mitigation of risks such as a lack of trust and understanding of the technology and its
effects.
In the mobility and transport sector, one of the most profoundly affected by the COVID-
19 pandemic, AI is perhaps the most pervasive in the public consciousness. AI has the
potential to greatly increase efficiency and safety as well as transform everyday
transportation with autonomous vehicles (AV) and smart traffic systems. The sector has
a considerable political imperative for governments given the centrality of transport in
everyday life and the fight against climate change but a lack of harmonized rules across
Europe and siloed focus on data and regulation has meant that the majority of
innovation and use of AI has taken place in the private sector. With a proven economic
track record from the success of the AV market and the environmental benefits of
improved transport systems and urban space, mobility will continue to be at the
forefront of AI development provided concerns around safety remain central.

2
 Adopt AI

In the health sector, government-funded research of AI is still the dominant paradigm

of public sector support although the use of AI solutions in healthcare system
management and frontline applications is growing. AI has myriad potential uses in the
health sector at every step in the value chain from R&D in the field of pharmaceuticals
to managing hospitals and healthcare systems and delivering care to the patients
themselves. Harnessing AI in health will assist governments in tackling the long-term
structural problems of ageing populations and protect against rising costs, especially
through automating administrative tasks, currently a common focus area for AI health
policy. As with other sectors, effective implementation will increase the quality and
accessibility of healthcare though it is vital to guarantee thorough security and
protection of data and that the human side of the provision of healthcare is not lost.
Finally, in education, AI has emerged as both a subject of learning and an instrument
to improve its delivery. AIEd and EdTech promise to transform education through
student-facing AI, enhancing teaching and improving the operation of education
systems. However, AI use cases in education are not yet widespread as a result of a
lack of tried and tested solutions and reticence from educators to adopt the technology.
Nevertheless, the current focus of most public policy throughout Europe focuses on
educating citizens on the use and potential of AI. Not limited to technical higher
education, the key link between employment and education has seen many
governments invest in teaching programmes on AI from early years to adult learning.
Chapter 3: Consultation of stakeholders’ analytical, comparative report
This chapter presents a comparative analysis of stakeholder consultations undertaken
as part of this study. The consultations included four sectoral workshops, dedicated to
exploring the challenges and measures in the uptake of AI by the public sector in the
health, e-government, mobility/transport and education sectors, and an online survey,
dedicated to assessing the challenges and recommendations in the uptake of AI in the
public sector.
The stakeholder consultations undertaken as part of this study show that the main
challenges in the uptake of AI centres on the procurement process, data, AI technology
and organisation.
The most significant procurement process challenges are related to burdensome
administrative requirements, lack of clarity from the public sector regarding their
needs/demands, an emphasis on price/cost-savings over service quality and non-
financial benefits, unclear regulatory requirements, and the complexity of writing
technical specifications.
The most significant data challenges are related to unsatisfactory sharing of data across
organisational boundaries, insufficient access to large volumes of high-quality data, lack
of data to understand where AI is needed/ best suited, underdeveloped data
governance, and lack of clear “data ownership/ data sovereignty”.
The most significant AI technology challenges are lack of transparency in AI systems'
decision support/making processes, difficulty to inspect and assess an AI solution before
their actual deployment, the potential for biases/discrimination within the systems,
difficulty in establishing liability and responsibility for the AI system, lack of limited
regulatory spaces ("sandboxes") for experimenting with AI solutions and monitoring
their impacts, and high requirements for the explainability of AI solutions.
The most significant organisational challenges are a lack of human resources for
managing the system, lack of political support, lack of understanding of the
capabilities/benefits of AI solutions, lack of human resources for procuring the system,
system complexity and lack of single-entry point, and lack of system interoperability.
The comparative analysis shows that these challenges are not significant across the
sectors. These challenges were raised and discussed by participants in all the
workshops, irrespective of the sector. In addition, the differences between the sectors
in the online survey were minor and cannot be considered significant.

3
 Adopt AI

Chapter 4: Policy recommendations on the uptake of AI in the public sector

This chapter explores the recommendations to increase the uptake and public
procurement of AI. It includes the results from the policy workshop dedicated to
identifying and exploring policy solutions and recommendations for the uptake of AI in
the public sector, and an online survey, dedicated to assessing recommendations for
the uptake of AI in the public sector.
The chapter also provides a comparative assessment of the recommendations. It
provides an assessment of the most important recommendations on a scale of
low/medium/high based on the criteria of Effectiveness (the extent to which
recommendations achieve their goals and the level of benefits they provide), Feasibility
(the extent to which recommendations can be implemented and whether they have
significant limits), Efficiency (the extent to which recommendations provide value for
money and if they require significant resources), and EU added value (the extent to
which recommendations provide benefits at a European level and the Single Market and
which level of implementation, European or Member State, is the most appropriate,
keeping in mind the principle of subsidiarity).
The chapter triangulates the recommendations from the policy workshop and the
online survey to arrive at the combined ranking. The recommendations are:
1. Increase funding and resources for AI in the public sector.
2. Reduce bias within AI and data sources.
3. Encourage coordination of AI procurement strategies.
4. Increase clarity and harmonization around cross-border data flows.
5. Promote alignment between industry and public sector expectations.
6. Establish a clear AI regulatory framework.
7. Promote the integration of new AI technologies and services into existing
systems in education, health and transport.
8. Promote interoperability, open data and data sharing.
9. Build trust in AI solutions through transparency and accountability.
10. Harmonize EU regulations to promote human-centric and trustworthy AI-
enabled public services.
11. Focus on long-term implementation in the use of AI in the public sector.
12. Develop dedicated AI-enabled solutions based on co-creation approaches.
13. Create a European marketplace for GovTech solutions in support of public
sector digital transformation.
14. Strengthen the role of the EU Artificial Intelligence Observatory.
15. Promote the development of sustainable AI.
The report concludes with a brief discussion of each recommendation, noting the current
EU initiatives and the potential policy actions that the European Commission can take
to address them.

4
 Adopt AI

Introduction
ICF, Carsa and Wavestone were awarded by DG CNCT to conduct the “Study for the
Adopt AI Programme”. The main purpose and objective of the commissioned study are
to support the European Commission with evidence and operational recommendations
on how public procurement of AI systems can be facilitated
This final report presents the results of the study. It consists of the following chapters:
 Chapter 1: State of play/ Scoping report. This chapter provides an analysis
of the state of play of the public procurement on AI, barriers to and benefits of
the adoption of AI in the public and private sectors, economic analysis of and
the selection of the four sectors for further analysis: e-government, mobility,
health and education.
 Chapter 2: Qualitative and quantitative analysis of key sectors. This
chapter provides a deep analysis of the four selected sectors (e-government,
mobility, health and education), consisting of challenges and solutions for AI in
the sectors, digital value chains, main drivers and barriers to the uptake of AI in
the sectors, case studies of selected projects, and economic analysis of the
impacts of AI on the sectors.
 Chapter 3: Consultation of stakeholders’ analytical, comparative report.
This chapter provides the results and the comparative analysis of four dedicated
sectoral workshops (e-government, mobility, health and education) and an
online survey conducted as part of this study.
 Chapter 4: Policy recommendations on the uptake of AI in the public
sector. This chapter presents the results of a dedicated policy workshop and an
online survey on policy recommendations conducted as part of this study. The
chapter provides a comparative assessment of policy recommendations,
considering effectiveness, efficiency, feasibility and EU-added value. It
concludes with a brief discussion of each recommendation, noting the current
EU initiatives and the potential policy actions that the European Commission
can take to address them.
The annexes include detailed methodologies for each chapter, outlining the data
collection efforts, approaches to data analysis, definitions, lists of projects and
initiatives, and references.

5
 Adopt AI

1 Chapter 1: State of play/ Scoping report

1.1 Introduction
The chapter aims to provide a review and analysis of available qualitative and
quantitative data on the public procurement of AI technologies in the EU, further
complemented by market research on priority economic sectors and AI typology
uptake.
The chapter consists of the following four sections, further detailed in the
Methodological note:
1. State of play of the public procurement of AI
2. Challenges and benefits in public procurement of AI
3. Economic analysis
4. Final selection of four priority public sector policy areas
1.2 State of play of the public procurement on AI
This section provides a brief contextual overview of political and legislative initiatives
on AI at the EU level; after which the focus will turn to the national level and the AI
strategies put forth throughout the Member States.

It must be noted that the information available through desk research does not cover
all countries to the same depth, therefore we will showcase the countries with the
highest level of information available.

1.2.1 Artificial Intelligence initiatives at the European Union level

The European Union aims to heavily invest in Artificial Intelligence (AI) to allow all
industries to benefit from it. In the period 2014-2017, around EUR 1.1 billion was
invested in AI-related research and innovation. The aim is to reach EUR 20 billion per
year in the decade following 2020.1 The EU’s role in AI is to facilitate and enhance
cooperation across the Union. The timeline, Figure 1 below, shows the EU AI initiatives
since 2018 when the Communication on Artificial Intelligence for Europe was
published.

1
Communication from The Commission to the European Parliament, the European Council, the Council, the
European Economic and Social Committee and the Committee of the Regions on Artificial Intelligence for
Europe. 25 April 2018. URL: https://ec.europa.eu/digital-single-market/en/news/communication-artificial-
intelligence-europe

6
 Adopt AI

Figure 1 Timeline of EU AI initiatives

Source: Authors’ elaboration based on A European approach to Artificial Intelligence 2.

The key goals of a number of the above-pictured initiatives are explained in Table 1
below.

Table 1 European Union initiatives on AI

Document title Date Key goals

(hyperlinked)

Communication on April 2018

The Communication set out to:
Artificial Intelligence
for Europe  Boost the EU's technological and industrial
capacity and AI uptake across the economy,
both by the private and public sectors;
 Prepare for socio-economic changes brought
about by AI by encouraging the modernisation
of education and training systems, nurturing
talent, anticipating changes in the labour
market, supporting labour market transitions
and adaptation of social protection systems;
 Ensure an appropriate ethical and legal
framework, based on the Union's values and in
line with the Charter of Fundamental Rights of
the EU.
Declaration of April 2018
The participating Member States agree to cooperate
cooperation on on:
Artificial Intelligence
 Boosting Europe's technology and industrial
capacity in AI and its uptake, including better
access to public sector data; are essential
conditions to influence AI development, fuelling
innovative business models and creating
economic growth and new qualified jobs;
 Addressing socio-economic challenges, such as
the transformation of the labour markets and
modernising Europe's education and training
systems, including upskilling & reskilling EU
citizens;

2
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

7
 Adopt AI

Document title Date Key goals

(hyperlinked)

 Ensuring an adequate legal and ethical

framework, building on EU fundamental rights
and values, including privacy and protection of
personal data, as well as principles such as
transparency and accountability.
Coordinated plan on December
Actions of the 2018 Coordinated plan:
Artificial Intelligence 2018 and
and its Review April 2021  Member States are encouraged to develop
their national AI strategies, building on the
work done at the European level;
 In the Union, public and private investments in
AI must be scaled up to reach the target of
EUR 20 billion per year over the next decade;
 Under the next programming period 2021-
2027, the Union invests in AI at least EUR 1
billion per year from Horizon Europe and the
Digital Europe programmes;
 Member States and the Commission will
engage in peer learning and discuss areas for
joint procurement of AI solutions, including
cybersecurity, as well as specific challenges for
the public sector.
Actions of 2021 reviewed Coordinated plan:

 Accelerate investments in AI technologies to

drive resilient economic and social recovery
aided by the uptake of new digital solutions;
 Act on AI strategies and programmes by fully
and timely implementing them to ensure that
the EU fully benefits from first-mover adopter
advantages;
 Align AI policy to remove fragmentation and
address global challenges.
Ethics guidelines for a April 2019
According to the Guidelines, trustworthy AI should be:
Trustworthy AI
 Lawful - respecting all applicable laws and
regulations
 Ethical - respecting ethical principles and
values
 Robust - both from a technical perspective
while considering its social environment
White Paper on February
The main building blocks of the White Paper are:
Artificial Intelligence – 2020
A European approach  The policy framework sets out measures to
to excellence and trust align efforts at the European, national and
regional levels. In a partnership between the
private and the public sector, the framework
aims to mobilise resources to achieve an
‘ecosystem of excellence’ along the entire
value chain, starting with research and
innovation, and to create the right incentives
to accelerate the adoption of solutions based
on AI, including by small and medium-sized
enterprises (SMEs).
 The key elements of a future regulatory
framework for AI in Europe will create a unique

8
 Adopt AI

Document title Date Key goals

(hyperlinked)

‘ecosystem of trust’. To do so, it must ensure

compliance with EU rules, including the rules
protecting fundamental rights and consumers’
rights, in particular for AI systems operated in
the EU that pose a high risk.
Source: Authors’ elaboration based on European Commission AI documents, hyperlinked in the ‘Document title’ column.

The European Commission in its 2018 Communication on Artificial Intelligence for

Europe recognised that public procurement is one of the key measures that have a
strong potential to facilitate AI adoption and help to stimulate demand and offer of AI
technologies in Europe.
In the European Commission’s Coordinated Plan on AI (2018; 2021), the Commission
and the Member States set the goal to support public administrations in procuring
trustworthy AI by developing a set of minimal capabilities for algorithms to be used in
contract conditions. The specific actions relevant to public procurement are presented
in Figure 2 below.
Figure 2 Actions of the Coordinated Plan on AI (2021) relevant to public procurement

Source: Authors’ elaboration based on Coordinated Plan on AI (2021).

The AI Watch report on the Overview of the use and impact of AI in public services in
the EU3 recommends the use of innovative public procurement to stimulate and speed
up AI adoption, which is likely to further extend the appropriateness and cost-
effectiveness of AI take-up in government, meeting the expectations of increasingly
proactive service providers and policy designers to the changing global landscape,
made more complicated by the Covid-19 crisis.

Finally, the need to support public procurement of AI systems was presented in the
White Paper on AI (2020). As presented in the Coordinated Plan on AI (2021), the
programme aims to help transform public procurement processes via open and
transparent sectoral dialogues to build a bridge between public procurers (who want to
know what solutions are available to address their needs) and European industry
(which wants to supply products/services to public administrations, and which needs
to know more about their plans). This will be organised on a European scale, allowing
the Member States to learn from each other. European Digital Innovation Hubs
(EDIHs) will be used to promote dialogue among industry actors throughout Europe.
The programme will thus stimulate industry investment in AI and the development of

3
European Commission. 2020. AI Watch Artificial Intelligence in public services – Overview of the use and
impact of AI in public services in the EU. URL:
https://publications.jrc.ec.europa.eu/repository/handle/JRC120399

9
 Adopt AI

new AI technologies and applications. Furthermore, the European Commission aims to

design a public procurement data space that will provide a comprehensive overview of
public procurement markets in the EU. A future IT tool will facilitate the use of AI
methods to analyse public procurement data. Available data combined with updated,
powerful analysis tools will be central to improving the governance of public
procurement.

1.2.2 Artificial Intelligence initiatives at the national level

This section presents the findings of the political and legal initiatives throughout the
27 EU Member States. These initiatives were researched based on the criteria of
whether Artificial Intelligence is a point mentioned within (Figure 3), and further
refined according to whether AI is the main subject matter (Figure 4), and how the
procurement of AI is envisaged within. The data collected and studied refers to non-
fixed periods, which include initiatives published between 2014 and 2020. The findings
are based on desk research of OECD data4, EU data5, and national-level data6.

As for the total overview of initiatives that include AI, a total of 72 initiatives were
found. Several initiatives foresee to be effective for multiple years. Figure 3 below
shows the breakdown by type of initiative for the total initiatives found. As can be
seen, most of the initiatives including AI are strategies (33 out of 72), legislation (13
out of 72) and Action plans (10 out of 72).

Figure 3 Types of national initiatives that include AI

Source: Authors’ elaboration.

The initiatives that include an AI point of view can be further broken into various
categories of focus. The most prominent ones include, for example, initiatives for
automated driving and vehicles7, touching upon, for instance, governmental guidelines
on testing on public roads and safety considerations. AI is also mentioned in
digitalisation8 plans, often from the data protection and human rights point of view.
Some countries have also published initiatives touching upon AI, specifically focusing

4
https://oecd.ai/en/dashboards
5
https://knowledge4policy.ec.europa.eu/home_en
6
Research in all Member State languages to identify and analyse AI initiatives.
7
For example: Austrian Automated Driving Regulation (2019); Belgian Autonomous vehicles code of
practice for testing (2016).
8
For example: Dutch Digitalisation Strategy (2021); Spanish Digital Rights Charter (2021).

10
 Adopt AI

on potential discrimination, human rights, and data disclosure.9 Legislations are only
present in the category of initiatives that are not all-encompassing about AI and
include types of initiatives as mentioned, on driving regulations and data ethics.

Of the total 72 initiatives, Figure 4 below shows the types of initiatives that focus
specifically on AI, adding up to a total of 38. These 38 initiatives on AI were published
between 2018 and 2020. They also include initiatives that have a long period of
validity, with the longest AI strategies from 2020 until 2030, namely in Hungary and
Poland.

Figure 4 Types of national initiatives that focus on AI

Source: Authors’ elaboration.

The large majority of initiatives are AI strategies. It emerges from the data analysed
that every Member State has a national AI-focusing initiative, except for Croatia (in
development), Romania (in development), and Slovakia (includes its AI priorities in
the general action plan for digital transformation). Conversely, there are no
legislations (yet) at the national level that focus particularly on AI. Furthermore, it has
been found that only Belgium has additional strategies/action plans in place for AI at
the sub-national level (more precisely, in both Wallonia and Flanders in addition to the
federal strategy).

The following Figure 5 provides an overview of the quantification of priority areas as

mentioned in the national AI initiatives, based on an analysis of the AI-focusing
initiatives and mentioned priorities within, carried out by the authors. The sizes of the
boxes depict the number of instances the policy sectors were mentioned within
national governmental AI programmes, categorised according to the COFOG
classification with added sub-categories as relevant.

9
For example: Danish Law on the disclosure of Data ethics Policy (2020); French Framework addressing
human rights concerns arising from facial recognition technology (2019).

11
 Adopt AI Study

Figure 5 Classification of public sector priority areas for AI

Source: Author’s elaboration.

12
 Adopt AI Study

The figure depicts the COFOG classifications and a breakdown of specific public sector
policy areas that Member State governmental AI initiatives have pointed out as a
priority for adopting AI. Given the high-level COFOG taxonomy, it is recommended
that the sub-categories within (as depicted in Figure 73) are used as takeaways to
identify the public sector policy areas which are most prioritised in national AI
initiatives. That is, according to COFOG, the government's broad objective (division) is
the 10 main categories, e.g., Economic affairs, which can be broken down into sub-
items (groups), e.g. agriculture, transport, and manufacturing within Economic affairs.
Therefore, the four most prioritised public sector policy areas are Health; General
Public Services (public administration and public services combined); Economic affairs
(transport and mobility); and Education.10

When it comes to the procurement of AI, it was found that among the Member States,
there are varying strategies and goals, and therefore lack of common denominators in
all countries’ approaches to public procurement of AI. We can extract three main
national approaches around AI procurement, focusing on i) internal development, ii)
R&D, and iii) the private sector, as depicted in Figure 6.

Figure 6 AI procurement national approaches

Source: Authors’ elaboration.

Under the internal development category, in Estonia, public sector civil servants
have access to guidelines and tools for better understanding and carrying out
procurements that may involve AI. These include a verification questionnaire;
instructional material; guidance material for data annotation; data protection impact
assessment; and explanation of technological concepts among other things.

For the R&D category, to exemplify, Hungary's AI strategy has set a dedicated R&D
and innovation procurement fund through the Artificial Intelligence National
Laboratory11, for the various policy sectors that it prioritises (manufacturing,
healthcare, agriculture, public administration, transportation, logistics and energy). In
Slovenia, AI R&D and innovation projects have been funded in priority policy areas
(health, industry 4.0, language technologies, and public administration).

Finally for the private sector-focus procurement of AI technologies mentions in

national AI initiatives, in Wallonia (Belgium) for instance, public funds are available to

10
Detailed information is not available on the selected public sector priority areas given that these are
usually listed in the initiatives without further context.
11
https://mi.nemzetilabor.hu/about-us

13
 Adopt AI Study

start AI projects since 2021, of which some projects supported public organisations in
their digital transformation into AI, thanks to the Start IA12 / Tremplin IA13 plans. In
Ireland, the government aims to lead the way and drive growth in AI by purchasing
and developing ethical and trustworthy AI applications, and by using public
procurement policies to stimulate and encourage the industry to provide AI-based
products and services to the government.

Main findings/implications
➔ The four most prioritised public sector policy areas in national initiatives for
AI are Health; General Public Services (public administration and public
services combined); Economic affairs (transport and mobility); and
Education.

1.2.3 Overview of AI projects in the public sector

This section provides an overview of the identified projects in which AI technology is
used. The collection of information is based on desk research from AI Watch 14,
OECD15, and CORDIS16. The projects are categorised by the following characteristics:

 Administrative level,
 Public sector policy area,
 AI typology.
Our research found a total of 182 projects that use AI as a technology.

The breakdown of the administrative levels of the AI projects in the public sector is
depicted in Figure 7 below.

Figure 7 Administrative levels of AI projects

Source: Authors’ elaboration.

12
https://www.digitalwallonia.be/fr/publications/appel-start-ia-05
The Start IA plan helps companies to exploit data thanks to Artificial Intelligence, with the help of expert
companies in the field.
13
https://www.digitalwallonia.be/fr/publications/dw4ai-tremplin-ia
The Tremplin IA plan is aimed at Walloon companies wishing to do a feasibility study or a Proof Of Concept
(POC) involving Artificial Intelligence technologies.
14
https://ai-watch.github.io/AI-watch-T6-X/catalog/10009.html
15
https://oecd.ai/en/dashboards
16
https://cordis.europa.eu/

14
 Adopt AI Study

It can be seen that most projects are implemented at the national level (98 projects
out of 182), for the whole government. Projects at this level may include, for instance,
chatbots on ministerial websites, automated classification of files, automated analysis
of water quality, etc. The local level projects closely follow in number (43 out of 182).
At this level, AI tools are mostly implemented to be tested out in a particular city,
such as for citizen opinion analysis and automated parking control. As for regional
projects using AI, these are few (10 out of 182) and concentrated in countries where
regions have high political power, which are Belgium, Germany, and Italy. Projects
include, for instance, agricultural subsidy monitoring, automated audio processing,
and regional mobility forecasting. The international level projects (31 out of 182)
represent the projects that were identified where multiple countries and stakeholders
were involved, that is, through Horizon 2020 funding. These projects may fall into the
category of general health appliances and automated vehicles, which can be adopted
at any other level.

The public sector policy areas in which the AI projects are implemented were also
classified according to the COFOG taxonomy and where relevant, these were further
broken down as done above for the national AI initiatives. Below in Figure 8 is a
classification of the public sector areas within which AI projects are implemented, to
be read in the same manner as for Figure 5 above (i.e. the size of the boxes
represents the quantification of instances projects fell into a category).

15
 Adopt AI Study

Figure 8 Classification of public sector areas of AI projects

Source: Authors’ elaboration.

16
 Adopt AI Study

According to the quantification depicted in the figure above, the COFOG public sector
policy areas in which projects are most implemented are General public services
(public administration and public services combined); Economic affairs
(transportation); Public order and safety (police services); and Health (medical
equipment). AI projects under the public services category include technologies that
contribute to providing digitalised services to citizens such as chatbots, intelligent
digital assistants, and virtual agents. The transportation category includes projects
that may help traffic flow with predictive analytics. Police services, in turn, include AI
technologies, more specifically computer vision and security analytics that support
ensuring regulations e.g. mobile phone detection while driving, or identifying fraud.
Finally, medical equipment is defined as technologies that are needed for health
purposes and also include predictive analytics for diagnostics.

The breakdown of AI typologies is based on the AI Watch17 AI typology (see

Figure 76). The categorisation and amount of projects falling into each category are
shown in Figure 9 below.

Figure 9 AI projects' typologies

Source: Authors’ elaboration.

The four types of AI technologies that are currently most used in the Member States’
AI projects are Chatbots, Intelligent Digital Assistants, Virtual Agents and

17
European Commission. 2020. AI Watch Artificial Intelligence in public services – Overview of the use and
impact of AI in public services in the EU. URL:
https://publications.jrc.ec.europa.eu/repository/handle/JRC120399

17
 Adopt AI Study

Recommendation Systems; Machine Learning, Deep Learning; Predictive Analytics,

Simulation and Data Visualisation; Computer Vision and Identity Recognition.

Main findings/implications
➔ The four public sector policy areas in which AI projects are currently most
implemented are General public services (public administration and public
services combined); Economic affairs (transportation); Public order and
safety (police services); and Health (medical equipment).
➔ The four AI typologies that the projects used are chatbots, computer vision,
machine learning and predictive analytics.

1.3 Barriers to and benefits of the adoption of AI in the public and

private sectors
Based on the findings of the desk research, this section identifies the main challenges
(issues/obstacles) and benefits that the Member States’ governments face when
considering the public procurement of AI technologies. Specific cases will be
mentioned, as well as general discussions as found in the literature.

1.3.1 Challenges
Four main axes of challenges have been identified through desk research that may
pose a barrier to the public procurement and adoption of AI technologies in the public
sector (and private sector when relevant), as shown in Figure 10 below and further
explored in the following sub-sections.

Figure 10 Macro categories of challenges

Source: Authors’ elaboration.

1.3.1.1 Procurement process challenges

Procurement process challenges are caused by internal organisational obstacles and
are mostly related to political and legal challenges. Furthermore, the procurement
process challenge also relates to public organisations deciding whether an AI tool is
necessary to procure for the needs at hand.18

18
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a Systematic
Understanding on the Challenges of Procuring Artificial Intelligence in the Public Sector. (Final version not
yet published)

18
 Adopt AI Study

According to the World Bank19, there are currently inadequate policies, legislation, and
incentives for AI. Indeed, at the moment, there is a lack of legal clarity over the use of
AI technologies which causes an obstacle underlined by national governments, such as
written in the Austrian Research and Technology Report 20. The French Mission Villani
report21 dedicates a section on leveraging public procurement for AI. To do so, it
emerges from the report that innovative procurement is not yet sufficiently utilised
due to a lack of information on the appropriate procedures, as well as the legal and
operational risks involved with these types of purchases such as liability on the public
purchaser. The Danish strategy22 also points to the lack of guidelines and an ethical
framework for AI as a challenge.

These challenges are also confirmed when looking at the private sector's point of view.
Legislation-related challenges are considered as stemming from external hindrances,
and the lack of laws or regulations poses an obstacle for 69% of the surveyed private
sector enterprises by the European Commission 23.

It emerges a need for alignment between the public and private sectors to understand
the market functioning and ensure that public procurement is aligned with what the
private sector may be able to offer. One of the points made by the Austrian
government24 is that to create a suitable framework for using AI in public
administration is “making use of public procurement (promoting innovation), i.e. so
that public administration can generate demand for ethical AI or applications in certain
industries such as healthcare or similar, enabling it to define markets and set
standards”.

The European Commission is undertaking several initiatives in this regard. One is a

community of practice for public buyers and the e-competence centre25, aiming to
spread good practices on the procurement of innovation, and tools and information to
help public buyers. Additionally, the Commission is developing a specific Community of
Practice on the development and testing of EU-wide AI procurement Clauses26.

1.3.1.2 Data challenges

There are multiple data-related challenges to the public procurement of AI relating to
data “availability, governance, ownership, and infrastructure” 27.

19
World Bank. 2020. Artificial Intelligence in the Public Sector : Maximizing Opportunities, Managing Risks.
URL: https://openknowledge.worldbank.org/handle/10986/35317
20
Austrian Federal Ministry of Education, Science and Research. 2021. Austrian Research and Technology
Report. URL : https://www.bmbwf.gv.at/en/Topics/Research/Research-in-Austria/Services/FTB.html
21
AI for Humanity. 2018. For a meaningful Artificial Intelligence (Mission Villani Report). URL:
https://www.aiforhumanity.fr/pdfs/MissionVillani_Report_ENG-VF.pdf
22
The Danish Government. 2019. National Strategy for Artificial Intelligence. URL:
https://eng.em.dk/media/13081/305755-gb-version_4k.pdf
23
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
24
Austrian Federal Ministry of Education, Science and Research. 2021. Austrian Research and Technology
Report. URL : https://www.bmbwf.gv.at/en/Topics/Research/Research-in-Austria/Services/FTB.html
25
https://commission.europa.eu/funding-tenders/tools-public-buyers_en
26
https://ec.europa.eu/newsroom/growth/items/746858/en
27
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a Systematic
Understanding on the Challenges of Procuring Artificial Intelligence in the Public Sector. (Final version not
yet published)

19
 Adopt AI Study

In the public sector, the lack of clear “data ownership” can create an issue of
sovereignty, as pointed out by the French Mission Villani report 28: for AI, a certain
dependency may arise on foreign technologies, therefore causing an issue of
sovereignty. Indeed, there is a challenge related to “data ownership” and “data
sovereignty”, i.e., where is the data collected and who is it sold to with or without
complete ownership.

The insufficient availability of high-quality digital data and digital

infrastructure in the public sector was also pointed out as a challenge by the World
Bank29 as well as the Danish National Strategy for AI30. Low data quality may lead to
poor performance of the AI, thereby not being able to fully automatise the selected
processes. A well-developed digital infrastructure is also more ready for AI adoption,
as it will lead to interoperable, successful and sustainable AI solutions. Public
administrations that lack this are unlikely to benefit from AI technologies.

1.3.1.3 AI technology challenges

AI technology challenges concern its “quality, transparency, and bias”31. These relate
to the necessary commitment that governments must make in their public
procurement processes, to overcome challenges associated with ethics and potentially
biased AI.

The European Parliament32 reports the following risks/challenges related to AI

technology adoption in the public sector:

• Discrimination due to data bias and hard coding of presumptions.

• Transparency and explicability. Good governance principles and the right to an

explanation of a decision become more difficult when the use of AI and ‘black-
box’ algorithms for decision support or algorithmic decision-making is
increasing.

• The dehumanisation of public services. Due to the hard coding of governmental

processes and decision-making, there is less and less leeway for exceptions or
case-by-case circumstances (automation in general and AI specifically do not
deal well with exceptions or boundary cases).

In the public sector especially, when procuring a technology, the potential for bias and
discrimination in AI, i.e., who bears responsibility and liability for the actions due to
the algorithm, is an important factor to address during the procurement/development
process.33 The question of understanding who bears responsibility and liability for AI

28
AI for Humanity. 2018. For a meaningful Artificial Intelligence (Mission Villani Report). URL:
https://www.aiforhumanity.fr/pdfs/MissionVillani_Report_ENG-VF.pdf
29
World Bank. 2020. Artificial Intelligence in the Public Sector : Maximizing Opportunities, Managing Risks.
URL: https://openknowledge.worldbank.org/handle/10986/35317
30
The Danish Government. 2019. National Strategy for Artificial Intelligence. URL:
https://eng.em.dk/media/13081/305755-gb-version_4k.pdf
31
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a Systematic
Understanding on the Challenges of Procuring Artificial Intelligence in the Public Sector. (Final version not
yet published)
32
European Parliament. 2021. Artificial Intelligence and public services. URL :
https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/662936/IPOL_BRI(2021)662936_EN.pdf
33
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a Systematic
Understanding on the Challenges of Procuring Artificial Intelligence in the Public Sector. (Final version not
yet published)

20
 Adopt AI Study

technology, as found by a European Commission study, is also raised by 59% of

enterprises in the EU, from a private sector point of view.34

The Austrian Research and Technology Report 35 for example raises the challenge of
understanding how to create a non-discriminatory AI decision-making tool while
ensuring traceability and transparency. On the other side, the Swedish National
Approach to AI36 aims to ensure that the country can manage the challenges
associated with AI, especially when it comes to the unintended consequences that
may arise from biased data, lack of transparency, and misuse. The priority of the
Swedish government in this respect is to avoid any discrimination, loss of trust,
financial damage and consequences on democracy. In the analysis of AI projects, it
was also found that privacy protection is a challenge faced when implementing the
projects. The transparency and explicability challenge seems to be fully confirmed also
when looking at the point of view of the private sector, which is considered a barrier
that 61% of those surveyed by the European Commission (2020).37

1.3.1.4 Organisational capacity challenges

Organisational capacity challenges relate to barriers at managerial, technical, and
human capital levels stemming internally and similarly, from both public and private
sector organisations.38

At the managerial level, leaders play an important role in the transition to AI, and
there may be a lack of leadership and awareness39 of AI’s possible uses that may
benefit organisational processes. Furthermore, communication with all involved parties
is necessary (i.e. all levels of government as well as outside stakeholders) who will
contribute to the development and/or running of the AI solution in the public sector. 40

At the technical level, building up AI expertise is also a challenge.41 If there is no

proper understanding of what AI entails that is context-specific (i.e. country-level,
public sector area-level), issues may arise, and non-appropriate strategies may be
adopted to deal with those issues.42 For instance, for the public sector, ensuring the

34
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
35
Austrian Federal Ministry of Education, Science and Research. 2021. Austrian Research and Technology
Report. URL : https://www.bmbwf.gv.at/en/Topics/Research/Research-in-Austria/Services/FTB.html
36
Government Offices of Sweden. 2018. National Approach to Artificial Intelligence. URL:
https://www.government.se/4a7451/contentassets/fe2ba005fb49433587574c513a837fac/national-
approach-to-artificial-intelligence.pdf
37
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
38
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a Systematic
Understanding on the Challenges of Procuring Artificial Intelligence in the Public Sector. (Final version not
yet published)
39
World Bank. 2020. Artificial Intelligence in the Public Sector : Maximizing Opportunities, Managing Risks.
URL: https://openknowledge.worldbank.org/handle/10986/35317
40
Campion, A., Gasco-Hernandez, M., Jankin Mikhaylov, S. and Esteve, M., 2020. Overcoming the
challenges of collaboratively adopting artificial intelligence in the public sector. Social Science Computer
Review, p.0894439320979953.
41
Austrian Federal Ministry of Education, Science and Research. 2021. Austrian Research and Technology
Report. URL : https://www.bmbwf.gv.at/en/Topics/Research/Research-in-Austria/Services/FTB.html
42
World Bank. 2020. Artificial Intelligence in the Public Sector : Maximizing Opportunities, Managing Risks.
URL: https://openknowledge.worldbank.org/handle/10986/35317

21
 Adopt AI Study

interoperability of an AI system with other government applications is an important

necessity.43

The human capital level relates to the lack of digital skills.44,45 “Public agencies find
themselves without core AI skills, which hinders their ability to deploy and operate AI
solutions.”46 However, once the AI experts with the required skills are hired, they
expect higher salaries, which is identified as a variable behind the high cost of
adopting AI solutions in the public sector. 47 In the private sector, the difficulty to hire
new staff with the right skills to handle AI, as well as the lack of skills among existing
staff, together pose very high barriers to the adoption of AI (85% and 81% of
companies respectively consider so).

1.3.1.5 Sectoral and AI technologies-related challenges

In a research carried out on European enterprises by the European Commission in
202048, several challenges, i.e. lack of public or external funding, strict standards for
data exchange and liability for damage caused by AI, were measured in light of their
impact on the economic sectors. Observing these impacts can shed light on
contributing factors that hinder the uptake of AI within the public sector sphere.

Figure 11 shows that the three types of obstacles do not vary greatly on average
across sectors. Lack of public or external funding is considered an obstacle throughout
the sectors in 36% of the cases on average; strict standards for data exchange at
34%; and liability for damage caused by AI at 33%.

43
Kankanhalli, A., Charalabidis, Y., & Mellouli, S. (2019). IoT and AI for smart government: A research
agenda. Government Information Quarterly, 36(2), 304–309.
44
World Bank. 2020. Artificial Intelligence in the Public Sector : Maximizing Opportunities, Managing Risks.
URL: https://openknowledge.worldbank.org/handle/10986/35317
45
The Danish Government. 2019. National Strategy for Artificial Intelligence. URL:
https://eng.em.dk/media/13081/305755-gb-version_4k.pdf
46
World Economic Forum. 2019. 5 challenges for government adoption of AI. URL:
https://www.weforum.org/agenda/2019/08/artificial-intelligence-government-public-sector/
47
European Commission Joint Research Centre. Unpublished. AI Watch: Road to the adoption of Artificial
Intelligence by the Public Sector.
48
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368

22
 Adopt AI Study

Figure 11 External obstacles to the adoption of AI

Source: Figure 21, European Commission, Directorate-General for Communications Networks, Content and Technology.
2020. European enterprise survey on the use of technologies based on artificial intelligence: final report.

The same survey also looked into internal obstacles to the adoption of AI, as shown in
Figure 12. On average across sectors it was found that the difficulty to hire new staff
was considered the most important obstacle (57%) followed by the cost of adoption
(53%); and the cost of adapting operational processes (50%).

Figure 12 Internal obstacles to the adoption of AI

Source: Figure 25, European Commission, Directorate-General for Communications Networks, Content and Technology.
2020. European enterprise survey on the use of technologies based on artificial intelligence: final report.

23
 Adopt AI Study

1.3.1.6 AI technological considerations of the challenges

Focusing on AI technologies, in a research carried out on European enterprises by the
European Commission in 202049, a number of the above-discussed challenges were
measured by looking at barriers to the adoption of AI by specific technologies (Figure
13)50. Interestingly, this perspective confirms the takeaways from the previous
analysis of the challenges, with i) liability considerations for damage, ii) poor skills
availability, iii) high costs of adoption, iv) (lack of) digital infrastructure and v)
difficulty to understand algorithms, being the most important barriers reported to
diffusion and use of AI, across different AI technologies.

Figure 13 Barriers to the adoption of AI by technology

Source: Table 7, European Commission, Directorate-General for Communications Networks, Content and Technology.
2020. European enterprise survey on the use of technologies based on artificial intelligence: final report.

Main findings/implications
➔ The barriers to the public procurement and adoption of AI technologies, as
found in the general literature and as identified at sectoral and technological
levels can be categorised as below:
o Procurement process challenges
o Data challenges
o AI technology challenges
o Organisational capacity challenges
Many of these challenges are also confirmed when looking at similar sources of
information for the private sector.

1.3.2 Benefits
Governments aim to use AI to provide better public services to citizens. This in
turn can translate into virtuous spillovers across different public sectors, such as

49
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
50
Please note that the AI typologies are not fully aligned with the classification made by AI Watch.

24
 Adopt AI Study

higher transportation safety, more varied cultural services, and improved healthcare
services, among others. The adoption of AI technologies may also support
governments’ internal processes, by digitalising and modernising public
administrations. This may take the form of a document, audio, and speech analysis
which speeds up proceedings and provide more time for the provision of other
services. These two main axes are depicted in Figure 14 below.

Figure 14 Macro categories of benefits

Source: Authors’ elaboration

The use of public procurement is hence a strategic and beneficial activity for AI
adoption, as it takes advantage of economies of scale in the public sector, and sets up
networks to learn from other use cases (be they best practices or issues). According to
the World Economic Forum, by “utilising public procurement, governments could not
only support AI innovation and economic growth but also set standards, with a
signalling effect on the market.”51

The European Parliament52 points out several generic benefits of AI in public services,
such as process optimisation, economic growth, increase in jobs, efficiency, time and
cost savings, service improvement, and improved accessibility and inclusion of
services. Certain benefits may only be perceived in the long term, such as the
eventual better allocation of public resources as a consequence of improved digital
services. The report summarises the main benefits of AI to be:

 Efficiency gains and internal process optimisation.

 Less human error and fraud, both internally and in services to businesses
and citizens.
 Possibility to deliver more accessible and inclusive services: personalisation.
 Increase of anticipatory governance and policy: more accurate predictions.

This is confirmed by the analysis carried out in this study on governmental AI

initiatives and projects. Looking at the positive outcomes of the AI projects
analysed in this study, enhanced data processing capabilities, better user
experience and less work for humans are the most important cited positive

51
World Economic Forum. 2020. How governments can use public procurement to shape the future of AI
regulation – and boost innovation and growth. URL: https://www.weforum.org/agenda/2020/06/artificial-
intelligence-ai-government-procurement-standards-regulation-economic-growth-covid-19-response/
52
European Parliament. 2021. Artificial Intelligence and public services. URL :
https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/662936/IPOL_BRI(2021)662936_EN.pdf

25
 Adopt AI Study

outcomes. Other notable mentions of positive project outcomes that utilise AI as

technology are: innovation, prediction, and optimisation among others.53

The use of AI in governments comes with a multitude of benefits and challenges. As

for the procurement of AI, governments must ensure that the technology is safe and
that ethical concerns may be solved. There are also benefits to the deployment of AI
in governments, which provides better decision-making, better public services to the
citizens, and better operations within. AI may even be used for public procurement
itself, thereby resulting in cost and employee working time savings. Furthermore, a
majority of governments do not develop AI solutions in-house54, for which reason
there is a dependency on public procurement and the private sector to deliver AI
solutions for public services. “Today, national governments should move beyond
talking about AI in principle, and put into action practical solutions to foster the
adoption of AI while mitigating its risks.” 55

1.3.2.1 Tools and ways to support the achievement of the benefits

Research has pointed out that to overcome the barriers and challenges, and to support
the achievement of the benefits of AI adoption, guidelines may be put in place for
the public sector specifically to better carry out public procurement for AI
technologies. Such guidelines could encompass standard templates, policies and
legal frameworks, as well as ethical considerations.

The World Economic Forum recommends that AI-specific tender templates, among
others, could boost innovation and economic growth. Public procurement processes
should also include young and small companies, including start-ups, which could learn
how to tender correctly. For the procurement of AI, accountability and ethics should
be embedded in the purchasing cycle. In fact, in general, the traditional approaches to
public procurement should be rethought. To support governments in doing so, the
World Economic Forum published two documents that support governments for the
procurement of AI: AI Government Procurement Guidelines 56 and Project Overview57.

Two country-specific examples can better illustrate the role of guidelines and
templates to better capitalise on the potential benefits of AI. The United Kingdom58
has published two relevant documents on AI, a Guide to using Artificial Intelligence in

53
A full list of keywords identified for the positive outcomes/objectives of an AI project is the following, in
alphabetic order: accuracy; adaptation to new information; analysis; automated detection; automation;
better analysis; better collaboration; better data processing; better decision making; better user
experience; building a service with citizen input; cost and time savings; cost efficiency; created by real life
needs; customer service; customer service around the clock; data processing; digitisation; efficiency; faster
service; high level of detail; high quality ; innovation ; knowledge development; low labour costs;
optimisation; prediction; real-time monitoring; reduced costs; reduced human labour; reduction of paper;
reliability; scalability; security; speed; transparency.
54
European Parliament. 2021. Artificial Intelligence and public services. URL :
https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/662936/IPOL_BRI(2021)662936_EN.pdf
55
World Economic Forum. 2020. How governments can use public procurement to shape the future of AI
regulation – and boost innovation and growth. URL: https://www.weforum.org/agenda/2020/06/artificial-
intelligence-ai-government-procurement-standards-regulation-economic-growth-covid-19-response/
56
World Economic Forum. 2020. AI Procurement in a Box: AI Government Procurement guidelines. URL:
https://www2.deloitte.com/content/dam/Deloitte/global/Documents/About-Deloitte/gx-wef-ai-government-
procurement-guidelines-2020.pdf
57
World Economic forum. 2020. AI Procurement in a Box : Project overview. URL:
https://www3.weforum.org/docs/WEF_AI_Procurement_in_a_Box_Project_Overview_2020.pdf
58
The United Kingdom is not a country included in the main analysis as it is not part of the EU-27. However,
it is showcased given the relevance of the governmental AI initiatives.

26
 Adopt AI Study

the public sector59 and Guidelines for AI procurement60. The Guide outlines that there
are several considerations to take into account when procuring AI, such as
understanding the needs, the maturity of commercially available products that meet
those needs, and how to integrate the AI solution with the existing infrastructure. It is
also pointed out that a data assessment should be carried out and ethical concerns
must always be considered from the start of the procurement process. In Ireland61,
the governmental public procurement of AI is governed by trustworthiness and safety.
Built upon the EU AI HLEG Ethics Guidelines, an Assessment List for Trustworthy AI
(ALTAI) has been put in place, which is a checklist of questions that both public and
private sector organisations may refer to. The AI strategy of Ireland aims to use public
procurement “as a catalyst for ethical and innovative AI”. Furthermore, “instruments
such as dialogues, hackathons and pre-commercial procurement of innovative
solutions will enable suppliers to respond better to public procurement requests, and
also assist public authorities to understand the market better and formulate targeted
procurements”.

Main findings/implications
➔ The benefits to the public procurement and adoption of AI technologies, as
depicted in the literature can be summarised below:
o Through public procurement, governments support AI innovation and
the economic growth of the market.
o AI technologies modernise public administrations’ internal processes
as well as governmental public services.
o By creating and/or following existing guidelines, European
governments ensure the continued adoption of ethical and
trustworthy AI.

1.4 Economic analysis

This section of the scoping report revolves around economic analysis. It provides an
analysis of the economic potential of further investments in AI by looking at existing
quantitative data on public procurement as well as market trends and maturity of AI
technologies, including their use in the private sector, to complement data available
on budgets dedicated to AI at the public procurement level. The latter is unfortunately
too scattered to conclude about the economic potential of investments in AI. The
following tasks of this study will work towards closing this gap by exploring four
sectors more in detail and by performing primary data collection on the subject of
public procurement of AI.
The main underlying concept of this economic analysis is that some AI technologies
are more ready than others for large-scale deployment and are used more intensively
in specific sectors, henceforth, using those technologies in certain public sector areas
would provide greater added value than not yet ready AI technologies. This section

59
UK Government Digital Service. 2020. A guide to using artificial intelligence in the public sector. URL:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/979892
/A_guide_to_using_AI_in_the_public_sector__Print_version_.pdf
60
UK Government Digital Service. 2020. Guidelines for AI procurement. URL:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/990469
/Guidelines_for_AI_procurement.pdf
61
Government of Ireland. 2021. AI – Here for Good. A National Artificial Intelligence Strategy for Ireland.
URL: https://assets.gov.ie/152580/b8ad2fa0-9ef2-44da-bab6-aaf8bb03c898.pdf

27
 Adopt AI Study

thus serves to provide an understanding of the market as it stands, and the potential
for the adoption of AI technologies in the different sectors.
Understanding the performance of the private sector and the overall market will allow
us to extract considerations for the public sector: adopting AI requires a high level of
digital readiness and digital transformation.
The private sector can be considered a leading actor in the uptake and development of
AI while the public sector is currently at low digital maturity as compared to other
sectors, as further detailed in Figure 15. According to the BCG (2021), the top four
sectors in terms of digital maturity are Tech and telecommunications, financial
institutions, the Consumer sector, and Industrial goods.
Figure 15 Likelihood of digital transformation success, 2021

Source: Boston Consulting Group. 2021. Which Sectors Perform Best in Digital Transformation?62

The importance of looking at the public sector adoption of AI in parallel with the
market dimensions of sectoral digital readiness and AI technologies maturity is
motivated by the fact that collaboration of the public and private sectors is key to
increasing knowledge and know-how on the procurement of innovative technologies.
For instance, as mentioned by the AI Watch63, the GovTech ecosystem may be
stimulated to develop AI for public sector use. According to Accenture64, AI innovation
for the public sector will emerge if the GovTech65 ecosystem works together to achieve
so. This passes from streamlining procurement processes as well as encouraging
feedback for improvements from the industry. Another essential condition is the
training and reskilling of employees both in the private and public sectors, so to
minimise job losses and maximise job creation. Along these lines, in a Global
Government Forum article66, it is also pointed out that the public and private sectors
should endorse each other to build employees’ skills.

62
Boston Consulting Group. 2021. Which Sectors Perform Best in Digital Transformation? URL:
https://www.bcg.com/publications/2021/learning-from-successful-digital-leaders
63
European Commission. 25 May 2022. AI Watch: Road to the adoption of AI by the Public Sector. URL:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1
64
Accenture, 2019. Transforming Public Service with AI. URL: https://www.accenture.com/_acnmedia/PDF-
110/Accenture-Govtech-POV.pdf
65
Governments, start-ups, medium and large companies, civil society.
66
Global Government Forum. 2021. Making Artificial Intelligence fit for a life inside government. URL:
https://www.globalgovernmentforum.com/making-artificial-intelligence-fit-for-a-life-inside-government/

28
 Adopt AI Study

1.4.1 An introduction to AI adoption

This section will provide detailed findings on the current state of play relating to AI
adoption in the private sector. This will be done according to the following indicators:

• Market growth: Defined as the rate at which a market’s size is increasing;

compound annual growth rate (CAGR).
• Market adoption: Defined as the rate at which new technology is acquired and
used by the market.
• Labour productivity: Defined as the output per unit of labour input.
• Employment: Defined as the number of people engaged in productive activities
in an economy.
Given the fact that we prioritised forecasts and there a single data source covering all
those indicators is missing, we relied on the most recent reports from top market
analysts and consulting companies and organisations (World Economic Forum),
whenever raw data at the Eurostat level was missing.

Market growth
The global AI market is growing at an incredibly fast rate. According to the industry
report published by Mordor Intelligence (2021)67, the market of global AI is forecasted
to increase from EUR 45.34 billion in 202068 to EUR 167.48 billion in 202669, with a
compound annual growth rate (CAGR) of 26.1%. Similarly, Statista forecasts the AI
software market size to reach USD 126 billion by 2025, corresponding to EUR 113.96
billion70.

Moreover, according to PwC Global (2017)71, the total contribution of AI to the global
economy is expected to reach EUR 14.2 trillion72 by 2030, corresponding to a 14%
increase in global GDP.

Market adoption
In terms of market adoption, the Worldwide Artificial Intelligence Spending Guide
(2021)73 by the International Data Corporation (IDC) forecasts global spending on AI
to hit EUR 184.51 billion74 in 2025, and European spending75 on AI systems will jump
from EUR 15.58 billion76 in 2021 to more than EUR 45 billion77 in 2025. The compound
annual growth rate (CAGR) for 2021–2025 will be 26.7%.

67
Mordor Intelligence. 2021. Artificial Intelligence Market - Growth, Trends, Covid-19 Impact, Forecast
(2022 - 2027). URL: https://www.mordorintelligence.com/industry-reports/global-artificial-intelligence-
market
68
USD 50.13 billion as of 17 March 2022.
69
USD 185.17 billion as of 17 March 2022.
70
As of 17 March 2022.
71
PricewaterhouseCoopers. 2017. Sizing the prize. What’s the real value of AI for your business and how
can you capitalise? URL: https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-
prize-report.pdf
72
USD 15.7 trillion as of 17 March 2022
73
International Data Corporation (IDC). 2021. Worldwide Artificial Intelligence Spending Guide. URL:
https://www.idc.com/tracker/showproductinfo.jsp?containerId=IDC_P33198
74
USD 204 billion as of 17 March 2022.
75
International Data Corporation (IDC). 2021. European Spending on Artificial Intelligence Will Reach $22
Billion in 2022, Supported by Strong Investments Across Banking and Manufacturing. URL:
https://www.idc.com/getdoc.jsp?containerId=prEUR148297521
76
USD 17.3 billion as of 17 March 2022.
77
USD 50 billion as of 17 March 2022.

29
 Adopt AI Study

Furthermore, according to Gartner (2019)78, the number of enterprises using AI in

business grew by 270% from 2015 to 2019. Along the same line, McKinsey&Co
(2021)79 reported that 56% of all respondents to the State of AI survey declared AI
adoption in at least one function, an increase of 6 percentage points from 202080.

To further complement these findings, Figure 16 below provides a country overview of

the percentage of enterprises using AI in 2021.

Figure 16 Enterprises using AI technologies by Member State, 2021

Source: Eurostat. 2021. Artificial Intelligence in EU Enterprises.81

In the EU, Denmark has the highest share of businesses using AI applications,
followed by Portugal, Finland and Luxembourg.

Labour productivity
In terms of labour productivity, Accenture (2016)82 found that AI as technology can
enhance productivity by up to 40%. More specifically, based on PwC’s Global Artificial
Intelligence Study (2017)83, increased productivity will contribute to approximately
40% of the previously mentioned increase in global GDP driven by AI of EUR 14.2
trillion. Moreover, based on the country-level analysis carried out by Accenture,
Sweden and Finland are the countries with the highest forecasted increase in

78
Gartner. 2019. Gartner Survey Shows 37 Percent of Organizations Have Implemented AI in Some Form.
URL: https://www.gartner.com/en/newsroom/press-releases/2019-01-21-gartner-survey-shows-37-
percent-of-organizations-have
79
McKinsey & Company. 2021. The state of AI in 2021. URL: https://www.mckinsey.com/business-
functions/mckinsey-analytics/our-insights/global-survey-the-state-of-ai-in-2021
80
McKinsey & Company. 2020. The state of AI in 2020. URL: https://www.mckinsey.com/business-
functions/mckinsey-analytics/our-insights/global-survey-the-state-of-ai-in-2020
81
Eurostat. 2021. Artificial Intelligence in EU enterprises. URL: https://ec.europa.eu/eurostat/web/products-
eurostat-news/-/ddn-20210413-1
82
Accenture. 2016. Why Artificial Intelligence is the Future of Growth. URL:
http://dl.icdst.org/pdfs/files2/2aea5d87070f0116f8aaa9f545530e47.pdf
83
PricewaterhouseCoopers. 2017. Sizing the prize. What’s the real value of AI for your business and how
can you capitalise? URL: https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-
prize-report.pdf

30
 Adopt AI Study

productivity levels for 2035. They are followed by the US, Japan and Austria,
Germany, the Netherlands and the UK.

Employment
In terms of employment, AI uptake will have an impact on the job market. According
to the World Economic Forum (2020)84, the ratio of job loss to job creation due to AI
will be 85 million to 97 million jobs by 2025, respectively. That is an overall addition of
12 million jobs, as presented in Figure 17 below.

Figure 17 Forecasted job landscape, 2020

Source: World Economic Forum. 2020. Don't fear AI. It will lead to long-term job growth.85

These findings also imply both the reskilling and upskilling of the workforce. On the
one hand, employees will have to either learn and be taught new skills (reskilling) or
additional ones (upskilling). On the other hand, employers will have to ensure such
processes. In this regard, it is interesting to look at the current level of human capital
in the EU Member States calculated as part of the Digital Economy and Society Index
(DESI)86 and presented in Figure 18 below.

84
World Economic Forum. 2020. Don’t fear AI. It will lead to long-term job growth. URL:
https://www.weforum.org/agenda/2020/10/dont-fear-ai-it-will-lead-to-long-term-job-growth/
85
World Economic Forum. 2020. Don’t fear AI. It will lead to long-term job growth. URL:
https://www.weforum.org/agenda/2020/10/dont-fear-ai-it-will-lead-to-long-term-job-growth/
86
European Commission. 2021. Digital Economy and Society Index (DESI) 2021. Human capital. URL:
https://Digital-Economy-and-Society-Index-DESI-2021%2F1007%2FDESI-2021-Thematic-chapters---
Human-capital

31
 Adopt AI Study

Figure 18 Human capital dimension by Member State, 2021

Source: European Commission. 2021. DESI 2021.

In the EU, Finland is the country with the highest level of human capital, followed by
Sweden, the Netherlands and Denmark.

Main findings/implications
➔ The global AI market is expected to grow at a fast rate in terms of size and
adoption, increase productivity, and generate employment.

1.4.2 Budget analysis of public procurement of AI

Following the desk research performed for this study on national governmental AI
initiatives, it was found that a total of 15 countries87 have publicly shared the funding
amounts dedicated specifically to AI, which may include funding for R&D and private
and/or public (projects) investment programmes. A total of 18 budgets have been
uncovered throughout the research on national AI initiatives. The public funding is
either expressed as a dated lump sum (i.e., a total portfolio for spending until a
foreseen date) or an un-dated lump sum (i.e., no clear indication of portfolio spending
timeline). In terms of dated lump sums, for 13 of the budgets, it was possible to
calculate the yearly average spending on AI, which amounts to ca. EUR 141 million 88.
For example, in Germany, EUR 5 billion is made available for federal investments in AI
until 2025, first announced at EUR 3 billion in 2018.

In terms of un-dated lump sums, with an unclear timeline for investment, the six
remaining budgets average EUR 529 million89. This average amount is influenced
(upward) by the Italian strategy that envisages a public investment of EUR 2.5 billion,

87
Austria, Belgium (federal, Wallonia, Flanders), Denmark, Estonia, Finland, France, Germany, Ireland,
Italy, Latvia, Lithuania, the Netherlands, Slovenia, Spain, Sweden.
88
With an exact amount of EUR 141,504,087.58. This amount is an average yearly amount, referring to
different base years depending on the initiative.
89
With an exact amount of EUR 529,583,333.33.

32
 Adopt AI Study

whereas, for instance, the Latvian government plans to spend approximately EUR 1.5
million on digitalisation with a focus on AI.

Following the desk research on AI projects, the dedicated budgets are available for 78
projects. Projects that received funding through Horizon 2020 amount to 29 of the 78
projects. The average budget dedicated to an AI project is ca. EUR 7.8 million90, which
refers to an average of the total budgets found. The budgets of AI projects are also
categorised according to dates (i.e., multi-annual budgets for several years) and un-
dated lump sums. For the average calculations, the total sum of the budget was used.
To further complement the budgetary findings, the average AI project budget amount
broken down by public sector area is shown in Figure 1991 below.

Figure 19 Average spending on AI projects classified by public sector policy area

Source: Authors’ elaboration.

The outlier in the graph refers to Public services for citizens (COFOG General public
services). The reason for this, as will be shown in Figure 20 below, is the number of
projects that fall into the AI chatbot category, which is the most popular AI project as
was also found in the section Overview of AI projects in the public sector.

Furthermore, the average budget spent according to AI typology within the AI project
analysis is depicted in Figure 20 below. For the average calculations, the total sum of
the budgets was used.

90
With an exact amount of EUR 7,846,566,72.
91
Please note that for the following public sector policy areas, budgets are not at all available: Economic
affairs; Education; Employment; Judicial system; Tax.

33
 Adopt AI Study

Figure 20 Average spending on AI projects classified by AI typology

Source: Authors’ elaboration.

The limited data available and lack of granularity of quantitative records (budgets) on
AI in public procurement can only provide the first insight into governmental
investments and current budgets for AI technologies and associated public sector
policy areas. The following analysis aims to complement these insights by delving into
the readiness of AI technologies when observed in their overall performance at the
market level.

Main findings/implications
➔ Preliminary indication of spending (to be noted that the findings are not
statistically significant) on AI projects, seen from the national initiatives and
projects point of view (both public sector area and AI typology for the
latter).
➔ The currently implemented AI projects budget is highest in the following four
public sector policy areas:
o General public services (Public services for citizens)
o Public order and safety (Police services)
o Housing and community amenities (Water supply)
o Social protection (Old age)
➔ The currently implemented AI projects’ budget is highest for the four
following AI typologies:

34
 Adopt AI Study

o Chatbots, Intelligent Digital Assistants, Virtual Agents and

Recommendation Systems
o Machine Learning, Deep Learning
o Computer Vision and Identity Recognition
o Cognitive Robotics, Process Automation and Connected and
Automated Vehicles

1.4.3 Market analysis on sector-level dynamics and AI technologies

uptake
This section reports selected innovation-related indicators aiming to provide a high-
level context of the private sector counterparts of the public sector areas, as AI
technologies adoption and their relative readiness intertwine with the sectoral
dynamics where these AI technologies are used. Different sources look at AI
adoption at a sectoral level, which is a central indicator for our analysis. Together
with this indicator, two other innovation-related indicators are reported below: Digital
intensity, defined as the degree to which the different economic sectors have
adopted digital technologies, and R&D intensity measured as expenditures by firms
on R&D (ratio of a firm’s R&D investment to its revenue).

AI adoption at the sectoral level

According to the findings of a global scale survey carried out by O’Reilly (2021)92, the
leading sectors for AI adoption at the company level are the ICT, financial services,
healthcare, and education sectors93. The following industries were found to have
adopted AI, as shown in Figure 21.

92
O’Reilly. 2021. AI Adoption in the Enterprise 2021. URL: https://www.oreilly.com/radar/ai-adoption-in-
the-enterprise-2021/
93
Though ‘other’ sectors have scored high, the survey report does not disclose the sectors that were
uncategorised.

35
 Adopt AI Study

Figure 21 AI adoption per industry, 2021

Source: O’Reilly. 2021. AI Adoption in the Enterprise 2021.94

In the same survey carried out by O’Reilly (2021)95, respondents were asked to map
their industry’s AI maturity levels on a three-level scale ranging from considering to
evaluating to mature. The top four sectors reporting mature practices are financial
services; telecommunications; retail; computers, electronics, and technology.
An additional finding is that the public sector and government score the highest for
evaluating the adoption of AI, as well as considering doing so.

According to a PwC study (2017) on the economic impact of AI by 2030 96, the sectors
that show the highest potential impact for AI adoption, listed from highest to lowest
are healthcare; automotive; financial services; transportation and logistics,
technology, communications, and entertainment; retail; energy; and
manufacturing.

As a third source about AI adoption at a sectoral level, the IDC Worldwide Artificial
Intelligence Spending Guide (2021), in Europe, reports banking and manufacturing
as the highest spending on AI solutions in the forecasted period until 2025, while
healthcare spending growing the fastest in that term. Retail AI spending growth is
expected in the 2021-2025 term as well.97

94
O’Reilly. 2021. AI Adoption in the Enterprise 2021. URL: https://www.oreilly.com/radar/ai-adoption-in-
the-enterprise-2021/
95
O’Reilly. 2021. AI Adoption in the Enterprise 2021. URL: https://www.oreilly.com/radar/ai-adoption-in-
the-enterprise-2021/
96
PricewaterhouseCoopers. 2017. Sizing the prize. Global Artificial Intelligence Study: Exploiting the AI
Revolution. URL: https://www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-
study.html
97
International Data Corporation (IDC). 2021. European Spending on Artificial Intelligence Will Reach $22
Billion in 2022, Supported by Strong Investments Across Banking and Manufacturing. URL:
https://www.idc.com/getdoc.jsp?containerId=prEUR148297521

36
 Adopt AI Study

As part of research carried out for DG CNECT by Ipsos (2020)98, in the EU, AI adoption
is most common among enterprises in the ICT sector, education, human health,
social work, and manufacturing. While the lowest AI adoption is in the sectors of
waste management, construction, transport, and food. The reason for this latter
finding is arguably the lower relevance of using AI in these sectors.99

Main findings/implications
➔ The global AI market adoption indicator analysis forecasts the following four
sectors to grow the most:
o Financial services
o ICT
o Healthcare
o Transport
➔ The forecast for the EU according to the AI adoption indicator is expected
especially in the two sectors below (chosen based on varying findings of
Europe-level data):
o Healthcare
o Manufacturing

Other indicators supporting the analysis of AI uptake

Digital intensity

The Digital Intensity Index measures the use of different digital technologies at the
enterprise level.100 This indicator is adopted by the European Commission (e.g.
Eurostat) as well as worldwide to measure the general degree to which an economic
sector has adopted digital technologies. This indicator is sensible for posting similar
results at a sectoral level as the one looking at the Adoption of AI. Despite broader,
digital intensity speaks about an important condition for AI adoption.

Based on the Eurostat Digital economy and society database, ‘Digital intensity’ for
EU27, in several available NACE-classified sectors, measured according to the
percentage of enterprises, is reported for Enterprises with high (DII version 3) and
very high digital intensity index (DII version 3), as per Figure 22 and Figure 23
below.

98
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
99
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368
100
Eurostat. 2021. How digitalised are EU’s enterprises? URL: https://ec.europa.eu/eurostat/web/products-
eurostat-news/-/ddn-20211029-1

37
 Adopt AI Study

Figure 22 Digital intensity per sector, 2021 (very high digital intensity index)

Source: Authors’ elaboration based on Eurostat 2021 data.

Note: *Not an official NACE category.

Figure 23 Digital intensity per sector, 2021 (high digital intensity index)

Source: Authors’ elaboration based on Eurostat 2021 data.

Note: *Not an official NACE category.

The most digitally intense sectors in the EU27 are information and communication;
the ICT sector; accommodation; and electricity, gas, steam, and air
conditioning supply. It is to be noted, Eurostat does not provide digital intensity
measurements for several sectors101. For this reason, further research was carried out.

In a study carried out by Mucha and Seppala (2021)102, the following digital intensity
scores were attributed to economic sectors, as shown in Figure 24 below.

101
Agriculture, forestry and fishing; mining and quarrying; financial and insurance activities; professional,
scientific and technical activities; public administration and defence, compulsory social security; education;
human health and social work activities; arts entertainment and recreation; other service activities;
activities of households as employers, undifferentiated goods- and services-producing activities of
households for own use; activities of extraterritorial organisations and bodies.
102
Mucha, T. and Seppälä, T., 2021. Estimating firm digitalization: A method for disaggregating sector-level
digital intensity to firm-level. MethodsX, 8, p.101233. URL:
https://www.sciencedirect.com/science/article/pii/S2215016121000261

38
 Adopt AI Study

Figure 24 Digital intensity score per sector, 2021

Source: Mucha and Seppala, 2021, Table 1: Reference data for sector-level digital intensity scores.103

Note: ISIC is a classification structure of economic activities (industries) based on a set of internationally agreed
concepts, definitions, principles and classification rules.

Note: According to the authors, firms with revenue-weighted digital intensity scores below 0.386 are classified as low
digital intensity, those with scores above 0.568 are classified as high digital intensity, and those in between are medium
digital intensity.

103
According to the methodology developed by Calvino and colleagues, the position of a given sector j in the
summary classification is computed as an average of sector j’s position in each considered dimension of the
digital transformation, where the weight is 1/36, i.e. the total number of sectors included in the taxonomy.
These sector and indicator-specific scores (i.e. a sector’s position divided by 36) are then averaged across
indicators, to yield one value per sector. More information: https://www.oecd-
ilibrary.org/docserver/f404736a-
en.pdf?expires=1647547698&id=id&accname=guest&checksum=21B7A687865053306A0ECD363169EF44

39
 Adopt AI Study

Though the digital intensity scores are calculated differently, the above findings
provide insight into the economic sectors that are not studied by Eurostat. The four
sectors with the highest digital intensity scores according to the above figure are
telecommunications; IT and other information services; finance and
insurance; and advertising and market research, other business services.

Main findings/implications
➔ Based on the analysis of the indicator on the digital intensity of the sectors,
the following four sectors perform best:
o ICT
o Financial services
o Electricity (not considered for this study)
o Advertising (not considered for this study)
➔ The finding that the two most digitally-intense sectors are ICT and Financial
services aligns with the findings of the AI adoption indicator at the global
level, however, Healthcare sector data lacks Eurostat findings to carry out
an aggregation of that sector.

R&D intensity

The R&D intensity is measured by the expenditures by firms on R&D, as the ratio of a
firm’s R&D investment to its revenues. When relevant, R&D investment growth will
also be analysed as an indicator as part of the economic analysis. R&D intensity serves
as a good proxy to indicate where innovation (including possibly AI) can take place
because it reflects where investments in innovation are done. In addition, R&D
intensity also expresses the pre-conditions for technologies such as AI to be
sustainably adopted. Sustainable R&D is defined as investments made by both private
and public sectors that are translated into commercialised technologies. This analysis
supports the understanding of which sectors are the readiest for large-scale AI
deployment.

In 2017-18, as found by McKinsey104, the R&D spend on the automobile industry was
highest in Europe. According to data gathered from Eurostat (2019) 105, the following
Figure 25 shows business enterprise expenditure on R&D per NACE sector.

104
McKinsey Global Institute. 2019. Reviving innovation in Europe. https://www.mckinsey.com/featured-
insights/innovation-and-growth/reviving-innovation-in-europe
105
The Eurostat data lacks information for a number of Member States in each sector.

40
 Adopt AI Study

Figure 25 R&D business expenditure, 2019

Source: Authors’ elaboration based on Eurostat 2019 data.

The above graph shows that in the EU, the largest R&D business expenditure was in
the Manufacturing; Professional, scientific and technical activities; Information and
communication; Wholesale and retail trade, and repair of motor vehicles and
motorcycles sectors.

To further analyse these findings with up-to-date data, as studied by the 2021 EU
Industrial R&D Investment Scoreboard 106, these findings partially align with the
findings on R&D investments' yearly growth rate in 2020. The highest R&D
investments in the EU were made for automobiles and other transport, followed by
health industries, ICT producers and ICT services, and can be observed in Figure
26.

106
https://iri.jrc.ec.europa.eu/scoreboard/2021-eu-industrial-rd-investment-scoreboard ; European
Commission. 2021. The 2021 EU Industrial R&D Investment Scoreboard. URL:
https://iri.jrc.ec.europa.eu/sites/default/files/contentype/scoreboard/2021-
12/EU%20RD%20Scoreboard%202021%20FINAL%20online.pdf

41
 Adopt AI Study

Figure 26 R&D investment per sector, EU, 2020

Source: European Commission, JRC/DG RTD, Figure S5. 2021. The 2021 EU Industrial R&D Investment Scoreboard.

These findings are also confirmed by the EIB107, which analysed the innovation
indicator, pointing out that the manufacturing sector of the EU has the largest share of
‘new to the world’s innovation. The EIB also underlines that the manufacturing sector
also has the largest average investment share for R&D.

Europe is a leading innovator in the automobile, health, financial and ICT sectors,
given the relative size of these sectors and their economic importance for the EU. This
is further confirmed by the European Commission108. It is also generally in line with
the strategic priorities of the EU in terms of innovation and growth. 109 As part of the
ongoing InvestEU programme110, the EU aims to strengthen investments for better
connectivity in the sectors of health, education, transport, logistics, and media, while
ensuring the widespread adoption of new technologies, among which AI.

Main findings/implications
➔ Based on the analysis of the R&D indicator, the following sectors are the
best performers:
o Healthcare
o ICT
o Financial services
o Transport

107
European Investment Bank. 2021. EIB Investment Survey. URL:
https://www.eib.org/attachments/publications/eibis_2021_european_union_en.pdf
108
https://ec.europa.eu/growth/industry/strategy/advanced-technologies/industrial-applications-artificial-
intelligence-and-big-data_en
109
Communication from the Commission to the European Parliament, the European Council, the Council, the
European Economic and Social Committee and the Committee of the Regions. A New Industrial Strategy for
Europe. COM/2020/102 final. URL: https://eur-lex.europa.eu/legal-
content/EN/TXT/?qid=1593086905382&uri=CELEX%3A52020DC0102
110
https://europa.eu/investeu/home_en

42
 Adopt AI Study

Main findings/implications of sectoral analysis

➔ Based on the market analysis of sectoral dynamics, the four most ready
sectors for AI adoption are the ICT sector; Financial services; Healthcare;
Transport.

1.4.4 Uptake of AI technologies by AI typology

This report will adapt the AI Watch classification when looking at AI technologies. The
AI Watch AI typologies classification is very detailed and includes multiple technologies
under each category. To increase the clarity of the analysis hereafter, the AI Watch
classification was compared to AI technologies that are commonly used in the
literature on AI industry analysis, as shown in Figure 27 below111.

Figure 27 Correspondence of AI typologies from AI Watch to Market-based classification

Source: Authors’ elaboration.

This section of the report conducts a market analysis of the uptake of specific AI
technologies by the private sector to identify those technologies considered mature by
the market by looking at three key indicators: market growth, market adoption and
investments. Shortlisted AI technologies will be then further explored by looking at
their adoption at the sectoral level. Figure 28 below summarises the approach for
studying the uptake of AI technologies and the funnel that will allow informing the
selection of the four sectors for the remainder of this study’s analysis.

111
Due to the lack of findings on Expert and Rule based Systems, Algorithmic Decision Making, this category
is not part of the analysis.

43
 Adopt AI Study

Figure 28 Approach for selection of sectors based on AI technologies analysis

Source: Authors’ elaboration.

Note: *This is only one dimension, complementary to other criteria of analysis.

The market growth rate allows for assessing the progress of a business, a product or
in this case, a technology. Figure 29 below presents the forecasted market size of
specific AI technologies in 2026, along with their respective Compound Annual Growth
Rates (CAGR). It is compiled based on different data sources and makes assumptions
based on CAGR calculation for missing data.

Figure 29 Forecasted market size by AI technologies, 2020-2026

Source: Authors’ elaboration based on secondary data collection.112

112
https://www.statista.com/statistics/1133875/global-voice-recognition-market-size/; Markets and
Markets. 2021. Chatbot Market by Component, Type (Rule Based & AI Based), Application (Customer
Service, Customer Engagement & Retention), Channel Integration, Business Function (ITSM, Finance),
Vertical, and Region - Global Forecast to 2026. URL: https://www.marketsandmarkets.com/Market-
Reports/smart-advisor-market-72302363.html;
KBV Research. 2021. Global Smart Robots Market By Component (Hardware and Software), By Mobility
(Mobile and Stationary), By Operating Environment (Ground and Underwater), By End User (Military &
Defense, Logistics Management, Inspection & Maintenance, Field/Agricultural, Healthcare, Industrial,
Domestic and Others), By Region, Industry Analysis and Forecast, 2020 – 2026/ URL:
https://www.kbvresearch.com/smart-robots-market/;
KBV Research. 2020. Global Computer Vision Market By Product Type (PC-Based and Smart Camera-Based),
By Component (Hardware and Software), By Application (Quality Assurance & Inspection, Measurement,

44
 Adopt AI Study

The technologies with the largest forecasted market size in 2026 are Natural
Language Processing (NLP), Predictive Analytics, Speech Recognition and
Deep Learning (or machine learning). The latter being the technology itself,
embedded in all other categories that are application fields of such technology, will be
considered as a stand-alone category that will have the purpose of confirming the
findings.

Market adoption indicates the pace at which new technology is acquired and used by
the market. A study conducted by Narrative Science (2019)113 identified the most
widely used AI-powered solutions. The findings are summarised in Figure 30 below.

Figure 30 Most widely used AI-powered solutions, 2019

Source: Narrative Science. 2019. Outlook on Artificial Intelligence in the Enterprise. 114

In alignment with the findings in terms of CAGR despite being in a slightly different
order, the AI technologies with the highest market adoption rate in 2018 were:
Predictive Analytics, Natural Language Processing, Voice Recognition and as a
consequence, Machine Learning. Moreover, the Hype Cycle for AI by Gartner
(2021)115 presented in Figure 31 below, confirms that Natural Language Processing

Identification, Predictive Maintenance, Positioning & Guidance and 3D Visualization & Interactive 3D
Modeling), By Vertical (Industrial and Non-Industrial), By Region, Industry Analysis and Forecast, 2020 –
2026. URL:
https://www.kbvresearch.com/computer-vision-market/; Orbis Research. 2022. Global Knowledge
Management Software Market Growth (Status and Outlook) 2022-2028. URL:
https://www.orbisresearch.com/reports/index/global-knowledge-management-software-market-growth-
status-and-outlook-2022-2028; BCC Research. 2020. Deep Learning: Global Markets. URL:
https://www.bccresearch.com/market-research/information-technology/deep-learning-market.html;
Mordor Intelligence. Year. Natural Language Processing (NLP) Market – Growth, Trends, COVID-19 Impact,
and Forecasts (2022 - 2027). URL:
https://www.mordorintelligence.com/industry-reports/natural-language-processing-market; Markets and
Markets. 2021. Predictive Analytics Market with Covid-19 Impact Analysis by Solution (Financial Analytics,
Risk Analytics, Marketing Analytics, Web & Social Media Analytics), Service, Deployment Mode, Organization
Size, Vertical, and Region - Global Forecast to 2026. URL: https://www.marketsandmarkets.com/Market-
Reports/predictive-analytics-market-1181.html; Mordor Intelligence. 2021. Threat Intelligence Market –
Growth, Trends, Covid-19 impact, and Forecasts (2022- 2027).
https://www.mordorintelligence.com/industry-reports/threat-intelligence-market
113
Narrative Science. 2019. Outlook on Artificial Intelligence in the Enterprise. URL:
https://narrativescience.com/wp-content/uploads/2019/02/Research-Report_Outlook-on-AI-for-the-
Enterprise.pdf
114
Narrative Science. 2019. Outlook on Artificial Intelligence in the Enterprise. URL:
https://narrativescience.com/wp-content/uploads/2019/02/Research-Report_Outlook-on-AI-for-the-
Enterprise.pdf
115
Gartner. 2021. The 4 Trends That Prevail on the Gartner Hype Cycle for AI, 2021. URL:
https://www.gartner.com/en/articles/the-4-trends-that-prevail-on-the-gartner-hype-cycle-for-ai-2021

45
 Adopt AI Study

and Deep Learning will reach the plateau of productivity, representing the beginning
of mainstream adoption soon (i.e. 2 to 10 years).

Figure 31 Hype Cycle for Artificial Intelligence, 2021

Source: Gartner. 2021. The 4 Trends That Prevail on the Gartner Hype Cycle for AI.116

The investment being a forward-looking operation, according to estimates by

Statista117, these two technologies are also among those receiving the most worldwide
funding in 2019. This further suggests that both technologies are ready to be invested
in today to meet the demand expected soon when the plateau will be reached. It must
be noted that Statista also reports Smart Robots and Computer vision being high in
terms of AI funding received, nonetheless, their rates of market growth are not as
high and their levels of market adoption are currently not available from the literature
analysed.

Based on these key market indicators, namely forecasted market growth rate and
market adoption rate, it can be stated that the readiest technologies to implement and
therefore to invest in AI technologies are: Predictive Analytics, Natural Language
Processing and Voice Recognition, as defined by the AI Watch classification. These
findings are also a good indication of the maturity level, intended as the level of
readiness of a technology to be implemented at a large scale, of the respective
technologies.

116
Gartner. 2021. The 4 Trends That Prevail on the Gartner Hype Cycle for AI, 2021. URL:
https://www.gartner.com/en/articles/the-4-trends-that-prevail-on-the-gartner-hype-cycle-for-ai-2021
117
Statista. 2019. Machine Learning Tops AI Dollars. URL: https://lb-aps-
frontend.statista.com/chart/17966/worldwide-artificial-intelligence-funding/

46
 Adopt AI Study

Taking these findings one step further, it can be noticed that the three AI technologies
that we have identified are also believed to be determinants of increasing labour
productivity, contributing to an overall higher generating economic potential.

For instance, Statistical Analysis System (n.d.)118 argues that Predictive Analytics, by
improving organisations’ operations, enables them to function more efficiently and
ultimately increases overall labour productivity. Concerning Natural Language
Processing, Wolters Kluwer (2020)119 argues that this technology leads to a boost in
productivity and overall improved quality in the employees’ work as it won’t replace
the workforce but rather change the way of working. As for Speech Recognition,
Forbes120 states that this technology can increase overall productivity in many
businesses such as in healthcare. In more general terms, the World Economic Forum
(2022)121 argues that Deep Learning can increase productivity as it can process
amounts of data that humans can’t and it “works like the human brain”, in the sense
that it learns from examples.

Furthermore, as previously shown under the An introduction to AI adoption, there is

evidence of a positive relationship between AI technology and employment, however,
similar evidence also exists against this statement, therefore implying that the
employment effect of AI remains controversial. Moreover, there are currently no
reliable data in the literature about the impact of specific AI technologies on
employment. In this regard, the only exception is provided by Deloitte (2015)122,
which claims that AI technology, in particular Machine Learning, will lead to overall job
creation, arguing that the common discourse on the job disruption effect of AI is
biased due to “the relative unpredictability of its creative aspects”. This finding can
ultimately be extended to other AI technologies as most of them rely on machine
learning algorithms, however, no study or evidence has been found in the literature to
confirm this statement. In this regard, the AI Watch also confirms that Machine
Learning is a broader AI category and that most AI technologies use some form of it.

Main findings/implications
➔ The AI technologies that have the highest economic potential and are the
readiest to be implemented based on market growth, adoption, productivity
and employment are Predictive Analytics; Natural Language Processing and
Voice/Speech Recognition.

AI technologies and sector-level analysis

This section will analyse how, the three identified as most mature technologies,
namely Predictive Analytics, Natural Language Processing and Voice

118
Statistical Analysis System (SAS). N.d. Predictive analytics. What it is and why it matters. URL:
https://www.sas.com/en_au/insights/analytics/predictive-analytics.html
119
Wolters Kluwer. 2020. How natural language processing impacts professions. URL:
https://www.wolterskluwer.com/en/expert-insights/how-natural-language-processing-impacts-professions
120
Forbes. 2021. A Market to Harness: Speech Recognition Artificial Intelligence (AI) Innovations On The
Rise. URL: https://www.forbes.com/sites/cindygordon/2021/12/23/a-market-to-harness-speech-
recognition-artificial-intelligence-ai-innovations-on-the-rise/?sh=5fe17773134d
121
World Economic Forum. 2022. How deep learning can improve productivity and boost business. URL:
https://www.weforum.org/agenda/2022/01/deep-learning-business-productivity-revenue/
122
Deloitte. 2015. Technology and people: the great job-creating machine. URL:
https://www2.deloitte.com/content/dam/Deloitte/uk/Documents/finance/deloitte-uk-technology-and-
people.pdf

47
 Adopt AI Study

Recognition are uptaken up by different market sectors. To ultimately confirm the

findings, the same analysis will also be applied to the general use of Machine
Learning. The analysis will be functional, together with other variables, to inform the
choice of the four priority sectors that are ready to uptake AI on a large scale in the
public sector.

Predictive Analytics

According to Allied Market Research’s report (2020)123, the sectors that will hold the
largest market share of Predictive Analytics technology by 2027 are Banking, Financial
Services and Insurance (BFSI), Retail, IT & Telecom and Manufacturing. These results
are shown in Figure 32 below.

Figure 32 Forecasted Predictive Analytics market share, 2019-2027

Source: Allied Market Research. 2020. Predictive Analytics Market. 124

This significant increase in the adoption of Predictive analytics is due to this

technology being largely used by organisations for conducting operations such as
customer analytics, risk reporting, threat management, product innovation and
enhancing customer experience by automating business operations and processes.

Natural Language Processing

According to KBV Research (2019), BFSI and IT&Telecom are the sectors that hold the
largest market share for Natural Language Processing technology, followed by
Healthcare and Retail&Commerce, see Figure 33.

123
Allied Market Research. 2020. Predictive Analytics Market by Component (Solution and Services),
Deployment (On-premise and Cloud), Enterprise Size (Large Enterprises and Small & Medium-sized
Enterprises), and Industry Vertical (BFSI, Retail, IT & Telecom, Healthcare, Government, Manufacturing,
and Others): Global Opportunity Analysis and Industry Forecast, 2020–2027. URL:
https://www.alliedmarketresearch.com/predictive-analytics-market
124
Allied Market Research. 2020. Predictive Analytics Market by Component (Solution and Services),
Deployment (On-premise and Cloud), Enterprise Size (Large Enterprises and Small & Medium-sized
Enterprises), and Industry Vertical (BFSI, Retail, IT & Telecom, Healthcare, Government, Manufacturing,
and Others): Global Opportunity Analysis and Industry Forecast, 2020–2027. URL:
https://www.alliedmarketresearch.com/predictive-analytics-market

48
 Adopt AI Study

Figure 33 Natural Language Processing market share, 2019

Source: KBV Research. 2019. Market Research Report.125

Being an easy-to-adapt and cost-effective cloud platform, NLP is found to be more

suitable for professional services.

Speech recognition

According to Allied Market Research (2020)126, Figure 34, Healthcare will be the sector
holding the largest market share of Speech Recognition technology by 2026, followed
by the Automotive sector.

Figure 34 Forecasted Speech Recognition market share, 2018-2026

Source: Allied Market Research. 2020. Speech Recognition Market.127

125
KBV Research. 2019. Global Natural Language Processing Market By Component (Solution and Services)
By Application (Text Classification, Machine Translation, Question Answering, Sentiment Analysis,
Information Extraction, Automatic Summarization and Others) By type (Rule Based, Statistical and Hybrid)
By Deployment Type (On-premise and Cloud) By Region, Industry Analysis and Forecast, 2019 – 2025. URL:
https://www.kbvresearch.com/natural-language-processing-market/
126
Allied Market Research. 2020. Speech Recognition Market by Deployment Mode (On Cloud and On
Premise) and End Use (Consumer Electronics, Healthcare, Enterprise, and Others): Global Opportunity
Analysis and Industry Forecast, 2019–2026. URL:
https://www.alliedmarketresearch.com/speech-recognition-market
127
Allied Market Research. 2020. Speech Recognition Market by Deployment Mode (On Cloud and On
Premise) and End Use (Consumer Electronics, Healthcare, Enterprise, and Others): Global Opportunity
Analysis and Industry Forecast, 2019–2026. URL:
https://www.alliedmarketresearch.com/speech-recognition-market

49
 Adopt AI Study

Moreover, by looking at the US market volume, Figure 35, it can be observed that
aside from Healthcare and excluding Consumers as it is not part of the NACE
classification, the other sectors that are predicted to increase their use of Voice
Recognition technology are Education, BFSI and Automotive.

Figure 35 Forecasted US Voice Recognition market share, 2014-2025

Source: Grand View Research. 2019. US Voice Recognition market.128

Machine Learning, Deep Learning

According to Grand View Research’s Global Industry Report (2019)129, the sectors that
had the greater market share of Machine Learning and therefore where this
technology is mostly deployed are Advertising & Media, BFSI and Healthcare, followed
by Retail and Automotive & Transportation. These findings are presented in Figure 36
below.

Figure 36 Machine Learning market share, 2019

Source: Grand View Research. 2019. Market Analysis Report.130

128
Grand View Research. 2019. U.S. Voice Recognition Market Size, Share & Trends Analysis Report By
Vertical (Automotive, Enterprise, Consumer, Banking, Financial Services & Insurance, Government, Retail,
Healthcare, Military, Legal, Education), And Segments Forecasts, 2019 – 2025. URL:
https://www.grandviewresearch.com/industry-analysis/us-voice-recognition-market
129
Grand View Research. 2019. Machine Learning Market Size, Share & Trends Analysis Report By
Component, By Enterprise Size, By End Use (Healthcare, BFSI, Law, Retail, Advertising & Media), And
Segment Forecasts, 2019 – 2025. URL: https://www.grandviewresearch.com/industry-analysis/machine-
learning-market
130
Grand View Research. 2019. Machine Learning Market Size, Share & Trends Analysis Report By
Component, By Enterprise Size, By End Use (Healthcare, BFSI, Law, Retail, Advertising & Media), And

50
 Adopt AI Study

Healthcare is expected to take the lead and holds the largest Machine Learning market
share by 2025. This is due to Machine Learning technology being used for quantitative
insights for better diagnosis and using it to prevent diseases, hence “moving the field
of medicine from reactive to proactive” (Grand View Research, 2019) and this will
drive the market.

By combining these last findings altogether and looking at the overall picture, it can be
concluded that the sectors that are the readiest to implement the chosen AI
technologies (Predictive Analytics, Natural Language Processing and Speech/Voice
Recognition) are: BFSI, Healthcare, Retail, IT&Telecom and Automotive. However,
since the ultimate purpose of this report is to identify four priority public sector areas
where accelerated public procurement could bring the highest EU-added value, retail
can be dropped from the current selection as it is not relevant from a public sector
perspective. Therefore, the suggested selection of the most relevant sectors focusing
on the uptake of AI technologies by the market is BFSI, Healthcare, IT&Telecom
and Automotive. This finding will be triangulated with additional criteria that
combined will provide the final selection of the four public sector areas for the current
study.

Main findings/implications
➔ Based on the market analysis of technology dynamics, the sectors that are
most mature and have the highest economic potential when adopting the
previously chosen AI technologies are Banking, Financial Services and
Insurance (BFSI); Healthcare; IT and Telecom; and Automotive.

1.5 Final selection of public sector policy areas

Within this scoping report, the i) analysis of the state of play of public procurement of
AI, the ii) analysis of the barriers and benefits for the adoption of AI, as well as iii)
market analysis, provided different perspectives and information valuable for a
selection of the four key sectors that are most ready for large-scale deployment of AI
technologies in the public sector. The selection of the four key sectors is, hence, based
on the information coming from these three high-level criteria. Figure 37 further
explains how these three different perspectives inform the selection.

Segment Forecasts, 2019 – 2025. URL: https://www.grandviewresearch.com/industry-analysis/machine-

learning-market

51
 Adopt AI Study

Figure 37 Criteria for selection of the key sectors

Source: Authors’ elaboration.

Note: *External factors analysed to understand what factors pose challenges/benefits for the uptake of AI.

Firstly, the qualitative and quantitative analyses of governmental AI initiatives

have contributed to providing an overview of governments’ priority policy areas for the
adoption of AI; and the analysis of AI projects to the current state of play of adopted
AI policy areas and technologies.

The economic analysis, in turn, contributed to the selection of the four key sectors
by assessing the sectors’ digital maturity levels by looking at AI adoption levels, the
digital intensity, and the R&D intensity of different sectors. The economic analysis
further looked at the economic potential of AI technologies according to their
market growth, adoption rates, labour productivity impact and employment
generation.

Analysis was also carried out on the barriers to and the benefits of the adoption
of AI in the public sector. This contributed to a better understanding of which
challenges are to be overcome by the public sector. The observed benefits and the
tools to support their achievements contribute to further considerations that can be
made at the public sector level. Since these findings provided higher-level and
horizontal explanations, they are not used to directly inform the selection of the
priority sectors but rather work as complementary to the above. Specifically, the
benefits, when materialised, should also be a measure of the European added value of
increased large-scale deployment of AI in the public sector areas.

Based on the above, the proposed selection of the four key sectors that are ready
for the large-scale deployment of AI are listed below and depicted in Figure 38.

52
 Adopt AI Study

Figure 38 Four key sectors

Source: Authors’ elaboration.

Health (COFOG Health / NACE Q Human Health and Social Work Activities): the
health sector has prevailed as a recurring key priority area for both the public sector
and the private sector. The adoption of AI technologies may benefit the public sector
by providing better health services, and the private sector to enhance the provision of
medical products, appliances, and equipment for public procurement.
According to the AI typology sector analysis, the health sector is the one expected to
take the lead in the adoption of Deep Learning and Speech Recognition technologies.
In the AI project mapping, it was found that there are also other existing technologies
in the sector such as cognitive robotics in addition to those mentioned.
Mobility (COFOG Economic Affairs / NACE H Transportation and Storage): the
mobility sector implies multiple facets related to transportation,
infrastructure/construction, and the manufacturing of vehicles. Both the public and
private sectors can benefit from the adoption of AI technologies in these. For instance,
the public sector may use AI to achieve smart cities’ objectives, infrastructure
monitoring, and better provision of public transport. The private sector may continue
to further enhance manufacturing by providing the underlying AI technologies (e.g. in
automotive) to the public sector. With Industry 4.0 (part of the Digital Single Market
Strategy131), governments in collaboration with the private sector may make Europe a
front-runner in the mobility sector.
Specifically, the AI technology that showed a particularly high rate of adoption in the
private automotive sector, which can be extended to the wider mobility category, is
Speech recognition. In the AI projects mapping, further technologies were also found
such as computer vision, predictive analytics, and process automation.
E-Government (COFOG General public services / NACE O Public administration and
defence; compulsory social security): the public sector aims to build the capacity to
seize the advantage of AI in policymaking and public service delivery, as found in the
AI initiatives and public sector projects. Indeed, governments can benefit from the

131
European Parliament. 2015. Industry 4.0. Digitalisation for productivity and growth. URL:
https://www.europarl.europa.eu/RegData/etudes/BRIE/2015/568337/EPRS_BRI(2015)568337_EN.pdf

53
 Adopt AI Study

adoption of AI to optimise internal processes and provide modern public services. To

further break down governmental tasks, five categories were identified by the
authors: enforcement; regulatory research, analysis and monitoring; adjudication;
public services and engagement; internal management. 132 These categories form the
initial basis upon which the following in-depth sectoral research will build.
The currently in place AI typologies used for internal public administration processes
are, among others, text mining, Natural Language Processing, and audio processing.
As for public services, chatbots are the majority group of AI technology, followed by
Natural Language Processing and audio processing among others.
ICT is incremental to the development of e-Government and is an underlying priority
sector to all AI governmental initiatives (e.g. investments in the software industry
developed by the private sector). To further digitalise public administrations and
therefore public services, governments can procure modern technologies and invest in
building a solid digital infrastructure.
Natural Language Processing is the AI technology that, according to the AI
technologies analysis, will be adapted to the largest extent by the ICT sector, and is
also showcased by the public sector AI projects that use AI technologies for language
analytics.
Education (COFOG Education / NACE P Education): the education sector is identified
as one of the top priority sectors for governmental adoption of AI, as depicted in the
Member States’ AI initiatives. The same importance has also been stressed at the EU
level, with the European Parliament stating that “the public sector should be leading in
demystifying AI and offering free and open education about AI”133, to take the wider
public on board, develop human-centric AI and increase public sector readiness for AI.
Along the same line, as also noted in the Maltese AI Strategy, the adoption of AI will
contribute towards making informed decisions for the implementation of relevant
policies to achieve the targets set within the education strategy. According to the AI
technologies analysis, Voice Recognition is the AI technology that is expected to have
the highest adoption rate in the Education sector.
The following Table 2 depicts the level of importance of the different key sectors that
were identified throughout the scoping report, as per the indicator studied. More
specifically, the crosses indicate the extent to which these sectors appeared in the
analysis.

132
Engstrom, D. F., Ho, D. E., Sharkey, C. M., & Cuéllar, M.-F. 2020. Government by Algorithm: Artificial
Intelligence in Federal Administrative Agencies. In SSRN Electronic Journal. URL:
https://doi.org/10.2139/ssrn.3551505
133
European Parliament. 2021. Artificial Intelligence and public services. URL:
https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/662936/IPOL_BRI(2021)662936_EN.pdf

54
 Adopt AI Study

Table 2 Sector importance vis-a-vis studied indicators

Source: Authors’ elaboration.

Based on the methodology adopted, the original proposal as the fourth key sector was
FinTech, however, due to the lack of governmental inclusion and quantifiable economic
impact of the FinTech sector, it was replaced by the Education sector.
The final proposal for the focus sectors balances the information available on public
administration data and market data. The four sectors constitute the basis upon which
the following research of this study builds.

55
 Adopt AI Study

2 Chapter 2: Qualitative and Quantitative analysis of key

sectors
2.1 Introduction
Artificial Intelligence (AI) is already transforming the public sector and the numerous
policy areas within its orbit. Rather than simply a steady natural progression of
digitalization, a desirable end in itself, the continued and increased uptake and use of
AI in the public sector is set to usher in a paradigm shift in the public sector with
profound political, economic and social implications.
In lockstep with a robust regulatory framework, this area of technology has the
potential to improve public administration and services both equitably and ethically if
handled correctly. With the major inflexion point of the pandemic, the European Union
has the chance to lead by example and responsibly champion the use of AI while
making further progress on its goals of strategic autonomy and achieving the twin
transition. Furthermore, with increased competition from abroad in key technologies
such as AI, it is crucial that the EU not only keeps up but makes up ground in the
sector. This chapter closely examines AI and the public sector through the lens of four
key sectors to provide additional insights into its benefits, barriers and the current
policy landscape.
Though still in its infancy, especially compared with its private counterpart which was
responsible for 67% of total AI investment between 2018-20134, the public sector´s
approach, and its historically cautious relationship with AI is starting to shift. Having
focused initially on strategy and regulation, the funding and use of AI applications are
increasingly commonplace with some public authorities even starting to develop
solutions themselves. In fact, in addition to the significant contributions to the health
sector (which will be covered later in the chapter), the pandemic has increased the
ambition of EU member states when it comes to AI135. Aided by the digitalization
component of NextGenerationEU funds, all but three of the EU´s member states have
published a national AI strategy by the time of writing136. Reflective of the variation in
the technology domain itself, the areas of emphasis of each member state have been
considerably different both in terms of the type of action (e.g., regulation, R+I) but
also sectoral focus.
Although use and development are on the rise, the transformative potential of AI in
the public sector is yet to be realized with applications generally focused on the so-
called “low-hanging fruit” of the public portfolio. However, with more public actors
both procuring and funding AI technology as well as spearheading development itself,
the AI space will only grow more dynamic and competitive. This chapter explores and
analyses four landmark public sectors which are both strategically crucial for the EU
and ready to receive an injection of public investment and focus. These sectors are
eGovernment, mobility, health and education.
Aims and objectives of the chapter
This chapter is part of a study by the European Commission to improve the
understanding of AI and the public sector. Insights into the dynamics of this
everchanging space should help increase investment in and the use of AI in the public

134
Tatjana Evas, Maikki Sipinen, Martin Ulbrich, Alessandro Dalla Benetta, Maciej Sobolewski and Daniel
Nepelski, AI Watch: Estimating AI investments in the European Union, EUR 31088 EN, Publications Office of
the European Union, Luxembourg, 2022, ISBN 978-92-76-53433-4, doi:10.2760/702029, JRC12917
135
Van Roy, V., Rossetti, F., Perset, K. and Galindo-Romero, L., AI Watch - National strategies on Artificial
Intelligence: A European perspective, 2022 edition, , EUR 31083 EN Publications Office of the European
Union, Luxembourg,2022, ISBN 978-92-76-52910- 1, doi:10.2760/385851, JRC129123.
136
Van Roy, V., Rossetti, F., Perset, K. and Galindo-Romero, L., AI Watch - National strategies on Artificial
Intelligence: A European perspective, 2022 edition, , EUR 31083 EN Publications Office of the European
Union, Luxembourg,2022, ISBN 978-92-76-52910- 1, doi:10.2760/385851, JRC129123.

56
 Adopt AI Study

(and indeed the private) sector. This study aims to support the broader goals of AI
adoption by providing evidence and operational recommendations. More specifically,
this report aims to build on past work carried out on AI in the public sector and answer
the call for more “deep dives at the country level through case studies and thematic
analyses”137. This means expanding previous research in the area and its calls to
address the contextual factors, institutional capacities, external and internal drivers
and barriers that underlie AI implementation and the transformation of public
administrations138.
Within each sector, the report evaluates closely the principal drivers, barriers and
trends as well as identify some best practices from interesting real-world AI use cases.
While the analysis studies four key sectors, some lessons and insights can be applied
to other sectors which might be less mature in terms of AI diffusion.
Likewise, as discussed in the Peer Learning Workshops on AI in the public sector, it is
crucial to collect and share any best practices between EU governments. Finally, a
coherent approach to studying is central to providing a “common reference point for
identifying and better understanding the underlying pillars for relevant and meaningful
assessment in the area”139.
Chapter structure
In terms of structure, each sector is analysed with the same set of methods in a
uniform order. A full methodology section including definitions and explanations of the
data collection and quantitative techniques can be found as an annex below. Using a
mapping of the pan-EU policy landscape as a base, each section starts with an outline
of how AI can resolve a set of sector-specific problems and a value chain analysis
followed by an assessment of the drivers and barriers in the form of PESTEL and
SWOT analyses. This is followed by a case study examining an insightful and
successful example of public policy on AI technology. The final part of each section
features a quantitative analysis which uses linear regression to highlight correlations
between the level of a country´s AI activity and a range of related variables.
The subsections are as follows:
 Introduction and policy mapping. Setting out the scope of the sector,
highlighting typical use cases and a brief account of the progression of AI in the
area. This section will give an overview of the policies and initiatives in each
sector.
 Challenges and Solutions. Identification of key challenges and corresponding
solutions in the selected sectors that could be solved using AI applications.
 Value Chain Analysis. Identification of key links in each sector’s AI value
chain.
 Drivers and barriers. A PESTEL (Political, Economic, Social, Technological,
Environmental and Legal) and SWOT (Strength, Weaknesses, Opportunities and
Threats) analysis.
 Case Study. A real-world example which examines a specific policy that uses
or focuses on AI. This section also presents a series of best practices which
made the policy successful and should be replicated.

137
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services, EUR 30255 EN,
Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-19540-5 (online),
doi:10.2760/039619 (online), JRC120399.
138
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services, EUR 30255 EN,
Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-19540-5 (online
139
Manzoni, M., Medaglia, R. and Tangi, L., AI Watch. Artificial Intelligence for the public sector. Report of
the “4th Peer Learning Workshop on the use and impact of AI in public services”, 28 October 2021,
Publications Office of the European Union, Luxembourg, 2021, ISBN 978-92-76-46347-4,
doi:10.2760/142724, JRC127944.

57
 Adopt AI Study

 Economic Analysis. Validation of the findings of the earlier section through

available indicators and provide quantitative evidence for AI investment from
the public sector. This compares categories of AI public sector activity on a
range of variables.
2.2 eGovernment
eGovernment, or digital public services, is defined in this chapter as the process of
“organizational change to improve structures and operations of government” 140. For
the majority of this century, the technological driver for this progress has been
Information and Communication Technologies (ICT) which has enabled “citizens,
enterprises and organizations to carry out their interactions with government more
easily, more quickly and at lower cost” 141. The increased use of ICT in government has
not diminished and remains a key policy goal at both the national and European levels
as a central tenet of the Digital Decade – the European Commission´s vision for a
digitalised EU. However, AI is also set to bring about even more profound and
disruptive “organizational change” to the public sector and digital transformation
rather than a mere continuation of digitalization 142.
Several recent pieces of research highlight this AI-driven paradigm shift which is
underway within the public sector. Ahn and Chen make a categorical distinction
between “IT-enabled bureaucracy” and “AI-augmented bureaucracy”143. The former
refers to the type of eGovernment outlined above whereby statistics, computers and
the internet are used to improve electronic service delivery while the latter combines
5G and IoT cloud computing with machine learning and algorithmic decision-making to
harness real-time, large and high-quality data. This blend of advanced ICT
technologies with powerful AI techniques can provide enormous public value including
a detailed understanding of citizens' needs and solutions, enhanced simulation and
planning capability and ultimately “smart government”. Reis et al. also deploy the
same distinction (and indeed terminology), noting that “digitization and digital
transformation have been occurring in organizations since the 1950s” 144 but that AI,
with its reliance on real-time and real data” is game-changing digital transformation.
The authors also point to a Capgemini report which advocates the economic and social
benefits of AI to the public sector and that “AI helps us to enter a new era of
sophisticated and smart public services”145.
This first sectoral analysis uses the more expansive European Commission definition of
eGovernment as “digital public services” 146 rather than something inherently involving
ICT technology. Furthermore, to avoid overlap with later sections as much as possible,
this sector focuses on the category of “general public services” 147. Using the COFOG
classification, this includes Executive and legislative organs, financial and fiscal affairs,

140
Field, T., Muller, E., Lau, E., Gadriot-Renard, H. and Vergez, C. (2003) The Case for E-Government:
Excerpts from the OECD Report “The E-Government Imperative”. OECD Journal on Budgeting, 3, 61-
131.http://dx.doi.org/10.1787/budget-v3-art5-en
141
https://digital-strategy.ec.europa.eu/en/policies/egovernment
142
Reis, J., Santo, P.E., Melão, N. (2019). Artificial Intelligence in Government Services: A Systematic
Literature Review. In: Rocha, Á., Adeli, H., Reis, L., Costanzo, S. (eds) New Knowledge in Information
Systems and Technologies. WorldCIST'19 2019. Advances in Intelligent Systems and Computing, vol 930.
Springer, Cham. https://doi.org/10.1007/978-3-030-16181-1_23
143
Ahn, Michael J. and Yu-Che Chen. “Artificial Intelligence in Government: Potentials, Challenges, and the
Future.” The 21st Annual International Conference on Digital Government Research (2020)
144
Reis, J., Santo, P.E., Melão, N. (2019). Artificial Intelligence in Government Services: A Systematic
Literature Review. In: Rocha, Á., Adeli, H., Reis, L., Costanzo, S. (eds) New Knowledge in Information
Systems and Technologies. WorldCIST'19 2019. Advances in Intelligent Systems and Computing, vol 930.
Springer, Cham. https://doi.org/10.1007/978-3-030-16181-1_23
145
Tinholt, D., Carrara, W., Linden, N.: Unleashing the potential of artificial intelligence in the public sector.
Capgemini Consulting (2017)
146
European Commission (2023) “eGovernment and digital public services”: https://digital-
strategy.ec.europa.eu/en/policies/egovernment
147
Eurostat (2023) “COFOG”: https://ec.europa.eu/eurostat/statistics-
explained/index.php?title=Glossary:Classification_of_the_functions_of_government_(COFOG)

58
 Adopt AI Study

external affairs, Foreign economic aid, General services, Basic research, R&D general
public services, General public services n.e.c., public debt transactions, transfers of
general character between different levels of government. 148 According to the
Innovative Public Services database149, 30% of AI use cases (207 out of 686 cases)
fall within the “General Public Services” domain. A “large and comprehensive
category” per AI Watch150, some AI applications which are explored in more detail
include:
• Chatbots and virtual assistants: used to speed up internal processes and
interact externally with citizens and business
• Comparison, detection and misinformation handling management
• Classification, storage and search of documents (even-hand written), videos
and/or recorded speeches with metadata and information extraction
• Several kinds of data anomalies detection or potential fraud
The policy landscape: eGovernment and AI
As in other sectors, the current and potential uses for public administration are myriad
for all levels of governance, something which is reflected in the variety of policies
around the EU. Uses range from image and video recognition technology to analysing
huge volumes of data to make predictions which are more comprehensive and
accurate and/or support human or automated.
Figure 39 Policy breakdown per mode taken from the mapping exercise

Source: Authors’ elaboration.

The most numerous uses are currently chatbots which allow citizens to interact with
services in a semi-automated manner through virtual assistants or conducting
sentiment analyses based on the interpretation of textual data (Chui et al., 2018;
Eggers et al., 2017) supported by Natural Language Processing (NLP) to understand
audio and text. Chatbots have been implemented throughout the EU at both the
national and regional levels and span several different thematic areas, highlighting
their versatility. This includes assisting tourists through the Plovdiv City Concierge

148
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services (2020), EUR 30255
EN, Publications Office of the European Union, Luxembourg, ISBN 978-92-76-19540-5 (online),
doi:10.2760/039619 (online), JRC120399.
149
Innovative Public Services Explorer: https://ipsoeu.github.io/ips-explorer/case/
150
AI Watch (2022) European landscape on the use of Artificial Intelligence by the Public Sector.

59
 Adopt AI Study

(Bulgaria), immigration and business-focused chatbots in Finland and even retired

people and interdepartmental communication in France and Germany.
Leading the way in the number of AI-focused policies is Italy (15) with larger national
projects such as the IBM Watson content hub and Evasometro Anonimizzato, a tool
which uses big data to analyse taxpayer behaviour, combined with a series of regional
initiatives in areas such as Piemonte and Lombardia. Other member states have
enacted larger programmes to improve public services through AI. In the case of
Portugal, this takes the form of multiple AI-related measures within a broader
digitalization programme SIMPLEX or a programme strictly dedicated to AI in Public
services like Finland´s AuroraAI network (part of the wider AuroraAI programme)
which links the services of central government organisations together with services in
other sectors151. Below is a summary of the results of the policy mapping for
eGovernment in the EU. A full explanation of the data sources and the precise nature
of the categories used can be found in the methodology section in the annex below.
Figure 40 E-government and AI policies in the EU-27

eGovernment and AI Policies in the EU-27

16
14
12
10
8
6
4
2
0
Austria

Finland
France
Cyprus

Italy
Latvia

Romania
Greece

Netherlands

Spain
Ireland

Malta

Slovakia
Luxembourg
Estonia

Lithuania

Sweden
Denmark

Hungary
Bulgaria
Croatia

Poland

Slovenia
Belgium

Germany
Czechia

Portugal

Regulation Facilitator Funder User

Source: Authors’ elaboration.

2.2.1 Challenges and AI solutions for e-government

The recent literature on the benefits of AI adoption by the public sector uses 3
dimensions derived from Twizeyimana & Andersson´s taxonomy 152: Improved
administration thanks in part to more efficiency, improved public services and
improved social value through more trust and confidence. Building on this, AI Watch´s
2022 landscaping report forwarded this conceptualization into 3 value drivers of AI
cases: Improved administrative efficiency, improved public services and open
government capabilities.
While these AI applications solve highly prevalent problems, their value can also pose
a risk to the public good if not handled with prudence. Often two sides of the same
coin, this section will also outline the main risks that accompany each value driver of
AI adoption in the public sector. The 3 AI Watch drivers are explored below with the
addition of a 4th major value driver of AI in cybersecurity. It is worth noting that the

151
DigiFinland (2022) “AuroraAI national artificial intelligence programme”: https://digifinland.fi/en/our-
operations/aurora-ai-national-artificial-intelligence-programme/
152
Jean Damascene Twizeyimana, Annika Andersson, The public value of E-Government – A literature
review, Government Information Quarterly, Volume 36, Issue 2, 2019, Pages 167-178,

60
 Adopt AI Study

value drivers themselves are mutually reinforcing and many of the solutions can
provide value in multiple ways (e.g., a chatbot can improve the accessibility of
government and free up resources simultaneously).
Figure 41 Summary of challenges and AI solutions for digital public services

Public sector Provide quality Improve the Improve the Defend

challenge services with understanding of relationship against
limited resources citizens´ needs between the cyber
and solutions state and citizen threats

AI Value driver Improved Improved public More open More secure

administrative services government public
efficiency services

AI Applications Automized Data Recommendation Chatbots Cognitive

entry systems security
Intelligent Digital
analytics &
Open AI process Assistants
threat
automation
intelligence
systems

AI Techniques Speech AI-powered NLP, Intelligent NLP and

recognition, algorithms agents, affective machine
machine computing and learning
translation, visual big data
form completion
checking

Improved administrative efficiency

The ability of AI to decrease the administrative burden of the public sector is likely the
most recognizable value driver to organizations looking to adopt the technology. Often
referred to as low-hanging fruit, indicating that the improvement is accessible,
valuable though not deeply transformative, AI technology can "take over mundane
tasks, freeing workers for more valuable work”153 and release valuable public funds
and budgetary savings which can be deployed elsewhere. According to “AI Multiple”,
53% of state and local officials surveyed in the US had excessive paperwork burdens
that impacted their ability to get their work done 154. Furthermore, Deloitte estimate
that automation of US federal government employee tasks could save between 96.7
million and 1.2 billion hours annually155. The same study reports that automation and
AI have the potential to save between $3.3 billion and $41.1 billion.
Delving into the inventory of public sector use cases provided by AI Watch, 28% of
cases improved the management of public resources (the highest number in the
“improved administrative efficiency” dimension) 156. Interestingly, only 3% provided
cost-reduction indicating that meaningful budgetary savings highlighted in the Deloitte

153
European Commission, Joint Research Centre, Manzoni, M., Medaglia, R., Tangi, L., et al., AI Watch, road
to the adoption of artificial intelligence by the public sector : a handbook for policymakers, public
administrations and relevant stakeholders, Publications Office of the European Union,
2022, https://data.europa.eu/doi/10.2760/288757
154
AI Multiple (2023) “AI in government: Examples, Challenges and Best Practices”:
https://research.aimultiple.com/ai-government/
155
Deloitte (2017) “AI-augmented government”:
https://www2.deloitte.com/us/en/insights/focus/cognitive-technologies/artificial-intelligence-
government.html
156
AI Watch (2022) European landscape on the use of Artificial Intelligence by the Public Sector

61
 Adopt AI Study

report are yet to be realized. Multiple applications to help increase efficiency already
exist and have achieved a high level of maturity, developed by both the public and
private sectors, to be implemented by public authorities.
However, more efficiency brings with it the possibility of job losses in the private
sector as AI is used for tasks that were previously executed by humans. While this
trend is nothing particularly novel and the overall impact of AI on the labour market is
contested, “some categories of jobs in the public sector are destined to disappear” 157.
Mitigating and softening the loss of certain jobs, especially those that are more
administrative, will require investment in up/reskilling initiatives for public sector
employees (which will be covered in the education section in further detail).

Use Cases: Efficiency Gains

Voice-to-text This AI application involves software for the intelligent

transcription recognition and processing of language in several forms. This
and real-time might be understanding and responding to natural language,
universal the transposition of spoken to written language or even real-
translation: time universal translation and natural language processing
systems158. The Finnish Tax administration has used this for
the automation of subtitling videos and audio.
The AI system is based on understanding speech and
transforming it into text and is used by the Finnish Tax
Administration. It is used to provide subtitles on videos and is
part of a wider initiative within the administration to use
Speech-to-Text technologies in various use cases.

AI process The value proposition of this application stems from its ability
automation to automate standard tasks. Using rule-based assessment,
systems: workflow processing, schema-based suggestions, data mining,
and care-based reasoning, what were formally human-led
processes can now be carried out by automated systems.
These can greatly enhance operational efficiency and support
humans with mundane tasks such as data entry and
processing requests for administrative application forms.
The Danish government has used an Intelligent Control
Platform that provides an automated assessment of how a
selected company/business is more likely to commit fraud
compared with others159.

Improved decision-making and public services

Interrelated to but distinct from efficiency gains for the internal functioning of public
authorities, AI can also drastically improve the quality of public services through more
informed decision-making. Returning to AI Watch´s inventory of cases, 41% of cases
offered “responsive, efficient and cost-effective services” followed by “increased
quality of information and services” (22%) and “public (citizen)-centred services”
(17%)160.

157
AI Watch (2022) road to the adoption of Artificial Intelligence by the public sector
158
Bernd W. Wirtz, Jan C. Weyerer & Carolin Geyer (2019) Artificial Intelligence and the Public Sector—
Applications and Challenges, International Journal of Public Administration, 42:7, 596-
615, DOI: 10.1080/01900692.2018.1498103
159
AI Watch (2022) European landscape on the use of Artificial Intelligence by the Public Sector
160
AI Watch (2022) European landscape on the use of Artificial Intelligence by the Public Sector

62
 Adopt AI Study

Experts point to the utility of algorithmic processes in scenarios where “staff find it too
difficult or time-consuming to externalize implicit knowledge”. AI can fill this need and
make up for the “limited cognitive capacity and struggle to process large amounts of
information”. AI applications have the potential to improve each step of the policy-
making process (simplified below as Birkland´s stages heuristic framework of policy-
making161).
Figure 42: The policy stages heuristic

Algorithms have been deployed to help set the policy agenda and improve the
background knowledge on an issue of a government department or the wider public 162.
Kolkman shows that public authorities have deployed algorithmic models for policy
negotiations such as the Land Registry of the Netherlands which used algorithms to
project the estimated revenue of different types of legal services which were in turn
used to defend against a raise in tariffs. The Dutch government has also used the
SAFFIER II algorithmic model for long-term financial planning and council on its
budgets. Used in tandem, “algorithmic models and analyst judgement has been shown
to improve the accuracy of shortcuts”. Proponents point to the value of an impartial
algorithmic model which is an “alternative to biased, subjective and otherwise flawed
human decision-making” and their ability to bring “reliability and objectivity to
otherwise uncertain procedures”163.
However, some experts have stated that algorithmic impartiality is overstated and, at
times, even compounds existing human bias. A notorious case of this occurred in the
UK in 2020 with an algorithmic-led grading fiasco for university entry exams which
were unable to take place due to the COVID-19 pandemic. The algorithm used
historical data and rewarded students in smaller classes in certain subjects with more
favourable grades and disproportionately affected students from lower socio-economic
backgrounds164. The saga also revealed another issue with algorithmic prediction in
the public sector in opacity and complexity 165. The so-called black-box issue, the lack
of explainability of the AI algorithm puts it at odds with public demands for
transparency and makes “it potentially impossible to account for specific AI-driven
outcomes, and to correct actions with unintended consequences” 166.

Use Case: Improved Public Services

Recommendation Using an information filtering system and software-based

systems systems which mine personalized information to predict
preferences, algorithmic models can make predictions which
are purportedly free from human bias if executed correctly.
“One of the main benefits of AI is the ability to identify

Birkland (2015) An introduction to the policy process: Theories, concepts, and models of public policy
161

making. Routledge, London (2015)

162
Kolkman (2022) The usefulness of algorithmic models in policy making, Government Information
Quarterly, Volume 37, Issue 3
163
Kolkman (2020) The usefulness of algorithmic models in policy making, Government Information
Quarterly, Volume 37, Issue 3
164
Kolkman (2020) “F**K the algorithm?: What the world can learn from the UK´s A-level grading fiasco”:
“https://blogs.lse.ac.uk/impactofsocialsciences/2020/08/26/fk-the-algorithm-what-the-world-can-learn-
from-the-uks-a-level-grading-fiasco/
165
AI Watch (2022) road to the adoption of Artificial Intelligence by the public sector
166
AI Watch (2022) road to the adoption of Artificial Intelligence by the public sector

63
 Adopt AI Study

patterns and make more accurate predictions based on

available datasets”167.
The Estonian government has used the OTT system – a
decision support tool for consultants an AI system which
assists its consultants with providing insights predicting the
chances of an unemployed person getting a new job.

Open government
An area where AI can surpass orthodox digitalization of public administrations is the
effect it can have on the openness, transparency and participation of citizens. With
more open government, the separation between the public sector and the population it
serves will be less impermeable thanks to “trust through higher participation by
citizens in public sector activities and decision-making processes”168. Parts of the
population which have historically found the public sector to be inaccessible whether
that be for economic, linguistic, physical or technological reasons will be able to
interact with the authorities thanks to AI applications such as chatbots or virtual
assistants. However, this value driver is currently the least common in AI Watch´s
inventory with only 12% of cases, with “increased transparency of public operations”
offered by all of those cases.
Public authorities also run the risk of appearing even more dehumanized and like a
faceless bureaucracy if AI applications are handled badly. Critics point to a loss of
accountability of public servants by outsourcing their communication with the public to
an AI application creating a “phenomenon where citizens are faced with impotence
when confronted with “the computer says no” responses” 169.

Use Case: Open Government

VDAB: Chatbot Conversational bots can reduce waiting times and

for job administrative bureaucracy freeing up valuable resources that
seekers170 can be deployed elsewhere. Using NLP, intelligent agents,
affective computing and big data, chatbots provide automated
answers to common questions and processes.
In the Netherlands, the Flemish employment agency promised
to make a job available to graduates within three months of
leaving school. They used Cognitive applications to provide
people with a personalised experience through digital
channels.

Increasing cybersecurity of digital public services

The increasingly digital nature of the public sector has also increased the importance
of cybersecurity. Public authorities are especially vulnerable to these new types of
threats because of the large number of agencies and stakeholders involved in the

167
Misuraca Gianluca et al, September 2020, The use of AI in public services: results from a preliminary
mapping across the EU,
https://www.researchgate.net/publication/345015463_The_use_of_AI_in_public_services_results_from_a_p
reliminary_mapping_across_the_EU
168
AI Watch (2022) road to the adoption of Artificial Intelligence by the public sector
169
AI Watch (2022) road to the adoption of Artificial Intelligence by the public sector
170
Deloitte (2017) “VDAB: Helping tackle youth unemployment through Cognitive Analytics with a friendly
digital job coach”: https://www2.deloitte.com/be/en/pages/impact-report-2017/articles/clients/vdab.html

64
 Adopt AI Study

process of governing a society. Each agency, bureau or interaction with the public is
subject to cyber-risks and cyber-intrusion. 171
This vulnerability has been demonstrated by numerous attacks against public
institutions. For instance, a group of hackers took control of the health data of French
hospitals, asking for €1.2 million in ransom to release the data.172 In addition, studies
have shown that 80% of websites related to e-government around the world are
susceptible to structure query injection (SQL) and cross-site scripting (XSS) because
of an absence of an appropriate mechanism for authentication.173 174
These difficulties are very salient and are shared with the private sector. AI for
cybersecurity is already a mature market which receives high levels of investment and
interest from the private sphere. However, the uptake of AI solutions for cybersecurity
in e-government remains low in Europe. The low level of investment is reflected by a
weaker European market and European cybersecurity firms are smaller and typically
underperform compared with their American counterparts, notably in fundraising.
There is a need for targeted funding of start-ups in Europe, but also to boost the
demand for AI solutions for cybersecurity applications 175and more synergy between
the interest of the European private market in AI and cybersecurity and the objectives
of the public authorities in eGovernment. AI applications can help to secure public IT
infrastructure with 24/7 system monitoring, automated response to threats,
identification and localisation of threats and intrusion or identification of malware. 176
Policy-makers could thus encourage the development of this market by following their
interest and increasing funding of AI projects for cybersecurity in the Horizon
programme (around €2.5 billion were dedicated to this topic in Horizon 2020) 177,
procuring and implementing existing solutions from European providers and using
their own IT e-Government infrastructure for pilot testing of innovative solutions.

Use Case: Cybersecurity

Cognitive Technology giant IBM provides AI cybersecurity solutions

security which use cognitive technologies to analyse security
analytics: IBM information through natural language processing and machine
Watson learning.
The solution can retrace a cybercriminal's actions for deep
insights into the breach, reconstruct the data involved in a
security incident for a step-by-step view of the offence, and

171
Conklin Arthur, White Gregory, February (2006), e-government and cyber security: the role of cyber
security exercices, https://www.researchgate.net/publication/4216146_e-
Government_and_Cyber_Security_The_Role_of_Cyber_Security_Exercises
172
LeFigaro, 22/04/2022, Des hackers demandent une rançon de 1.2 million d’euros à un hôpital de la
Marne, https://www.lefigaro.fr/flash-eco/des-hackers-demandent-une-rancon-de-1-2-million-d-euros-a-un-
hopital-de-la-marne-20220422
173
Bala Kiran et al, 30/06/2021, Analysis of Cyber Security in e-government utilising blockchain
performance, https://assets.researchsquare.com/files/rs-938929/v1/bb9de2f7-7de9-4057-a7e1-
e02dba97937b.pdf?c=1633016778
174
Large agencies tend to be better equipped and drilled to face cyber-attacks. However, a large level of
unreadiness remains to be observed. The conduction of exercises on how to react to cyber-attacks has
demonstrated their efficiency in raising awareness and prompting modernisation of management structures
and procedures. Conklin Arthur, White Gregory, February 2006, e-government and cyber security: the role
of cyber security exercises, https://www.researchgate.net/publication/4216146_e-
Government_and_Cyber_Security_The_Role_of_Cyber_Security_Exercises
175
European Court of auditors, March 2019, Challenges to effective EU cybersecurity policy,
https://www.eca.europa.eu/Lists/ECADocuments/BRP_CYBERSECURITY/BRP_CYBERSECURITY_EN.pdf
176
Data Center Knowledge, 03/02/2022, Top 3 use cases for AI in cybersecurity,
https://www.datacenterknowledge.com/security/top-three-use-cases-ai-cybersecurity
177
European Court of auditors, March 2019, Challenges to effective EU cybersecurity policy,
https://www.eca.europa.eu/Lists/ECADocuments/BRP_CYBERSECURITY/BRP_CYBERSECURITY_EN.pdf

65
 Adopt AI Study

give IT security teams greater visibility even without special

skills or training178.

2.2.2 Digital value chain of e-government

A CEPS report on the European AI value chain defined it as the “organizational process
through which an individual AI system is developed and put into use” 179, something
which typically involves different types of stakeholders such as public authorities,
technology providers and academia.
As with other emerging technologies in the past, some experts argue that the public
sector needs take on the role of the “entrepreneurial state”. This idea, coined by
Mariana Mazzucato, highlights the role of the state in co-creating world-changing
technology such as the internet and GPS 180. Rather than simply relying on technology
from the private sector, Mikhaylov et al. argue that government needs to “play a
significant role in innovation”181. They argue that cross-sectoral collaborations with
businesses and universities are integral to ensure that the power of AI can become a
boon for the economy and society. They even point to recent policy paradigms such as
the UK´s AI sector deal which has “institutionalized the partnership between
government, industry and academia”. Many experts point to the competence gap 182
between the public and private sectors whereby the latter has the specialization and
funding to offer something such as an AI application.
More specifically, an AI Watch183 report points to the argument that collaboration
between public and private entities is desirable and likely take the form of either
partnerships or procurement. It highlights 3 types of interaction in the implementation
of AI solutions.
The implementation of AI solutions implies an interaction among different actors. The
public sector is mainly involved in three types of relations:
• Government-to-Government (G2G). Processes between and within public
organisations, like services and information transactions between the central-
state government, state-local governments, and between department-level and
attached agencies and bureaus.
• Government-to-Citizen (G2C). Services and information transactions by the
government interacting with private users (citizens).
• Government-to-Business (G2B). Services and information transactions by the
government to private organisations and other economic activities.
The report observed that 45% of AI solutions are designed to support direct
interaction with the users and 43% support G2G relations of the governmental
backend and public organizations. At 12%, G2B solutions were comfortably the least
numerous.
Nevertheless, as a result of the complexity of AI applications, collaboration can take
several forms as highlighted in the recent AI Act 184. Understanding the complexities

178
IBM (2023) “IBM Security Qradar SIEM): https://www.ibm.com/products/qradar-siem/addons
179
Engler & Renda (2022) Reconciling the AI Value Chain with the EU´s Artificial Intelligence Act
180
Mazzucato & Li (2020) The entrepreneurial State and public options: Socializing risks and rewards:
https://www.ucl.ac.uk/bartlett/public-purpose/sites/public-
purpose/files/final_the_entrepreneurial_state_and_public_options.pdf
181
Mikhaylov, Esteve and Campion (2018) Artificial intelligence for the public sector: opportunities and
challenges of cross-sector collaboration
182
Mikhaylov, Esteve and Campion (2018) Artificial intelligence for the public sector: opportunities and
challenges of cross-sector collaboration
183
AI Watch (2022) European landscape on the use of AI by the public sector.
184
Engler and Render (2022): Reconciling the AI Value Chain with the EU´s Artificial Intelligence Act

66
 Adopt AI Study

and which type of interaction will suit different scenarios in the public sector will soon
become a prerequisite for public authorities looking to adopt AI applications. This is an
area where information exchange and sharing of best practices is invaluable between
public actors at the member state and European level.

Type 1: Internal AI development and deployment

Type 2: One entity develops an AI system for another entity (AI system contracting)
Type 3: One entity writes the code and trains the system then sells access through a
branded application or API (restricted AI system access)
Type 4: A vendor writes code for an AI system but does not pre-train it or provide
training data to purchasers (software with AI code)
Type 5: Vendors of learning AI systems
Type 6: initial development by one entity and fine-tuning by another (AI system-fine
tuning)
Type 7: One entity integrates different AI systems into a new one (AI model
integration)

Public procurement process of AI

With the complexity of AI solutions as well as the limited resources and generalist
disposition of public administrations, procurement is becoming increasingly
commonplace – demonstrated by the revised procurement processes of 13 member
states in their National AI Strategies. As AI becomes more pervasive in the lives of
ordinary citizens, member states need to be able to supply a high quality of service in
line with what is provided by the private sector. As with procurement writ-large,
governments rely on the expertise and previously developed models of technology
providers and may lack the necessary skills to fully understand or trace algorithmic
causality185.
In the words of a World Economic Forum guidelines on AI procurement,186 “technology
providers understand these challenges and look to governments to create clarity and
predictability about how to manage them, starting in the procurement process”. While
governments can build on institutional knowledge and adapt previous procurement
processes, there are key considerations for the process of the procurement of AI.
While there are fundamental differences between partners in the value chain,
cooperation and “common sense frameworks can help governments overcome
reluctance to procure”187.
Successful procurement of AI solutions is bookended by crucial steps in the overall
adoption of the technology. Before the procurement, authorities should carry out a
thorough analysis of the precise value proposition of the AI solution and “should
consider in-house knowledge on AI for the– partial or complete – internal development
of AI, for the direction and adjustment of the system developed by external suppliers,
and/or for ensuring proper management of procurement activities”188. Similarly, the
procurement of the technology needs to be followed and backed up by successful
implementation following a piloting phase with a long-term strategic vision and in-built
flexibility for the continued engagement of the technology provider. The Adopt AI
programme will help member states with the intricate process of AI procurement and
below is a synthesis of two pieces of advice from the World Economic Forum´s AI

185
WEF (2020): AI Procurement in a Box: AI Government Procurement Guidelines
186
Ibid
187
Ibid
188
WEF (2020): AI Procurement in a Box: AI Government Procurement Guidelines

67
 Adopt AI Study

procurement in a box toolkit – the SDLC stages and some corresponding

recommendations.

Figure 43: Procurement process of AI

2.2.3 Main drivers and barriers

The table below presents a PESTEL analysis of the uptake of AI technologies in public
procurement in the e-government sector.

68
 Adopt AI Study

Table 3 PESTEL analysis of the uptake of AI technologies in public procurement in the e-government sector

Criteria Drivers Barriers

 At a time of low trust in political institutions,  The growing complexity of AI solutions makes it
there is a need to increase transparency, difficult to predict their concrete impact on
anticorruption and accountability. redefining governance. There is a lack of
consensus on how to handle the challenges
 Public authorities want better information to related to AI in the public sector as well as a
make better decisions and AI-powered tools can
scarcity of research on AI governance, policy
increase the capacity of the government to
and regulatory issues.
meet the needs of the population and serve
them by improving the quality of decision-  AI-powered eGovernment requires a re-design
making. This can mean an increase in of existing vertical and hierarchical-oriented
Political
anticipatory governance and policy through administrative structures. For a fruitful
more accurate predictions. Digital modelling of implementation of AI solutions in public
policy options can help make more effective administration, an organisational transformation
governmental decisions, such as in the context is required.
of urban governance and planning.
 An appropriate strategic approach to the
 There is high potential in the cybersecurity implementation of AI needs to be secure and
market which needs to see more funding for AI ethical which requires constant updating at the
projects for cybersecurity. same pace as the development of the
technology.

 The promise of decreased administrative  Adoption of AI involves a high initial set-up

burden for both administrators and society can cost, and public resource allocation, especially
lead to important time and cost savings. Most in planning, programming and budgeting the
of the AI initiatives undertaken by public resources for the public sector.
Economic
authorities are done to achieve efficiency gains.
AI applications free up precious cognitive
resources of public workers, which can then be
allocated to tasks of higher added value.

69
 Adopt AI Study

 Citizens want more service efficiency and  Caution is required in the implementation of AI
service quality which incentivises the usage of solutions as AI can either increase or decrease
AI by public authorities. trust in public authorities from their citizens.
There is a need to investigate not only the
 AI applications offer new avenues of impacts of AI on levels of trust but also to
participation and to better involve citizens in
understand mechanisms that build the
planning decisions. AI systems could synthesise
Social trustworthiness of AI, which is closely related to
millions of responses and identify the most
providing transparency, explainability and
important points of view and their weighting.
reliability.
 Attracting a sufficient number of experts with
the required communications skills by public
authorities can communicate the intricacies of
AI technologies.

 There is often high scalability and replicability in  Lack of interoperability in AI systems can be a
AI solutions. A successful solution implemented shortcoming especially when moving beyond
in a specific country (at the national level) can pilot phases. Interoperability can ensure the
be easily adapted and applied in other countries compatibility and integration of different data
or even at regional or European levels. and, therefore, guarantee a smooth
implementation and adoption of AI.
 Many AI techniques for improved delivery in e-
government are still relatively immature
Technological (except well-known solutions such as customer
chatbots). In this aspect, relatively new
solutions (for example neural networks for deep
learning for handwriting recognition and fraud
detection) need further R&D efforts. The
development of appropriate technical standards
and encouraging experimentation are crucial for
the increase of AI deployment.
 Many local government bodies or smaller
communities do not have the means to be part

70
 Adopt AI Study

of a vivid data ecosystem, which is a minimal

requirement to develop AI.
 The shortage of in-house knowledge on AI for
the internal development, direction and
adjustment of the systems developed and to
ensure proper management of procurement
activities. Algorithms can be biased and this
might lead to erroneous suggestions that
require correcting when they occur.

 AI solutions can be used to optimize the use of  Data centres are critical for storing the large
resources to deliver general public services. amounts of data needed to power AI systems,
Subsequently, the digitalisation of but demand a huge amount of energy. In
Environmental administrative operations can result in the addition, training advanced artificial intelligence
minimisation of materials used and waste systems, including deep learning models, can
reduction. require high-powered GPUs to run for days at a
time.

 Public procurement has the opportunity to take  There are still important concerns that require
a leading role in establishing rules and technical clarification and regulation related to biases,
standards for AI solutions that could be then transparency and attribution of responsibilities
taken and implemented by the private sector. and accountability.
Legal
 A common requirement is the simplification of
regulatory frameworks enabling and facilitating
data sharing (for AI) which leads to improved
data accessibility.
Source: Authors’ elaboration based on the challenges and the value chain analysis.

71
 Adopt AI Study

Figure 44 Strengths, Weaknesses, Opportunities and Threats of the uptake and public procurement of AI-technologies in
the e-government sector

S
Political will - There is general acceptance by most member states that administrative
issues under the responsibility of public authorities shoykd be automatized with AI
systems. The public sector presents adequate readiness and willingness to operationalize
existing technologies, develop alternatives and acquire the expertise. Additionally, the
involvement of citizens in planning decisions that AI applications can offer, can increase
this social acceptance as inhabitants feel part of the process. However, it is crucial to
adopt trustworthy AI solutions
Large amounts of data of the public sector - Governments and public organizations
have large amounts of available data and the full potential of these datasets has yet to be
unlocked. Across countries, initiatives to share data and open data exercises between
public entities and other collaboration partners are emerging. These initiatives will ensure
the ability to combine data sources and increase the validity of data. It should be

W
considered that available data volume of smaller countries is smaller.

Lack of skilled workforce – There is a lack of expertise and digitally skilled employees
in public authorities and AI solution providers with knowledge in governmental and
administrative procedures in order to provide them efficient tools and technologies
capable of answering to real needs.
Unsatisfactory data sharing across organisations – In order to achieve high volumes
of data and the requirements to combine different types of data from key thematic issues
is required. The absence of data standards, lack of data interoperability due to technical
standards and hardware and software variations, among others, make this data sharing
more complex.
Data quality and management – The outcomes of the application of AI solutions
strongly depend on the quality of the input data as well as its management. In this

O
aspect, the large volume of available data needs to be treated and organized in order to
provide the required results.

Increase on the investment - Investment in R&D for AI technologies and the launch of
pilot projects can incentivise the further development of more mature AI technologies and
the scale up of the existing ones, as well as the enhancement of the cybersecurity
market, which is a current need for the implementation of AI tools in the digitalization
pathway of services provided by public authorities.
Bring services closer to people – Adoption of AI technologies in general public services
can make those services available, inexpensive and easy to use for citizens. Additionally,
AI presents the opportunity to improve interaction between citizens and governments
through the provision of better and more inclusive services and the enhancement of
citizen participation in the activities of the public sector.
Replicability- Adoption of AI technologies in general public services presents high
scalability and replicability. A successful solution implemented in a specific country (at
national level) can be easily adapted and applied in other countries or even at regional or
European levels. Common needs at different levels can be addressed through the same
tool by adapting it to the new environment. AI technology providers can reproduce the
solution in any other entity with similar nature and necessities.
Accelerating the uptake of legal and ethical frameworks- Regulations and laws
aren’t keeping up with the rapid development of the technology. The adoption of AI
technologies by the public sector has the potential to speed up the simplification of

T
regulatory frameworks, as well as clarifying several aspects in terms of biases,
transparency and accountability.

Privacy concerns - Collection, storage, disclosure, processing, and dissemination of

personal data is required to power AI based solutions in general public services.
Algorithmic bias – As the usage of AI-based algorithms by public organisations
increases, the possibility of amplifying discriminatory biases increases. As algorithms are
implemented in heterogeneous socio-technical systems, it is important to consider the
potential propagation of existing social biases across gender, race, sexuality and ethnicity.

Source: Authors’ elaboration based on the challenges, value chain and PESTEL analysis.

72
 Adopt AI Study

2.2.4 Case Study: Mercè - citizen science for better urban life (Spain)
Key actors
Project User: 300.000 Km/s
Contributors: Ministry of Science and Innovation of Spain and Fundación Española
para la Ciencia y la Tecnología
Supported by: Barcelona’s town hall, COAC, COTEC, UrbanNext and S+T+ARTS
Goals
The Mercè project forwards a new line of work that applies new machine-learning
techniques to the disciplines of urban planning, geography, sociology, economics,
science and urban health to build objective knowledge and open data about urban
environments. Mercè is a citizen science experiment that aims to involve citizens in
the training of an algorithm that allows the design of more liveable cities. Through this
experiment, it is possible to translate the individual perceptions of many citizens about
the habitability of cities into objective knowledge, reusable in urban planning. The
relevance and innovation of the project are based on applying information and
communication technologies to different transversal fields of knowledge. The project
develops a workflow based on public/open data and machine learning techniques to
promote algorithmic transparency in data science and citizen participation with great
social impact.
Description
Mercè is an artificial intelligence algorithm in which citizens can show their preferences
and generate a knowledge bank that will make it possible to identify the patterns that
exist between different opinions, making the subjective measurable and defining
metrics on the qualitative to put the liability on urban agendas as something tangible.
In this context, city halls have multiple challenges related to urban planning that have
to be solved to make cities environments in which human life can be developed
optimally: mobility, land management, the provision of the correct proportions of
housing and its relationship with services and facilities, the location of economic
activities, the balance with the environment and the surroundings, among others.
Therefore, liveability appears as a concept with a very high degree of subjectivity,
which varies according to the cultural context and the vital condition of each person.
In this line, the main challenge for the development of Mercè relied on how to
measure this phenomenon.
To overcome this challenge, Mercè project experimented with new channels of digital
participation to convert the subjective perception of many citizens into objective
information, making the subjective measurable and defining metrics on liveability and
making it something quantitative for urban agendas.
For the application of these technologies, Mercè is powered with two information
sources:
 Individual citizens create data either from their mobile devices, through
social networks, online shopping, video calls and digital photographs.
 Administrations and private companies generate data on their
management of urban services, administrative procedures, citizens' use of their
services or internal operations, and on the use of their services by citizens.
With all this information and the application of machine learning algorithms, it was
possible to recognise common patterns from input data. The Mercè project applied this
set of technologies to urban planning through a citizen science experiment that
allowed participants to express their preferences and, thanks to an artificial
intelligence algorithm, to classify these opinions and associate them with the
characteristics of the urban environment according to the habitability they generate.

73
 Adopt AI Study

The next challenge was related to the training of the algorithm. The data model
developed in the first part of the experiment consisted of 15,611 streets with 212
variables associated with each of them. The next step was to be able to generate a
simplification of the model to maximise the participation result: if only 20% of the
streets were evaluated, the number of times a citizen can vote for it increases.
Therefore, the streets were clustered based on their characteristics. Another
fundamental part of the training process is the design of participation: how can
citizens interact in a scientific process of data classification? In this sense, they opted
for an online voting app189 whose design and operation design and operation must
guarantee the participation of different groups: from an individual citizen of any age,
sex and origin to a specialised public related to the subject of the experiment.
Another key pillar of the strategy of citizen involvement strategy has been to build
alliances for the dissemination190 of the project with several relevant entities in the
world of citizen science and architecture.
Results
The final result of the Mercè experiment is a map of the habitability of the city191
that shows us, according to five typologies, which are the most and least habitable
streets. Additionally, the application of the machine learning algorithm enabled the
identification of the characteristics of the most habitable streets, as well as the
characteristics of the least habitable streets generating a complete report with the
results192. In this line, streets were classified and characterised in five typologies: very
unhabitable streets, not very habitable streets, streets with medium habitability,
habitable streets and highly habitable streets.
According to these five typologies, the Mercè tool generates a habitability map of the
city of Barcelona, in which the streets are coloured depending on their habitability.
Thanks to this characterisation of the streets, the Mercè tool enables the identification
of demographic patterns. If the most habitable streets are mostly residential, have
little traffic and have a quality public space thanks to trees and lush vegetation, it is
interesting to ask whether these urban characteristics are equally shared by men and
women, young and old, locals and foreigners. The tool allows the identification of
priorities and differences between gender, age and origin.
The results obtained from the Mercè project enabled the validation of the tool itself
and helped the city hall of Barcelona to understand the factors that make a street
more or less habitable. In this line, it helped in the identification of the less habitable
streets, which can be classified as action areas, and the most habitable streets
considered as real examples of what inhabitants want.
Data-informed urban planning facilitates greater clarity in the planning process
because what motivates it and how it is formulated is expressed through parameters
and conditions that have been validated, agreed upon and traceable. This means that
urban planning documents must be constructed based on concrete objectives
and indicators, which are measurable and which make it possible to determine their
success or their need for reformulation. And this evaluation can also be carried out by
citizens.

189
Mercè App, link
190
The Mercè project counted with the collaboration of the College of Architects of Catalonia, the Urban
Next platform, a global content network focused on architecture and urbanism and the Ibercivis
Foundation, a benchmark organisation in Spain that develops, promotes, makes visible and researches
citizen science.
191
Habitability map of Barcelona generated with Mercè tool, link
192
300.000 Km/s (in collaboration with Ministry of Science and Innovation of Spain and Fundación Española
para a Ciencia y la Tecnología), December 2020, Mercè- A citizen science experiment: citizens training
algorithms to make more liveable urban environments, link

74
 Adopt AI Study

Successes and Key lessons:

 The participation and involvement of all citizens are key to addressing
the radical transformation of lifestyles that we will experience in the next
decade and the challenges it entails. This participation must be accompanied by
technology, making planning processes incorporate new shared standards
between cities, but above all, it must be guided by the liveability of cities as the
ultimate goal and driver of change.
 The aspect of giving voice to citizens ensures that the actions carried out by the
public administration are in line with the needs and requirements of inhabitants.
Additionally, the involvement of citizens allows strategic plans and agendas to
be focused on real necessities and design pathways towards the actual
challenges, which will lead to the implementation of efficient solutions and
measures. In the case of the Mercé tool, it transforms urban planning and
urban agendas into participatory, transparent, evaluable and fair strategies.
 The importance of real indicators for decision making. Mercé provides
cities with a tool and a methodology based on collective and participatory
artificial intelligence and open data models to build bottom-up citizen indicators
that allow both the design of a more habitable city and the offer of strategies to
evaluate projects, actions and urban policies already deployed.
 In this line, Mercè enables city planning that moves from using information for
diagnosis to using knowledge for forecasting. The different indicators used by
Mercé allow urban planning documents to be constructed based on concrete
objectives and indicators, which are measurable and which make it possible to
determine their success or their need for reformulation. And this evaluation can
also be carried out by citizens.
 Collaboration is crucial for the creation and later adoption of AI
technologies by public entities. In the case of the Mercè project, the tool was
fully built and created by 300.000 Km/s a private architecture studio in
Catalonia. However, as it was considering a strategic solution that could help in
the development of urban agendas and strategies in different towns, cities and
even regions, the project was supported by the Ministry of Science and
Innovation of Spain and Fundación Española para la Ciencia y la Tecnología.
Additionally, as the creation of the tool has taken Barcelona as the pilot city,
the town hall of Barcelona has also been implied in the project.
 Furthermore, the contribution of private entities has been essential for the
successful conclusion of the project, which ended up with a fully developed AI
solution, Mercè. These entities supported the project from a private and
specialised point of view in terms of ICT and sectorial areas that the project
addresses.
 Therefore, it can be said that each entity, regardless of their nature, had a key
and specific role in the project and the effective contribution among them
guaranteed the creation of an efficient and highly applicable AI-powered tool.
2.2.5 Economic Analysis
Proponents of AI in the delivery of public services point to its ability to increase
efficiency and accessibility. The variables chosen for use in the linear regression
models test these supposed benefits of AI in general public services with more
efficient government. In line with Misuraca and Van Noordt´s framework in the
introduction, this asks whether there is any evidence that supports the expected
outcome that countries using more AI will have more digitalized public services.

75
 Adopt AI Study

The three models use the DESI´s eGovernment indictor group and measure citizens´
use of eGovernment services (in the last 12 months) 193. To test the online government
services accessibility, model 1 uses the “citizens´ use of eGovernment services” DESI
indicator which measures: the individuals who have used the internet, in the last 12
months, for interaction with public authorities 194. To test whether countries with more
AI investment in eGovernment have better interaction with both citizens and
businesses, models 2 and 3 use the DESI´s “digital public services for citizens” and
digital public services for businesses” indicators. These measure the extent to which a
service or information concerning service for citizens and businesses is provided
online, and via a portal 195.
Figure 45 Member states typology

N.B Cyprus is categorized as a Facilitator and Malta as a planner

193
Eurostat, Table isoc_r_gov_i: Individuals who used the internet for interaction with public authorities
194
Eurostat, Table isoc_r_gov_i: Individuals who used the internet for interaction with public authorities

76
 Adopt AI Study

Table 4 Regression table – E-government

Use Citizens Businesses

User 71.50 77.76 85.02

Facilitator -14.25 -21.62 0.78

Planner -23.71 -9.00 -12.89

R² 0.28 0.16 0.18

Observations 27 27 27

*Intercept is the figure for the User category but also the baseline to interpret the rest of the
coefficients in the model

 Use: AI User countries are the category with the highest use with 71.5% of
citizens using the internet for public services. Facilitators have 14% lower use
than users. Planners, countries with no investment in AI in general public
services, have almost 24% lower use than those that have already developed
AI solutions.
 Citizens: Users of AI also have the best rate of information for citizens online
for “citizen life events” such as moving or small claims. With a score of nearly
78, AI Users are 9 points better than planner countries.
 Businesses: This trend is similar for business life events where Users and
Facilitators, which are roughly the same, are nearly 13 points higher than
those countries without AI in their public services.

2.3 Mobility
As streets, train carriages and airports emptied seemingly overnight, the advent of the
COVID-19 pandemic served as a reminder of the everyday reliance on mobility but
also as a chance to take stock of how this cornerstone of society might be improved.
No longer limited to the realm of science fiction, the era of driverless cars, drone taxis
and other uses of AI in the mobility space is on the horizon with the global market for
AI in the sector expected to reach $3.5 billion by 2023 196.
This growth in AI-enabled mobility, defined by the ELTIS glossary as the “potential for
movement and the ability to get from one place to another using one or more modes
of transport to meet daily needs”197, can be attributed to AI´s transformative potential
at every level of the mobility ecosystem and multiple modes of transportation. On a

196
Joshi (2019): How AI can transform the transportation industry:
https://www.forbes.com/sites/cognitiveworld/2019/07/26/how-ai-can-transform-the-transportation-
industry/?sh=16eb45374964
197
ELTIS (2002): “Mobility”: www.eltis.org/glossary/mobility

77
 Adopt AI Study

broader level, AI can help plan, design and manage mobility networks 198, processing
past and real-time data to administer flows of traffic and people on road networks and
public transport. AI is already being utilized for detection and prediction in areas such
as aviation, with estimated more precise arrival times at Singapore´s Changi Airport 199
and road maintenance on Germany´s Deutsche Bahn200 . In addition to infrastructure
and systems, AI in mobility is also developing quickly within the modes of transport
themselves, especially in the area of autonomous vehicles (AVs). In cars, lorries and
buses, AVs deploy deep learning in a range of capacities including localization dynamic
scene understanding, path planning, control and user interaction 201. Even at the level
of the individual user, mobility-related applications for smartphones and an increased
range of micro-mobility options like e-bikes and electric scooters are already
harnessing the power of AI and seem like a harbinger for an age of integrated and
cleaner mobility, particularly in urban areas.
Despite the clear growth of the sector that has seen $51.5 billion invested in
automotive AI alone since 2010202, uptake is still languid in the EU with only 37.5% of
respondents to an Ipsos survey203 (for European SMEs in the transport sector) stating
there were using at least 1 AI technology. Despite 55% planning to incorporate AI into
their enterprise in some capacity over the next 2 years, issues that are stifling uptake
in other sectors can also be found in mobility such as budget constraints 204.
To this effect, public leadership has a key role to play, especially given the current
state of the sector which is highly siloed in the private sector with 76% of interactions
occurring within the B2B space205. Up to this point, much of governmental bandwidth
has been taken up by the prescient task of regulation, especially crucial given the
quotidian nature of the mobility sector and the risks concerning personal safety.
Indeed, at the European level, the OECD described the EU´s approach to AI ethics and
regulation as “comprehensive” but also cited the need for speed and malleability,
especially transient with the high pace of innovation in mobility. Lagging behind other
geographies such as the US and China, the EU must go beyond regulation and play a
more active role through procurement. The improvement of data infrastructure, for
example, is critical given that around half of those surveyed in an EIT Mobility report
highlighted poor data availability and quality as a bottleneck to the sector.
In the overall landscape of AI technologies, according to the European Investment
Bank, the EU needs to fill a €10 billion gap206 to compete with the USA and China who
together currently account for around 80% of the total investment in AI and

198
EIT Urban Mobility (2021) Urban Mobility Next #3 AI Mobility Landscape in the EU:
https://www.eiturbanmobility.eu/wp-content/uploads/2021/09/EITUM-UrbanMobilityNext3_Final.pdf
199
Lee and Miller (2019) AI gets real at Singapore´s Changi Airport
https://ink.library.smu.edu.sg/cgi/viewcontent.cgi?article=1115&context=ami
200
DB (2020) Predictive maintenance using artificial intelligence: https://www.dbsystel.de/dbsystel-
en/about-us/news/Predictive-maintenance-using-artificial-intelligence-5569190
201
Fernandez Llorca & Gomez Gutierrez (2022) Artificial Intelligence in Autonomous Vehicles: towards
trustworthy system: https://publications.jrc.ec.europa.eu/repository/handle/JRC128170
202
Cornet et al (2017) The road to artificial intelligence in mobility – smart moves required:
https://www.mckinsey.com/industries/automotive-and-assembly/our-insights/the-road-to-artificial-
intelligence-in-mobility-smart-moves-required
203 European Commission (2020) European enterprise survey on the use of technologies based on artificial
intelligence: https://digital-strategy.ec.europa.eu/en/library/european-enterprise-survey-use-technologies-
based-artificial-intelligence
204
EIT Urban Mobility (2021) Urban Mobility Next #3 AI Mobility Landscape in the EU:
https://www.eiturbanmobility.eu/wp-content/uploads/2021/09/EITUM-UrbanMobilityNext3_Final.pdf
205
EIT Urban Mobility (2021) Urban Mobility Next #3 AI Mobility Landscape in the EU:
https://www.eiturbanmobility.eu/wp-content/uploads/2021/09/EITUM-UrbanMobilityNext3_Final.pdf pdf
206
EIT Urban Mobility (2021) Urban Mobility Next #3 AI Mobility Landscape in the EU:
https://www.eiturbanmobility.eu/wp-content/uploads/2021/09/EITUM-UrbanMobilityNext3_Final.pdf

78
 Adopt AI Study

blockchain207. Globally, the mobility sector is estimated to be worth €6500 billion

annually208 and, with its enormous potential and environmental implications, should
form a core pillar of the EU strategy on AI usage and value.
The policy landscape: Mobility and AI
Perhaps as expected, the OECD´s AI observatory´s policy catalogue shows that the
principal regulatory focus has been on automated vehicles. Indeed, the only
intervention of certain countries, such as Spain and Austria, has been limited to
regulations that have passed legislation on automated driving and a code of practice
(AT) and authorizations to test automated driving systems and parking systems (ES).
Figure 46 Mobility and AI – breakdown by policy mode

Source: Authors’ elaboration.

With three of the top ten automotive manufacturers worldwide in terms of revenue 209,
Germany´s approach to AI and mobility has been a leading example of public
intervention in Europe. In 2018, rather than merely a section of a broader AI strategy,
the Federal Ministry for Transport and Digital Infrastructure published a focused Action
Plan for Digitalization and AI in Mobility. The subsequent policy outputs include a
series of AI centres for mobility, a dedicated data initiative for mobility and the
mFund, a €200 million programme which funds R&D projects in Mobility 4.0.
Likewise, France, Belgium and Finland have also launched procurement programmes
mostly in the form of R&D or business grants through their respective innovation
agencies such as Innoviris210 (Brussels region) and the “challenges IA” initiative in
France. These have focused more on providing SMEs and start-ups with a public
investment such as the Brussels cooperative urbike211, a project seeking to prove the
viability of bicycles as a solution for the infamous last-mile problem in logistics.

207
EIT Urban Mobility (2021) Urban Mobility Next #3 AI Mobility Landscape in the EU:
https://www.eiturbanmobility.eu/wp-content/uploads/2021/09/EITUM-UrbanMobilityNext3_Final.pdf.pdf
208
Business Finland (2018) “Smart Mobility Program Starts”: https://www.businessfinland.fi/en/whats-
new/news/2018/smart-mobility-program-starts
209
Companies market cap (2023) “Top publicly traded automakers by revenue”
https://companiesmarketcap.com/automakers/largest-automakers-by-revenue/
210
Innoviris.brussels (2023): https://innoviris.brussels/
211
Urbike (2022): https://urbike.be/

79
 Adopt AI Study

Meanwhile, countries like Slovakia and Hungary are focusing more on early-stage R&D
with the Smart Mobility Lab 212and the National Laboratory for Autonomous Vehicles 213
which link academia and industry to design and develop innovative mobility solutions.
In addition to the more recent push in national-level procurement, the concept of
living labs in mobility is well established throughout the EU. Recognized in 2006 by the
Commission and its Sustainable and Smart Mobility Strategy for Europe´s Urban
Mobility. The concept seeks to encourage the development and testing of innovative
mobility solutions in the real-world environment in which they would be used.
Typically, these are 2–4 year projects that bring together academia/RTOs, end-users
and public authorities. Urban mobility initiatives, including living labs, test beds and
lab-like initiatives, are most numerous in Spain (25), the Netherlands (22) and
Germany (18) inevitably in large urban centres such as Madrid (5), Amsterdam (4)
and Hamburg (3). Taking the City Flows Milan living lab in the city´s Central Station
as an example, a monitoring decision support system uses historical and real-time
data to measure crowd movements from multiple modes of transport entering the
station including two subways lines, bus and tram stop, bike, scooter, taxis and car-
sharing traffic.214
Figure 47 Mobility and AI – breakdown by Member State

Mobility & AI: Breakdown by Member State

7

0
Austria

Finland
France

Romania
Cyprus

Italy
Latvia
Greece

Malta
Netherlands

Spain
Ireland

Luxembourg

Slovakia
Estonia

Lithuania

Sweden
Denmark

Hungary
Bulgaria
Croatia

Poland

Slovenia
Belgium

Germany
Czechia

Portugal

Regulation Facilitator Funder User

Source: Authors’ elaboration.

2.3.1 Challenges and AI solutions/applications in the mobility sector

Several key trends have profoundly transformed and impacted transportation &
mobility (T&M). Broadly speaking, T&M companies are faced with the need to cut costs
while maintaining high-level quality for their transportation offer. For example,
protectionism’s impact on T&L led global airlines to slash profit forecasts out of fear of
higher protectionism while M&A intensity dropped by 22.6% in the sector. 215 216

212
Smart Mobility Lab (2023): https://smartmobilitylab.sk/
213
Autonomous Systems National Laboratory (2022) https://autonom.nemzetilabor.hu/
214
City Flows Project (2022): https://cityflows-project.eu/milan/
215
Win-win solution, 2019, Global airlines slash profit forecast 21% on protectionism fears, http://win-
winsolution.com/feed-items/global-airlines-slash-profit-forecast-21-on-protectionism-fears/
216
PwC china, 2017, Key findings from the transportation & logistics industry,
https://www.pwccn.com/en/research-and-insights/ceo20/transport-and-logistics.html

80
 Adopt AI Study

Similarly, the COVID crisis caused many transportation companies to accumulate

217

debt to maintain activities in a context of extremely limited market demand.

In addition to these difficulties, the legal and societal requirements to lower energy
consumption and CO2 emissions are particularly pertinent for the transportation
sector. Considerations such as reducing congestion of transport infrastructure and
networks are extremely important for road and rail transportation. Even though AI
applications represent a partial solution to address these challenges, the
transportation industry remains relatively underdeveloped in terms of its AI
progression. For example, a PwC 2016 global survey showed that only 9% of T&L
companies have a dedicated data & analytics department.218 Furthermore, 23% of
surveyed companies mentioned they had no significant data analytics capabilities. 219
Transportation players are focused on addressing their two most urgent digital
challenges: improving data quality and standardising data formats. Data remains
fragmented and stored in different data silos that correspond to the different transport
modes that compose the industry. The harmonising data format is made even more
difficult because of the poor quality of the available data (inconsistent data collection,
heavy use of traditional paperwork, etc). Consequently, cross-modal efforts to address
this challenge are key for the deployment of AI solutions. 220
Figure 48 Summary of challenges and AI solutions for mobility

Public sector Reduce costs Reduce Reduce Reduce

challenge environmental congestion of accident rates
footprint infrastructure
and networks

AI value driver Efficiency gains Improved Congestion Road safety

through cost- environmental control and and accident
cutting performances prediction prediction

AI applications Smart- Vehicle control Smart traffic- Intelligent

maintenance (autonomous management systems for
driving, emission systems visual
control systems) monitoring,
vehicular
accident
modelling

AI techniques Computer vision Artificial neural Genetic Expert

networks, fuzzy algorithms systems
logic

217
PwC, 2018, M&A in the transport & logistics industry,
https://www.pwc.com/sg/en/publications/assets/transport-logistics-ma-2018.pdf
218
PwC, 2016 (B), 2016 Global Industry 4.0 Survey, transportation and logistics key findings,
https://www.pwc.se/sv/pdf-reports/industry-4-0-building-the-digital-enterprise.pdf
219
PwC, 2016 (B), 2016 Global Industry 4.0 Survey, transportation and logistics key findings,
https://www.pwc.se/sv/pdf-reports/industry-4-0-building-the-digital-enterprise.pdf
220
YourEdi-Transmetrik, 2018, Big Data and big roadblocks: how the logistics industry can overcome its big
data challenges, https://www.youredi.com/blog/logistics-industry-can-overcome-big-data-challenges and
International Journal of logistics systems and management, 2019, Challenges and opportunities for logistics
standardization in Asia-Pacific countries: a descriptive case-study, https://decisionsciences.org/wp-
content/uploads/2019/06/p584364.pdf

81
 Adopt AI Study

Efficiency gains through cutting costs

AI offers solutions and opportunities for efficient cost-cutting strategies in most
transport modes. Uses include such as last-mile delivery robotics,221 fleet management
solutions, AI solutions for less-than-truckload optimisation222, automation of client-
facing interfaces, smart maintenance, etc. 223 Smart infrastructure.
In this sector of the economy, public and private mobility providers are confronted
with the daunting challenge of having to cut their costs to remain competitive while
providing the same level of service at the same or higher quality level. Despite this,
the level of uptake of AI solutions that could help achieve this complex goal can
greatly vary due to factors such as a lack of business incentives for greater uptake of
AI solutions or a lack of IT infrastructure to enable faster deployment.
Development of trans-modal solutions and transfer (e.g., fleet maintenance and
autonomous vehicles solutions from the automotive to railway sector) could generate
economies of scale and increase the market potential for the development of new
solutions. Public procurement will have to play the leading role in the railway industry
for cost-cutting strategy. This is because public authorities are the main actors in both
operating the infrastructure and rolling-stock in railway ecosystems. Greater
interoperability and transfer of technologies from road transport could greatly benefit
the rail sector.
In air, road and maritime transport, public procurement of advanced AI systems for
smart infrastructure is expected to generate important cost-cuts. This is the case of
automated ports for berth activities and optimisation of routes. Smart infrastructure
would also accelerate the deployment of new innovate solutions from private players
and generate new market opportunities (e.g., start-ups to provide machine learning
and automated analytics for port and road fleet management).

Use Case: Efficiency gains

Deutsche Bahn This rail giant uses computer vision from a network of cameras
smart and sensors to inspect the roofs of trains, relieving employees
maintenance and reducing the time “from several hours to a few
minutes”.224
Furthermore, DB is currently testing AI processes, for
example, to predict material requirements in the factories or
the right time to maintain or replace wheelsets which further
increases vehicle availability.

Improved environmental performances

The challenges raised by key environmental questions are central to the future of
mobility. Transport accounts for 27% of greenhouse gas emissions, the majority of

221
Last-mile delivery represents a disproportionate share of time and expenses for the T&L road value
chain. Aside from sharing apps, R&D initiatives also explore robotics solution to automatise this segment of
the value chain. InsiderIntelligence, 15/04/2022, The challenges of last mile delivery,
https://www.insiderintelligence.com/insights/last-mile-delivery-shipping-explained/
222
Less-than-truckload represents an important burden for road transport. Loading a truck with several
shipments causes much more complexities to organise delivery and quickly escalating costs.
223
For example, empty fleet management in shipping represents 8% of operational costs and remains
under-digitalised. AI solution could generate quick and important savings and optimisation. Cloud one, sin
dato, AI Case study 3: cost-saving AI in manufacturing logistics, https://www.tradecloud1.com/en/ai-case-
study-3-cost-saving-ai-in-manufacturing-logistics/
224
Global Railway Review (2021) “Deutsche Bahn expands use of artificial intelligence to improve
punctuality”: https://www.globalrailwayreview.com/news/123981/deutsche-bahn-artificial-intelligence/

82
 Adopt AI Study

which (71.7%) stems from road transport.225 Artificial intelligence can greatly
contribute to decreasing the environmental impact of mobility. For example, a pilot
test in Pittsburgh showed that smart traffic-management systems could reduce travel
time by 25% and cut polluting emissions by up to 21%. 226
Applications such as Intelligent Travel Systems (ITS) offer this type of environmental
upside. However, on top of their potential for the reduction of CO2 emissions,
modernised infrastructure is a prerequisite for the uptake and deployment of
autonomous vehicles in Europe. Clear marking for smart cars, a new 5G
communication network and transmitters installed on the transportation grid are
necessary for an autonomous car to show its entire potential.227
However, the implementation and deployment of Intelligent Transport Systems (ITS)
greatly differ between countries in Europe. South-East European countries, for
example, have a different level of deployment with difficulties in integrating
(interoperability) with other levels (cross-national and European integration).228
Several measures could help the modernisation of transport networks across Europe
and the standardisation of smart systems for the deployment of autonomous vehicles
across borders. Machin et al highlights several other AI applications focused on
environmental benefits such as fuel efficiency:
• Anti-lock braking systems
• Lateral and frontal control of vehicles
• Adjusting trajectories in Remotely Operated Vehicles (ROV)
• Parameters optimization for unmanned vehicles
• Stability improvement in vehicles
• Optimization of vehicle suspension systems
• Minimize consumption and emissions of Plug-in Hybrid Electric Vehicles (PHEVs) and
Electric Vehicles (EVs)
• Human-knowledge integration into Automatic Guided Vehicles (AGVs)

Use Case: Improved environmental performances

HOPU The HOPU solution brings a suite of dashboards with CO2

indicators related to mobility founded on predictive models
based on real-time air quality measurements and AI to
optimise the traffic flows in cities and ports over FIWARE open
platform
It allows third parties to provide high-value green and
innovative mobility services for citizens, companies, and public

225
EEA, 18/12/2020, Greenhouse gas emissions from transport in Europe,
https://www.eea.europa.eu/data-and-maps/indicators/transport-emissions-of-greenhouse-gases/transport-
emissions-of-greenhouse-gases-
12#:~:text=In%202017%2C%20transport%20(including%20aviation,increased%20by%200.7%20%25%2
0in%202018.
226
Sacyr, sin dato, Smart traffic lights to reduce air pollution, https://www.sacyr.com/en/-/semaforos-
inteligentes-para-reducir-la-
contaminacion#:~:text=The%20pilot%20tests%20in%20Pittsburgh,emissions%20by%20up%20to%2021
%25.
227
EC, May 2017, Public Supports measures for connected and automated driving,
https://op.europa.eu/en/publication-detail/-/publication/0f3e5c98-66ad-11e7-b2f2-
01aa75ed71a1/language-en
228
Rijavec Robert et all, January 2013, Intelligent Transport Systems deployment and integration in South
East Europe,
https://www.researchgate.net/publication/270571783_Intelligent_Transport_Systems_deployment_and_int
egration_in_South_East_Europe

83
 Adopt AI Study

administrations. This tool is focused on the decision-makers,

providing a set of functionalities to analyse and simplify large
amounts of data related to mobility and its impact. By
leveraging high-value open data, private data and satellite
data sets available in the cities, this process provides a
context for the real-time data that allows the definition of new
correlations and, for that reason, new mobility and urban
design aspects that affect air quality.229

Congestion control and prediction

Experts estimate around €100 billion in costs could be saved from traffic congestion
annually if traffic management systems could be improved 230. This improvement would
have other positive externalities such as a decrease in the number of accidents
(decreasing the human-factor in driving decisions, which is responsible for 90% of all
crashes) and lower emissions.
AI systems have the potential for improvement in traffic management that are far-
reaching such as forecasting demand, identifying passenger behaviour, automating
cumbersome tasks, help to design transport systems and networks.231Even though the
different transportation modes (maritime, airway, road and railway) are differently
affected by the challenges and difficulties related to traffic management systems, all
modes require a ramp-up in investments and the acquisition of new systems. In some
cases (e.g., port management and road management system) new technologies are
lacking and require new R&D investments (e.g., new sensors for an automated train in
opened areas). Cross-modal collaboration through an open ecosystem (data exchange
between connected infrastructure across the entire value chain) and technology
transfer (e.g., application of automated vehicles for rail) could improve the network
utilisation rate and decrease costs. The creation of an “Information Vessel” could
exchange along the entire value chain from a cross-modal perspective. Documents for
the transportation of merchandise can be hundreds of pages long and cumbersome.
232 233
There are efforts to standardise Electronic Data Interchange Systems (EDI).234
However, these approaches are too focused on B2G exchange, leaving aside B2B
documents that entail much more information and data. A bottom-up approach should
be preferred, including supporting the upscaling of initiatives already launched by the
industry (E.g., the ECLIC initiative in chemical transportation). Areas of smart traffic
management are not sufficiently investigated or acquired. This is the case, for
example, of smart parking management. Machin et al. provide a list of the most
common AI topics in this area:
• Traffic flow prediction in urban networks (short and long terms)

229
EIT Urban Mobility (2021) Urban Mobility Next#3 AI mobility Landscape in the EU
https://eit.europa.eu/sites/default/files/eitum-urbanmobilitynext3_final.pdf
230
Optimisation of transportation grid could generate savings of approximately €100 billion annually by
improving trafic congestion. European Parliament, January 2021, Artificial Intelligence in Road Transport
cost of non-europe report,
231
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
232
Transinfo, 2019, Standardisation in logistics is the inevitable future,
https://trans.info/en/standardisation-in-logistics-is-the-inevitable-future-sscc-number-can-make-things-a-
lot-easier-162905
233
Wartsila, 2019, Blockchain: the case for digitalising shipping,
https://www.wartsila.com/twentyfour7/innovation/blockchain-the-case-for-digitalising-shipping
234
International Journal of logistics systems and management, 2019, Challenges and opportunities for
logistics standardization in Asia-Pacific countries: a descriptive case-study, https://decisionsciences.org/wp-
content/uploads/2019/06/p584364.pdf

84
 Adopt AI Study

• Vehicles speed and route prediction

• Traffic volume prediction
• Route planning to avoid traffic jams
• Reduction of the time stopped at intersections
• Traffic signals control
• Traffic congestion reduction

Use Case: Congestion Control

Siemens AI’s processing, control and optimisation capabilities have

mobility – been applied to traffic management and decision-making
forecasting systems to enhance and streamline traffic management and
traffic and make our roads smarter, i.e., smart traffic light systems 235.
optimise
The project uses AI to optimise traffic light control and reduce
logistics
the waiting time at an intersection. Simulations suggest it can
decrease waiting times at lights by up to 47% compared to a
traditional pre-timed signal plan.

Road safety and accident prediction

Improvements in the management of the traffic system in conjunction with in-vehicle
AI technologies would have other positive externalities such as a decrease in the
number of accidents (decreasing the human-factor in driving decisions, which is
responsible for 90% of all crashes) and lower emissions. 236 Machin et al. highlights 3
principal ways in which AI can increase road safety and reduce accidents: prevention
of traffic accidents, analysis and processing of the circumstances that cause accidents
and the mitigation of the severity of accidents 237.
Risks posed by driver behaviour, traffic, the state of the roads and considerations such
as weather are data points which can be gathered by sensors in the vehicle and
processed using AI whether that be fuzzy logic or artificial neural networks – the two
most used AI techniques in this area per Machin et al. More specifically they forward a
list of systems and applications for increasing road safety which can be seen below:
• Intelligent systems for visual monitoring
• Vehicular accident modelling
• Accident frequency analysis
• Determining the causes of the accident
• Traffic accidents evaluation
• Driver fatigue detection
• Dangerous driving identification

235
EIT Urban Mobility (2021) Urban Mobility Next#3 AI mobility Landscape in the EU
https://eit.europa.eu/sites/default/files/eitum-urbanmobilitynext3_final.pdf
236
Optimisation of transportation grid could generate savings of approximately €100 billion annually by
improving trafic congestion. European Parliament, January 2021, Artificial Intelligence in Road Transport
cost of non-europe report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf
237
Machin, Mirialys & Sanguesa, Julio & Garrido, Piedad & Martinez, Francisco. (2018). On the use of
artificial intelligence techniques in intelligent transportation systems. 332-337.
10.1109/WCNCW.2018.8369029.

85
 Adopt AI Study

• Automatic incident detection

• Automated braking systems

Use Case: Road Safety

AI Aware – Using predictive AI algorithms, this collaboration between

Predicting and Volvo Car Corporation, HERE, Carmenta Automotive, Ericsson,
prevention Zenseact, Trafikverket and the Gothenburg City and Swedish
road crashes Transport Association, to detect when there is a risk of a
with AI vehicle colliding with another vehicle, a pedestrian or other
obstacle. It then sends predictive alerts to vehicles to prevent
the crash from happening238.

2.3.2 Value chain analysis

Schematically, the transportation & mobility value chain can be divided into three
stages between the origin and the point of destination: “storage and handling”,
transport execution” and “distribution/unloading”. The value chain is dominated by the
execution stage, which globally represents revenues of €2.3 trillion for a total world
market of €2.7 trillion. Road transport represents the biggest part of this segment at
€1.4 trillion. Sea transport ranks second with €312 billion in revenue (€199 billion for
the container, €60 billion for tanker and €51 billion for bulk).239
Figure 49 Transportation and mobility – value chain

Source: Three steps logic from L.E.K., other elements from author’s elaboration.

Equally important is the modal split for the organisation of the transportation &
mobility industry. Road, airway, railway and maritime transport face very different
challenges and realities. Furthermore, the competitive nature and type of players
operating differ across transportation modes. Maersk, for example, is a very large and
important company for sea transportation but has virtually no operation in the land
and rail segment. Consequently, the analysis examines each mode separately. For
each mode, infrastructure and vehicle fleet will be looked at separately. Closely linked,
these two elements face different challenges and are also operated by different types
of operators. The state typically plays a much more important role on the

238
Drive Sweden (2020) “AI Aware”: https://www.drivesweden.net/en/project/ai-aware
239
In a “T&L (Transportation & logistics) perspective, the postal segment would have to be added. BCG,
2016, Transportation and logistics in a changing world, https://www.bcg.com/publications/2016/corporate-
development-finance-value-creation-strategy-transportation-and-logistics-in-a-changing-world.aspx

86
 Adopt AI Study

infrastructure side (funding, organising, or directly owning them) 240. The different
opportunities and recommendations presented in this section are displayed in the
figure below.
Figure 50 Summary of conclusions for value chain analysis.

Source: Three steps logic from L.E.K., other elements from author’s elaboration.

Road transportation
Road transportation is the segment that displays the most potential for AI applications
with rapid progress being made even though the uptake of AI solutions is focused
mostly on vehicle fleets.
Indeed, AI systems for road transportation already exist and represent attractive
opportunities for operators and companies: “AI systems applied in the transport sector
can already now detect patterns in a large volume of data and model complex
solutions that enable increased efficiency in decision making and better resource
allocation. For example, AI technologies are used for 'real-time or predictive matching
of supply and demand for rides or goods, predicting traffic speeds or dangerous road
segments and behaviours, and managing supply chains”. 241
Most expectations and efforts to develop AI in road transportation are focused on
autonomous vehicles and the deployment of smart traffic management systems.
Consequently, most of the initiatives and projects led by the EC are focused on
enabling the deployment of CAVs (Connected and Autonomous Vehicles), e.g.,

240
In T&L, infrastructure is the first stage on which the rest of the T&L pyramid is built and it is also the
most asset-intensive one. The second stage includes the logistics execution players, which entails relatively
important asset-intensity as it requires owning and operating a transportation fleet. The third stage
includes the freight forwarder and the contract logistics. Contract logistics providers usually cover all the
T&L activities while freight forwarders cover the steps ranging from picking up to delivery, thus leaving
aside warehousing and other services. The freight forwarder and contract logistics allow their customer to
outsource a larger share of their T&L activities, however, they do not directly own assets but rather rent or
subcontract them. The fourth and final stage is advisory services that require only regular offices and a very
low asset-intensity. These players are focused on optimising and outsourcing activities but not on directly
operating T&L processes or owning assets. BCG, 2016, Transportation and logistics in a changing world,
https://www.bcg.com/publications/2016/corporate-development-finance-value-creation-strategy-
transportation-and-logistics-in-a-changing-world.aspx
241
European Parliament, January 2021, Artificial Intelligence in Road Transport cost of non-europe report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf

87
 Adopt AI Study

updating and developing cooperative intelligent transport systems (C-ITS), including

digital traffic management systems.242
However, greater effort is needed to address the “EU-level liability on AI and on
enhancing the trust of users of AI in the road no further action is taken at the EU level
on liability in AI and on enhancing the trust of users of AI in road transport could
amount to between €231 097 and €275 287 million, were none of the gaps and
barriers analysed addressed. This EU action would be also beneficial for employment
and could create between 5,181 and 6,147 million jobs”.243 Aside from autonomous
vehicles, several other main areas for the uptake of AI solutions can be identified in
the Road transport segment:244
 Truck platooning;
 Road traffic management;
 Automated traffic lights;
 Sharing ride platforms;
 Automated vehicles and last-mile delivery;
As public authorities do not operate fleets or fleet management systems themselves,
there is a limit in terms of what public authorities can achieve through direct
procurement. Moreover, road transportation is a very competitive and dynamic market
where investments from the private sector are already very important. 245
Most efforts should therefore be concentrated on the legal framework, infrastructure
modernisation and digital seamlessness. In that regard, numerous projects already
exist in the field with Horizon funding and other initiatives such as the development of
electronic freight transportation documents. 246
An overhaul is necessary to revise existing liability legislation, cybersecurity, and
ethical framework for trustworthy AI and to develop a “common approach to liability
rules and insurance for connected and autonomous vehicles”247 Acceleration of the
adoption curve of CAV generated by revised EU liability framework could generate
around €148 billion in added-value. “EU joint legislative action on ethical standards for
AI systems could 'boost the internal market and establish an important strategic
advantage” representing a potential gain of “€294.9 billion in additional GDP and 4.6
million additional jobs by 2030.” 248

242
Autonomous vehicles could increase EU’s GDP by 5.3% for the 2016-2050 period and generate €17
trillion of income for the same period. Most of the gains would come from improvements of internal
processes and overall functioning of the transportation sector. Moreover, CAVs could generate reduction in
fuel consumption, CO2 emissions (by 1.2% annually) and improve road safety. Optimisation of
transportation grid could generate savings of approximately €100 billion annually by improving traffic
congestion. European Parliament, January 2021, Artificial Intelligence in Road Transport cost of non-europe
report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf
243
European Parliament, January 2021, Artificial Intelligence in Road Transport cost of non-europe report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf
244
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
245
The total global investment in autonoumous vehicle only already exceded $200 billion in 2022. Forbes,
14/02/2022, Autonomous vehicles and their impact on the economy,
https://www.forbes.com/sites/forbestechcouncil/2022/02/14/autonomous-vehicles-and-their-impact-on-
the-economy/?sh=53f9202f60de
246
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
247
European Parliament, January 2021, Artificial Intelligence in Road Transport cost of non-europe report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf
248
European Parliament, January 2021, Artificial Intelligence in Road Transport cost of non-europe report,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/654212/EPRS_STU(2021)654212_EN.pdf

88
 Adopt AI Study

Efforts are already granted for the deployment of smart national road infrastructure
systems. However, cross-borders interoperability and cross-model interoperability are
still missing. Furthermore, most efforts are focused on B2G exchanges and leave aside
B2B, which represents both more data and more market potential for new
applications.
Rail
Passenger and freight segments of railway transportation are displaying different
dynamics. Passenger traffic is growing on average by 1.7% per year while freight
volume still hasn’t fully recovered from its 2009 drop. The railway is an energy-
efficient mode of transportation and only accounts for 2% of the total EU energy
consumption in transport and only 0.5% of CO2 emissions. Contrastingly, rail
transported 12.2% of all freight and 6.6% of all passengers in 2016. 249
The main challenge for the rail business model is the competition from road
transportation and the pressure it represents on its competitiveness and prices.
Indeed, rail freight is more expensive than road freight. Furthermore, the revenue
generated by passengers decreased with the number of kilometres. 250 In addition, the
sector suffers from a lack of competitiveness caused by the lack of interoperability and
cross-border cooperation between rail operators.251
Consequently, railway operators must focus on cost leadership strategy without
jeopardizing quality, something that AI systems can contribute to. 252253 However,
railway operators also have to simultaneously address the congestion of their network
in the context of growing traffic. Aware of these difficulties, public authorities have
increased their investment in railway infrastructure, rising from €29 billion in 2011 to
€ 50 billion in 2015 (for both maintenance and improvement of infrastructure). The
European Fund for Strategic Investments represented €3.5 billion of that sum. 254
Nonetheless, European rail infrastructure tends to be much older than in other
regions, leading to more important costs and difficulties in fully modernising them.255
AI can apply to each process involved in operating railways, including chatbot to assist
customers, smart ticketing for demand forecast, robotics in railway and maintenance
(e.g., drone for track monitoring and automated inspection), predictive maintenance,
warehouse robotics, etc. However, despite the potential represented by AI to address

249
European Commission, February 2019, Sixth report on monitoring development of the rail market,
https://transport.ec.europa.eu/system/files/2019-02/staff_working_document_-_6th_rmms_report.pdf
250
Rail passenger remains domestic and its share in all modes grew from 7% in 2007 to 7.6% in 2016 —
even though personal care remains above 80% of all transport. Freight volume remains important at 17%
but show some decline compared to road. Track access charges represent 80% revenues for railway
operators, this represents less than €3 per train-kilometre in average in the EU. Shift2Rail, 8/12/2021,
Summary of existing relevant projects and state-of-the-art of AI application in railways,
https://www.researchgate.net/publication/352440235_RAILS_Project_Deliverable_D12_Summary_of_existi
ng_relevant_projects_and_state-of-the-art_of_AI_application_in_railways
251
European Commission, February 2019, Sixth report on monitoring development of the rail market,
https://transport.ec.europa.eu/system/files/2019-02/staff_working_document_-_6th_rmms_report.pdf
252
UIC, March 2021, Artificial Intelligence case of the railway sector,
https://uic.org/IMG/pdf/artificial_intelligence_case_of_the_railway_sector_state_of_play_and_perspectives.
pdf
253
Shift2Rail, 8/12/2021, Summary of existing relevant projects and state-of-the-art of AI application in
railways,
https://www.researchgate.net/publication/352440235_RAILS_Project_Deliverable_D12_Summary_of_existi
ng_relevant_projects_and_state-of-the-art_of_AI_application_in_railways
254
Typically, public authorities currently focus their efforts on public service obligations followed by
international passenger’s flow and freight services. European Commission, February 2019, Sixth report on
monitoring development of the rail market, https://transport.ec.europa.eu/system/files/2019-
02/staff_working_document_-_6th_rmms_report.pdf
255
Interview with a representative from a standardisation body in railway.

89
 Adopt AI Study

the challenges met by railway operators, there is a lack of market uptake for these
solutions.256
Mainly, two areas can be identified as especially attractive for railways: “Intelligent
train automation” (ATO) and predictive maintenance. ATO transfers train control
responsibilities from the driver to the computer (with different levels of attribution
based on the level of automation). To deploy ATO on a larger scale in Europe, the
development and deployment of European rail traffic management systems (ERTMS)
aimed at harmonising rail control systems is an absolute prerequisite.257 However,
more R&D is necessary to make these systems applicable in open areas and not solely
in the closed metro environment.258
Predictive maintenance is another promising field for AI applications in Railways.
Especially on the infrastructure side. 259 Rail infrastructure is more mature with a
higher level of digitalisation meaning that data are already there to implement AI-led
predictive maintenance applications. Rolling-stock, however, is less mature and
requires more R&D development (for instance, to create a solution to identify wheel
tread defects).260 261
One uniquely European difficulty for wider uptake of AI solutions is the fragmentation
of its rail market causing a lack of interoperability. Operators sometimes have to
“guess” what systems and standards are used in another country as each operator has
its ticketing systems, applications, etc. This is a very important challenge as AI
systems require a high volume of good-quality data to be efficient. Without this,
Machine Learning and AI applications are impossible. Consequently, the Shift2Rail
initiative was therefore launched to develop a common interoperability framework at
the European level. 262
The competitive dynamics within the railway sector also cause difficulties in
encouraging wider uptake. Railway operators are competing with road transportation
rather than with other railway companies. Consequently, investing more efforts in
achieving higher interoperability and sharing data is sometimes perceived more as an
expensive and complex administrative burden. Developing new incentives to

256
UIC, March 2021, Artificial Intelligence case of the railway sector,
https://uic.org/IMG/pdf/artificial_intelligence_case_of_the_railway_sector_state_of_play_and_perspectives.
pdf
257
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
258
“In addition to ensuring technical compatibility between national rail systems, the ERTMS combined with
ATO can reduce rail operators' costs and energy consumption, and increase rail speed (up to 500 km/h),
punctuality, safety and line capacity.” European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
259
Upgrade and maintenance of infrastructure should focus on switches which “are the most critical part of
the rail infrastructure, causing approximately 20% of infrastructure-related delay minutes [of the trains]
and costing €12bn a year globally to maintain and replace”. UIC, March 2021, Artificial Intelligence case of
the railway sector,
https://uic.org/IMG/pdf/artificial_intelligence_case_of_the_railway_sector_state_of_play_and_perspectives.
pdf
260
UIC, March 2021, Artificial Intelligence case of the railway sector,
https://uic.org/IMG/pdf/artificial_intelligence_case_of_the_railway_sector_state_of_play_and_perspectives.
pdf
261
For example, SNCF in France started to use predictive maintenance to pantographs and enable to
forecast 80% of all catenaries incidents (the piece that supply electricity to the train) and reduce incidents
related to train switches by around 30%. Ongoing project are developing the possibility for train to transmit
“health diagnostic” to the fleet supervisor. We can also mention Devices for long-term assessment of asset
performance and development of digital twins (e.g., the Rete Ferroviara Italiana developed digital twinning
for its network). European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
262
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en

90
 Adopt AI Study

encourage Door2Door logistic competitiveness would be the key to encouraging more

uptake from railway operators. 263
Railway infrastructure remains mainly state-owned and 70% of investment in
infrastructure still comes from national budgets. 264 Furthermore, train operators face
“Public Service Obligations” for their operations while the state-owned or formerly
owned company still represent the main operator in most countries (100% of market
operation in France for example).265 This situation leads to low competitive intensity in
the market. Public authorities will have to play the leading role in purchasing,
installing and deploying AI solutions for rail. ATO and predictive maintenance are the
most interesting areas to focus investments and acquisitions. More R&D efforts are
needed to tackle specific barriers such as the lack of technical solutions for the
maintenance of rolling stock and implementation of ATO in open areas. 266
To generate a greater ripple effect and potentially decrease costs, the development of
cross-modal collaboration and development has to be explored. Few examples exist
yet the railway industry could benefit from solutions developed in road transportation
for autonomous vehicles. Joint research efforts and transposition of these solutions in
the rail sector is an interesting lead to explore, especially for cross-modal barriers to
AI such as in cybersecurity, modular warehousing robotics, workforce upskilling, etc.
267

The fragmentation of the European market is a barrier to business competitiveness

and the deployment of AI solutions. Greater interoperability is a key prerequisite for
the development of new AI applications in the railway. Legislation on the B2B and B2G
side of data exchanges and interoperability is needed. These legislations should focus
on making exchanges and interoperability mandatory while leaving aside the technical
question to railway operators. 268 Data sharing however also involves the question of
data sovereignty. Something that could be protected, for example, through the
development of a data space specific to railway operations or if the operator (data
owner) is left in charge of sorting who can access and use their data.
Maritime
Sea trading represents 76% of Europe’s external trade. Some 90% of all of the goods
consumed were shipped at some point in their value chain and the EU shipping
industry represents a total of €147 billion. 269 These figures highlight the importance
taken of maritime transportation for all industrial sectors and the European economy
as a whole.
As for other transportation modes, shipping vessels have turned into data factories
with the deployment of new digital solutions such as radar, electronic navigation

263
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
264
European Commission, February 2019, Sixth report on monitoring development of the rail market,
https://transport.ec.europa.eu/system/files/2019-02/staff_working_document_-_6th_rmms_report.pdf
265
McKinsey, October 2019, Navigating the EU rail-market liberalisation,
https://www.mckinsey.com/industries/travel-logistics-and-infrastructure/our-insights/navigating-the-eu-
rail-market-liberalization and McKinsey, July 2019, The liberalisation of the EU passenger rail market,
https://www.mckinsey.com/~/media/mckinsey/industries/travel%20logistics%20and%20infrastructure/our
%20insights/navigating%20the%20eu%20rail%20market%20liberalization/the-liberalization-of-the-eu-
passenger-rail-market-vf.pdf
266
improvement of asset geolocalisation is necessary for automation of trains. Improved sensors have a
range of 50-70 meters while shorter range of 1-5 meters are needed to roll in opened area (currently under
development by the SNCF and the French Space Agency). European Parliament, March 2019, Artificial
Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
267
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
268
Interview with a representative of a European public railway company.
269
ECSA, 2019, European Shipping sets ambitious goals for its next chapter,
https://www.ecsa.eu/sites/default/files/publications/ECSA%20-%20Priorities%202019-2025.pdf

91
 Adopt AI Study

charts, autopilot systems, wave radars, oil-spill detectors, and high-accuracy

sensors.270
Pressured by the fear of protectionism and the lasting consequences of the 2008 crisis
on their profitability, shipping actors have ramped up their efforts in increasing the
digitalisation of their value chain and operations mainly to achieve greater cost-
efficiency and increasing competitiveness. 271
However, several important differences with other sectors can be observed. For
instance, automated shipping is of low interest to shipping companies as it would
involve developing a new business model. Instead, shipping companies are especially
interested in the development of AI for data utilisation for smart shipping to optimise
operations, and onboarding and increase cost-efficiency.272 273
This strategy led to the development of cooperation initiatives for the development of
common standards and sharing ecosystems (E.g., the Blockchain in Transport
Alliance). However, technology uptake greatly varies between countries and shipping
companies. Non-EU European countries, for example, have a much lower uptake of
distributed ledger and less modern ICT systems.274 This fragmentation, in turn,
creates a lack of interoperability and difficulty for B2G and B2B data exchanges
(something necessary to enhance data utilisation).
Automatic Identification Systems (AIS) have been developed to streamline B2G and
infrastructure to vessel data exchanges (including data about the ship's identification
number, position, course, speed and destination). Datasets that represent the
important potential for analytics and that are still under-used so far.275 On the
infrastructure side, the “smart port” concept has arisen for the digitisation and
organisation of activities on the port (including loading and unloading). In total, the
port call optimisation platform (such as the one implemented in Rotterdam) can use AI
to reduce waiting times by 20%. However, challenges remain for wider
implementation stemming from low data quality and quantity (erroneous data from
human or sensor mistakes).276
Similarly, port activities could benefit from greater automation and standardisation.
However, there is a lack of market demand for these solutions and no genuine Europe-
wide coordination. Most efforts and investments are done at the port level and then
coordinated through port alliances or cooperation. Most of the discussions on the
matter still take place at a high level while there is still a lack of implementation and
understanding of what field applications could achieve for the industry. 277

270
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
271
This however does not mean that AI-support to pilots won’t be created. But full automation is not the
focus of shipping companies. Ichimura Yuki et all, March 2002, Shipping in the era of digitalisation,
https://www.sciencedirect.com/science/article/pii/S2666954422000023
272
Ichimura Yuki et all, March 2002, Shipping in the era of digitalisation,
https://www.sciencedirect.com/science/article/pii/S2666954422000023
273
We can however name the NOVIMAR and MUNIN projects in Europe for the development of “shipping
vessel platooning” and autonomous ships. European Parliament, March 2019, Artificial Intelligence in
Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
274
Kapidani Nexhat et all, 15/09/2020, Digitalisation in developing maritime business environments towards
ensuring systainability, https://www.mdpi.com/2071-1050/12/21/9235
275
Initiatives were also launched to simplify exchange of information between ships and port for electronic
freight transport information and cross-border operations. Advanced use could help to gain in energy
efficiency, detection of anomalies, navigation optimisation, predictive maintenance, traffic congestion and
forecasts for bunker needs. European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
276
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
277
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en

92
 Adopt AI Study

In general, despite genuine intent, the level of digitalisation of the shipping industry
(infrastructure and vessels) remains limited. The shipping industry business model
relies on secrecy. Disclosing information (e.g., availability of shipping capacity, the
exact position of ships) would change the bargaining power to the detriment of
shipping companies. Big players also have such a large reach that they prefer to
develop and keep their internal solutions and tend to not use APIs enough. Despite
discourse about collaborating practices, most cases for standardisation take place
bilaterally or even internally.278
Nonetheless, large shipping companies are already engaged in standardisation and
digitalisation efforts while ports tend to lag. The impact of AI systems on port
automation is especially important and interesting to decrease the consumption of
fuel. This is especially important in the context of rising prices and stricter regulations
to make the industry greener. 279
Public procurement can play an important role in the implementation of AI solutions in
port activities. It would represent an important gain in operational efficiency and fuel
savings. Greater purchases, coordination at the European level and development of
pilot projects could generate replications efforts. 280 Authorities could play a role in
making specific data (either B2B or B2G) exchanges mandatory. This could address
the lack of exchange between ports to vessels and encourage the development of
sharing practices. 281
European collaboration in the matter is very limited as activities are left to port
operators. European-scale coordination for the acquisition, implementation and
standardisation of AI systems for port operations could give a competitive advantage
to the EU. Implementation of these solutions could generate replication efforts from
the industry. 282
Air
Recent decades have seen tremendous progress in the uptake of AI solutions in air
transportation. For example, the development of private data spaces for analytics and
machine learning (e.g., Skywise from airbus) or the use of digital twinning for the
design of aeroplanes.
However, many areas remain lowly digitalised and represent important opportunities
for increased use of AI. For instance, the use of AI systems in air traffic control and air
fleet management is still in its infancy.283 There is still a lack of initiative to improve
air traffic management and ensure a high degree of seamless air-ground
integration.284

278
Shipping contracts can be done based on route length or used bunker. Consequently, shipping companies
are discouraged to optimise their routes and can sometimes purposedly prolong a trip. This leads to higher
bunker consumption, inefficiency and greater waiting time for port operations because of miss-
communications. Interview with a representative from a startup in platform for port operations.
279
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
280
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
281
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
282
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
283
European Parliament, March 2019, Artificial Intelligence in Transport,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/635609/EPRS_BRI(2019)635609_EN.pdf
284
For example INTUIT to explore the potential of machine learning and visual analytics: COP>TRA on
trajectory prediction. MALORCA speech recognition to encore in recognition software for human to machine
and ground to air coordination.

93
 Adopt AI Study

For aeroplane fleets, Smart Maintenance is the most interesting area for the
development of new impactful AI solutions. Maintenance has already been digitalised
by many MROs (Maintenance and Repair Organisations) internally using “Electronic
Maintenance Records”.285 However, information remains siloed and kept in-house. The
developed data space platforms in the field lack cross-platform interoperability. SMEs
are especially affected by this problem as they face more challenges in accessing
these tools. Furthermore, maintenance is mandated by law, but the way it is
documented is not. This leads to inconsistent reporting of maintenance information
and difficulties to implement automation. 286 Air Traffic Management (ATM) is the area
where public procurement would be the most impactful. Public procurement of
modules, systems and support for an R&D project for Air Traffic Control (e.g., for
developing explainability of AI systems) could transform ATM. 287 This would represent
gains in operational efficiency, and ground-air connectivity and encourage the
deployment of on-board solutions to exploit new opportunities opened by enhanced
connectivity with ground operations. The development of standards requirements for
safety and maintenance would encourage standardisation and uptake of automated
smart maintenance systems.
2.3.3 Main drivers and barriers
The table below presents a PESTEL analysis of the uptake of AI technologies in public
procurement in the e-government sector.

285
Fernadnez Antonio, 16/02/2019, How Blockchain could enhance aircraft maintenance,
https://datascience.aero/blockchain-enhanceaircraft-maintenance/
286
CARSA et al, 02/12/2021, Industry agreements in current value chain,
https://op.europa.eu/en/publication-detail/-/publication/8c021023-53ee-11ec-91ac-
01aa75ed71a1/language-en
287
Degas AUgustin et all, December 2022, A survey on artificial intelligence and explainable AI in air traffic
management, https://www.mdpi.com/2076-3417/12/3/1295

94
 Adopt AI Study

Table 5 PESTEL analysis of the uptake of AI technologies in public procurement in the mobility and transport sector

Criteria Drivers Barriers

 The promise of establishing a niche in the  Lack of harmonised rules and regulations in the
market related to AI in transport, especially in adoption and integration of AI in mobility and
areas such as AI-driven Autonomous Vehicles transport across Europe.
technology deployment.
 Need for the inclusion of urban AI in EU
Political  The adoption of AI in mobility and traffic research programs addressing data exchange,
monitoring can play an important role in the communication networks, and policy on
decision-making of public organisations to mobility.
leverage the implementation and development
of new or existing public transport services,
mobility infrastructures, etc.

 The ability of AI solutions to reduce costs and  A major constraint on the growth and
significantly lower public spending: Deciding development of AI in the transport market is
whether to build a new road, how much money the high cost of some AI systems.
should be allocated to maintenance and
rehabilitation activities and which road
segments or bridges to maintain, and whether
to divert traffic to an alternative route in an
Economic incident situation.
 AI can also help to manage and utilize large
amounts of data and help to plan, design and
control road transport networks. It can also
help optimize movements to maximize
efficiency contributing to significant cost
reduction for instance in terms of logistics,
public transport or traffic management.

95
 Adopt AI Study

 The new market for automated and connected

vehicles is expected to grow exponentially and
large economic benefits are expected.

 Safety: Road safety for both drivers and  Sharing data and changing the rigid transport
pedestrians are a major public health issue. business models in separate modes towards a
While inadequate infrastructure—in particular, more dynamic network that joins technological
poor roads and vehicles without modern safety platforms, mobility providers and customers, is
equipment—plays a role in the high death toll, a difficult task. Establishing commitment from
human error is an important contributor. top management to drive the cultural and
process change required is a current barrier.
 Potential to enhance personal autonomy,
especially for people with impaired driving  Concerns over loss of jobs including delivery
Social abilities. Additionally, it can contribute to and heavy truck drivers, bus drivers, taxi
providing timely and accurate transit travel drivers, and chauffeurs. AI is likely to accelerate
time information, which can attract ridership the transition toward a service economy,
and increase the satisfaction of transit users. upending established economic development
models by speeding up job losses for low-skilled
 AI is helping to take the personalization of the workers in many fields, including transport.
mobile user experience to the next level.
Personalization is increasingly relevant as
mobility systems evolve toward greater human
centricity and sustainability.

 The introduction of AI enables the improvement  For any AI use, the collection, quality,
of vehicle efficiency in terms of minimising fuel coherence and volumes of data available are
consumption. Additionally, it presents the paramount. Some data quality and quantity
opportunity to invest in research of alternative issues arise with transportation data collected
fuels. by sensors and data acquisition systems.
Technological Missing or incomplete data and subpar accuracy
 Empowering electrical vehicles, as the and availability of data directly impact the
application of AI technologies in transport
quality and trust in urban AI systems.
presents opportunities to further develop
aspects such as battery research and  In several trials and pilots over recent years,
some limitations of current technologies have
arisen and they will need to process even more

96
 Adopt AI Study

development, battery energy management, data to realize the expectations in the years
vehicle-to-grid algorithms, etc. ahead.
 Opportunity to create a unified mobility model,  Demand for AI experts has grown over the last
in which there is open data sharing leading to few years in developed countries and EMs
the generation of reliable, robust and high- where AI investment is increasing. A lack of
quality data sets and unified governance across skilled AI talent has been widely cited as the
all mobility modes, which allows going far largest barrier to AI adoption in developed
beyond traditional solutions. countries. When it comes to AI within local
authorities, the implementation is often
outsourced towards commercial parties
providing tailored AI as a service. However,
digitalisation and AI-related expertise are
necessary for local governments to adequately
assess internally or externally developed AI on
their aspects of trustworthiness.

 Management of traffic proactively with data-  Lack of clear cost-benefit analysis for the entire
driven insights; to automatically detect set of AI solutions. Environmental analysis is
incidents for faster response; to more efficiently usually focused on the gain and positive
and effectively manage bottlenecks on the aspects, while negative impact, such as the
road; and to identify and target traffic implementation of larger server farms, is
violations. This results in smoother traffic, usually left aside. Therefore, the positive impact
Environmental reduced congestion and carbon emission. might be backfired on the real and global view
Therefore, it directly contributes to the environmental impact of AI.
reduction of greenhouse gas emissions,
pollution and noise related to traffic.
 Better management of urban space and
reclaiming specific urban areas for residents.

 Development of an updated and harmonised  The regulatory requirements for AI are difficult
EU-wide liability framework for connected and to predict, especially when it comes to who
Legal autonomous transport. would ultimately be held liable if an AI-powered
service were to cause an accident, harm, or
fatality. Setting out the responsibilities of

97
 Adopt AI Study

 Build and maintain a balanced regulatory manufacturers, operators and drivers depending
environment, based on existing applicable on the level of automation is another important
regulations, that enables and stimulates future common challenge, to finding appropriate
technological innovation and evolution. answers to new legal issues concerning liability
and ethics.
 Asking users to opt-in and provide more
personal data for machine learning requires
robust privacy laws. These laws must be
balanced against the benefit of having more
data in a telecommunications network.
Source: Authors’ elaboration based on the challenges and the value chain analysis.

98
 Adopt AI Study

Figure 51 Strengths, Weaknesses, Opportunities and Threats of the uptake and public procurement of AI-technologies in
the mobility and transport sector

S
Better quality of life for citizens – Adoption of AI in transport has the potential to
improve life quality of all citizens by: increasing road safety for both drivers and
pedestrians, enhancing personal autonomy especially for people with impaired driving
abilities and personalizing the mobility user experience.
Growing market - The new for automated and connected vehicles is expected to grow
exponentially and large economic benefits are expected. The automated vehicles market
has propelled a significant shift in automation and connectivity and the integration of
advanced technologies, including AI, is taking place to help and automate driving
operations in different means of transport.

W Rigid business models – Mobility and transport sector presents rather rigid business
models. Therefore, reluctancy towards data sharing, more dynamic technological
platforms and application of AI technologies is considered a barrier that needs to be

O
treated in order to change the current cultural and process paradigms.

Efficient decision making – AI provides the opportunity to significantly improve decision

making in several aspects such as deciding whether to build a new road, how much money
should be allocated to maintenance and rehabilitation activities and which road segments
or bridges to maintain, and whether to divert traffic to an alternative route in an incident
situation.
Environmental improvements - AI in mobility present vast opportunities to detect
several aspects of traffic and mobility, such as incidents, bottlenecks, etc. that can be
quickly and efficiently treated to guarantee smoother traffic, reduced congestion and
carbon emissions, with a direct impact in the environment.
Growth of urban areas – AI enables a better management of urban space and
promotion of specific urban areas for residents together with leverage the implementation
and development of new or existing public transport services, mobility infrastructures, etc.

T
that will enhance the development of cities and towns.

Job loss – Part of the society believes that the adoption of AI will lead to job loss, as
machines will replace humans.
Privacy concerns - Asking users to provide more personal data for machine learning
requires robust privacy laws and the existence of a balanced regulatory environment and
liability frameworks.

Source: Authors’ elaboration based on the challenges, value chain and PESTEL analysis.

2.3.4 Case Study: The Mobility Data Space (Germany)

Key Actors
The MDS initiative already encompasses 200 stakeholders coming from multiple
different parts of the mobility industry. These members can be split into the following
categories:288
 Data users and providers: science, business, and public administration.
 Founder and coordinator: the Acatech National Academy of science and
engineering.
 Co-partners to support the coordinator (since October 2021): Caruso, Deutsche
Bahn, Deutsche Post DHL, HERE, HUK-Coburg Versicherung, Mercedes-Benz as

288
The Mobility Data Space, initiative’s website, https://mobility-dataspace.eu/

99
 Adopt AI Study

well as BMW via the BMW Intec Beteiligungs GmbH and Volkswagen via
Volkswagen Group Info Services AG.
 Managing body: the “DRM Datenraum Mobilität GmbH“. Created as a
sponsoring company (non-profit limited liability company). The states of North
Rhine-Westphalia and Baden-Württemberg are co-shareholders since 2022.
MDS is a member of the IDSA as well as it is the first operating project of Gaia-
X.
Goals
The development of a Mobility Data Space (MDS) was instigated as an initiative of the
German federal “Concerted Action on Mobility Committee” in 2019. The objective of
the initiative was to create one common data space for mobility data (a one-stop
shop). Quickly gaining traction, the German initiative has now reached a European
scale.
Description
The MDS acts as a marketplace and builds on the architecture and principles defined
by Gaia-X and IDSA. The MDS provides a secure architecture that acts as an
intermediary to put data buyers and providers in contact. The initiative does not have
any centralised data lake. Instead, the system is organised around a series of
“connectors”. These connectors are not large “pipes” where data can transit and
instead data remains warehoused on the participants’ servers. These connectors are
closed environments where data from other members can be accessed through a
system of API.289
Up and running, the MDS already generates added value for participants. With this
safe and trusted ecosystem for data sharing, participants have access to the data they
need for new use cases and AI applications. Even though the initiative remains
somewhat immature in terms of business model, the promise it holds in terms of use
case, in breaking data silos and bringing together private and public players already
makes it a shining example in Europe. The initiative provides template licence
agreements for data exchanges. Furthermore, the connectors have a built-in system
to monitor activities and verify compliance by partners to terms and agreements. 290
However, data exchange has to be agreed upon bilaterally. Participants have to
contact each other to agree on the terms of the exchange. There is no standard in
terms of data format, content, quality or volume. Participants also have to agree on
the temporality of the exchange: i.e., access to data can be provided for a specific
period (one month, 2 weeks, etc) or even with regular updates (e.g., new data sets
will be provided each month). Similarly, price and valuation mechanisms are left to
the participants. They can decide to provide data for free, decide bilaterally on the
price or proceed to barter. 291
Such a decentralised network requires a “central directory in which data sources and
services are published and which can be searched either manually or automatically by
data users. With different operator and business models, one or more central
components for the data space can be offered”:292 The directory is therefore a
repository that provides a list and description of the data available on the data space,
their characteristics, formats, etc.
AI use case. The connector system does not place any limit on the use that can be
made for the data acquirer. These limitations have to be defined by the partner before

289
Ibid. Also see Fraunhofer, March 2021, Mobility Data Space,
https://cefic.org/app/uploads/2022/01/Cefic-position-on-transport-and-logistics-digital-collaboration-and-
data-sharing.pdf
290
Second interview with a representative of the Mobility Data Space.
291
First interview with a representative of the Mobility Data Space.
292
Ibid

100
 Adopt AI Study

the exchange takes place. Participants can use connectors as closed and safe
environments to run their algorithms and machine learning systems. Mainly, four
types of data are shared on the MDS: 293
 Weather data: to link with mobility data such as traffic jams;
 Infrastructure data: e.g., e-mobility infrastructure, smart traffic light, etc;
 Road safety data: data on hazards as identified and analysed by the vehicles
to adapt traffic management and warn users);
 Environmental data: a combination of environmental information with other
data (e.g., road utilisation) to generate positive environmental externalities.
As the initiative acts as a marketplace, the possibilities opened to AI depend on the
creativity of the participants, and their ability to conclude data exchange agreements.
However, the main use case is related to autonomous driving. To be efficient, an
autonomous vehicle needs to be able to connect and exchange with other vehicles and
the smart infrastructure. For example, in poor weather conditions, a vehicle can have
difficulties identifying if the light is red or green. Accessing the data from the
infrastructure to identify under what condition (e.g., level of brightness), these signals
can be correctly identified can help manufacturers in improving their systems. 294 In
addition, we can also identify other potential use cases that will grow in the future
for traffic management, improvement of embarked electronics and preventive
maintenance.
Successes and Key Lessons
 Breaking Data Silos in the mobility ecosystem which had been noted as a
barrier in both the public and private spheres 295. Until then, dataspace
initiatives were mainly developed at the regional and local level for smart cities,
smart traffic management systems and similar types of initiatives. 296 Similarly,
the MDS was set to absorb the Mobility Data Marketplace initiative launched in
2017. The MDM is a marketplace for data exchange in the mobility industry. 297
Indeed, the MDS was launched to build an integrated data ecosystem bringing
SMEs, start-ups, public authorities and OEMs together to restore German
competitiveness against the US tech giants. 298
 The federal government played an important role in driving the
development of the MDS and in convincing industrial partners to get on board.
The federal government for example granted €18 million for the development of
the MDS299 and €3 million for the development of the preceding initiative, the
Mobility Data Marketplace.300 In addition to this financial support, the
government was active in convincing large OEMs — without which the

293
Ibid.
294
Second interview with a representative of the Mobility Data
295
Next-mobility, 17/01/2022, Datenraum mobilität: wie Europa sich gegen die US-Tech-Giganten rüstet,
https://www.next-mobility.de/datenraum-mobilitaet-wie-europa-sich-gegen-die-us-tech-giganten-ruestet-a-
1088354/
296
Mobility Data Space, sin dato, Connection of data platforms, https://www.mobility-data-
space.de/en/connection.html
297
Next-mobility, 17/01/2022, Datenraum mobilität: wie Europa sich gegen die US-Tech-Giganten rüstet,
https://www.next-mobility.de/datenraum-mobilitaet-wie-europa-sich-gegen-die-us-tech-giganten-ruestet-a-
1088354/
298
Next-mobility, 17/01/2022, Datenraum mobilität: wie Europa sich gegen die US-Tech-Giganten rüstet,
https://www.next-mobility.de/datenraum-mobilitaet-wie-europa-sich-gegen-die-us-tech-giganten-ruestet-a-
1088354/
299
Handelsblatt, October 2020, Merkel Drängt autokonzerne: BMW, Daimler und VW sollen Datenschatz
teilen, https://www.handelsblatt.com/politik/deutschland/autogipfel-merkel-draengt-autokonzerne-bmw-
daimler-und-vw-sollendatenschatz-teilen/26308418.html?ticket=ST-1376973-6msjtMrmiOfcreSP5PK1-ap2
300
APCO Worldwide, 7/12/2020, Has Germany set the European Transport sector on the path to a digital
transformation?, https://apcoworldwide.com/blog/has-germany-set-the-european-transport-sector-on-the-
path-to-a-digital-transformation/

101
 Adopt AI Study

initiative would have little sense — to embark on the digital journey This was
achieved through public statements held by the then Chancellor Angela Merkel,
participation in public events and fair by officials and dissemination of a
covenant to be signed by joining partners.301
 Another important step to build trust between private and public players
was the selection of the Federal Academy for Science and Engineering
(Acatech) as a Trust Anchor to develop and launch the initiative.
Responsibilities that the Academy partially transferred to the organisation
created for the occasion, the “LLD DRM Datenraum Mobilität GmbH”. 302 Up and
running, the MDS already generates added value for participants. With this safe
and trusted ecosystem for data sharing, participants have access to the data
they need for new digital innovative business cases, which solve problems of
the current through data solutions. The GmbH already reached a sufficient level
of maturity with the implementation of concrete use cases. By breaking data
silos and bringing together private and public players, the MDS already is a best
practice in Europe.

2.3.5 Economic analysis

Despite its enormous promise, the number of real-world use cases of AI in mobility in
the public sector is still relatively few with uses often in localized settings such as
single cities or smaller transport networks. Therefore, large datasets across multiple
locations are not yet available. This makes an economic analysis which compares the
performance of users and non-users of AI in the public using data frames built around
member states impossible. As the use of the technology in cases like transport system
management becomes more widespread, future research could use metrics like
congestion data and safety statistics to test the theory, but with the novelty of the
technology in the public sector, this approach is not currently possible.

Figure 52 Mobility & AI: Member state classification (Malta and Cyprus are categorized as “planners”)

301
APCO Worldwide, 7/12/2020, Has Germany set the European Transport sector on the path to a digital
transformation?, https://apcoworldwide.com/blog/has-germany-set-the-european-transport-sector-on-the-
path-to-a-digital-transformation/
302
The Mobility Data Space, initiative’s website, https://mobility-dataspace.eu/

102
 Adopt AI Study

2.4 Health
In a 2020 survey, 18 European Member States designated healthcare as the sector
which should be prioritized the most going forward in terms of accelerating AI
uptake303. The speed and severity of the COVID-19 pandemic have undoubtedly
expedited technological innovation in a sector that has been searching for solutions to
structural challenges such as ageing populations. Few subjects have received more
attention and expectation as potential solutions for both the short- and long-term
issues facing the health sector than Artificial Intelligence (AI). From Natural Language
Processing to image analysis and predictive analytics, AI is used in multiple areas of
healthcare including care management, diagnosis, medical diagnostics, clinical
decision support and many more304. Beyond patient-centric delivery, AI is also being
deployed to revolutionize health system management with secure patient data at its
heart.
Yet despite its promise, reticence at the user level, a lack of skills, governmental
coordination and incentive models305 have meant that healthcare is the third most
advanced sector in terms of businesses using AI behind ICT and education. This has
been particularly acute at the European level where a cloistering of data and failure to
make use of its significant advantages306 has meant that the EU is yet to fulfil its
considerable potential as a high-skilled powerhouse of AI innovation in health.
The health sector incorporates multiple subsectors and units of analysis, each with its
dynamics, considerations and relationship with AI. From a patient at home receiving
advice from a chatbot to a hospital overhauling its administrative system and a
national government´s drawing up its regulations, AI has a multidimensional role to
play in health. According to a joint EITHealth and McKinsey report, in the short term,
AI will increasingly take over operational and administrative tasks while in the medium
and long term, AI will facilitate a rebalancing of care between hospitals and homes,
easing the burden on health systems, and becoming an increasingly integral part of
clinical trials and by extension clinical practice. Commercially, the healthcare sector in
the EU has seen Venture Capital AI investment of around 5 billion EUR annually,
reaching 13% of global Venture Capital (VC) investments around the world behind the
US and China307. While this share looks to be increasing, there is still a large
disconnect between the highly touted goals of national AI strategies and the everyday
use of AI technologies in health. A 2020 Survey found that only 44% of respondents
working in enterprises in the health sector are using AI technology in any form with a
50-50 split between those that are planning to use AI in the future and those that are
not308.
The policy landscape: Health and AI
A clear signal of intent from the raft of national strategies on AI that were released at
the start of the 2020s was the importance that most member states placed on
healthcare as one of a few critical areas for AI growth 309.

303
Misuraca, G., and van Noordt, C., (2020) Overview of the use and impact of AI in public services in the
EU, Publications Office of the European Union, Luxembourg
304
EIT Health and McKinsey, 2020. ´Transforming healthcare with AI´: https://eithealth.eu/wp-
content/uploads/2020/03/EIT-Health-and-McKinsey_Transforming-Healthcare-with-AI.pdf
305
PWC Netherlands (2017) Adoption of artificial intelligence in healthcare
306
PWC Netherlands (2017) Adoption of artificial intelligence in healthcare
307
AI Watch (2021) ´How can Europe become a global leader in AI in health´:
https://knowledge4policy.ec.europa.eu/sites/default/files/JRC123420_AI_health_Policy%20Brief_FINAL_0.p
df
308
European Commission, Directorate-General for Communications Networks, Content and Technology
(2020) European enterprise survey on the use of technologies based on artificial intelligence : executive
summary, Publications Office, https://data.europa.eu/doi/10.2759/40940
309
AI Watch (2020) National strategies on Artificial Intelligence: A European perspective, 2022 edition, ,
EUR 31083 EN Publications Office of the European Union, Luxembourg,2022, ISBN 978-92-76-52910- 1,
doi:10.2760/385851, JRC129123.

103
 Adopt AI Study

As a part of and in addition to the strategies and plans published throughout the EU,
the strategic procurement of AI technologies and investment in the organization that
produce them in the health sector is increasingly common. This has never been more
evident than with the success of BioNTech, the company behind the world´s first
vaccine against COVID-19, which benefited from several EU R&D programmes310. The
BioNTech vaccine was evidence of how collaboration between the public and private
sectors can create extraordinary health benefits as well as commercial success. One
of these support mechanisms that assisted BioNTech´s work was a European Research
Council Grant311 in 2018, a funding model that some member states have
subsequently applied at a national level.
Figure 53 Health and AI – breakdown by policy type

Source: Authors’ elaboration.

Similar programmes include France´s “Breakthrough innovation challenge on AI in

health”312 Germany´s “Research on AI technologies in Health” 313 and Lithuania´s
National Research Programme for Healthy Aging.
Alternatively, Denmark and the Netherlands have used their Innovation Fund and
National Growth Fund as investment vehicles to assist areas such as regenerative
medicine and health data infrastructure. However, the public sector is also developing
its initiatives such as Portugal´s CCM-SNS Verification of medical prescriptions,
Sweden´s Analytic Imaging Diagnostic Arena and the Datakalab in France which
permitted COVID-19 mask-wearing detection in some French cities (France). On a
regional level, bodies such as Belgium´s Brussels and Flanders have launched their
boutique AI strategies (with Health as a focal point) albeit with an inevitably smaller
scale. The Brussels region´s AI policy for example, through the regional innovation
agency Innoviris, is emblematic of the different forms that public procurement and
financing of AI technology. As part of a series of funding mechanisms, the organization
has financially supported two start-ups focused on the “from the therapeutic medicine
to preventative medicine: Prediction, Prevention, Identification” involved in genomics

310
https://researchprofessionalnews.com/rr-news-europe-horizon-2020-2021-1-biontech-chief-eu-r-d-
funds-helped-develop-covid-19-vaccine/
311
Research Professional News, BioNTech chief : EU R&D funds helped develop Covid-19 vaccine,
2021:https://researchprofessionalnews.com/rr-news-europe-horizon-2020-2021-1-biontech-chief-eu-r-d-
funds-helped-develop-covid-19-vaccine/
Hoppen (2022) “The digital health acceleration strategy explained” https://www.hoppen.care/strategie-
312

dacceleration-sante-numerique/
313
German Research Centre for AI - https://www.dfki.de/en/web

104
 Adopt AI Study

and radioactive medicine314. In addition to these projects, which received 500,000

euros each, Innoviris is also responsible for HubBrussels, a MedTech accelerator which
provides start-ups in the sector with support and access to expertise. Nevertheless,
these programmes are certainly outliers rather than indicative of similar programmes
throughout Europe. While adoption may be sluggish thanks to the unique
idiosyncrasies of the health sector, these programmes at the European, national and
local levels might be a blueprint to be replicated by other (sub)national governments.
Figure 54 Health and AI policies – breakdown by Member State

Health & AI Policies: Breakdown by Member State

6
5
4
3
2
1
0
Finland
France

Italy
Austria

Cyprus

Latvia

Netherlands

Romania
Greece

Malta

Spain
Ireland

Slovakia
Estonia

Luxembourg
Lithuania

Sweden
Denmark

Hungary
Bulgaria
Croatia

Germany

Poland

Slovenia
Belgium

Portugal
Czechia

Regulation Facilitator Funder User

Source: Authors’ elaboration.

2.4.1 Challenges and AI solutions/applications in the health sector

Healthcare systems around the world are facing the structural challenges of
demographic changes as populations and public spending on healthcare balloon along
with higher expectations of the quality of healthcare. In the short, medium and long
term, AI is well placed to address the daunting challenges of the health sector from
resolving the low-hanging fruit of improving operations to boosting R+I and even
shifting healthcare to a more continuous and proactive model rather than one that is
based on mitigation and ex-post treatment.
Figure 55 Summary of challenges and AI solutions for health

Public Improve medical Make Improve Mitigate

sector decision-making administrative understanding surges in
challenge work and deeper medical admissions,
procurement science staff
more efficient shortages

AI value Computational Improving Strengthening Improving

drivers assistance for operations innovation population-
medical decision- health
making management

314
https://innoviris.brussels/brussels-city-innovators

105
 Adopt AI Study

AI AI-assisted AI-assisted Disease state and Prediction of

Applications diagnostics, triage procurement target staff
and diagnosis, understanding, shortages
AI-software for
clinical decision
back-office work lead selection and Identification
support
optimisation, clinical of risk of
dose and endpoint unplanned
selection, therapeutic hospital
tailoring and portfolio admissions
management
Drug Discovery

AI SaMDs NLP and machine Machine learning to AI-powered

Techniques learning help design protocols models
Dual-layered
based on
neural networks
clinical data
from a large
population

Improved medical decision-making

The more recent interest surrounding the topic of Artificial Intelligence (AI) in
healthcare delivery hides the fact that computational systems to support medical
decision-making are not new. Computational technologies already existed using
probabilistic models to rationalise decision-making (e.g., QMR-DT) such as a system to
support the diagnosis of abdominal pain was already created and tested in Leeds in
the 1960s. However, applications are harnessing a new generation of tools and
solutions using larger data sets, web-based technologies (e.g., ADM or Dxplain
models) and recent advances in machine learning (ML) and AI. 315 316
The potential held by new systems is both promising and varied. This includes imagery
recognition for faster diagnoses and the development of IBM Watson for co-diagnostic
or automatic robots for surgical procedures. Taken together, these innovations could
change virtually all stages of the delivery of healthcare to patients. It would help
health practitioners to have a more comprehensive approach to diagnosis, facilitate
the presentation of options to patients and make more accurate drug-prescription —
especially for rare diseases. 317
A joint EIT-Health and McKinsey report identified 6 patient-facing impact areas in
which AI is set to transform healthcare:
• Self-care/prevention/wellness
• Triage and diagnosis
• Diagnostics
• Clinical decision support
• Care delivery
• Chronic care management

315
Souhir Ben Souissi, 15/01/2018, Vers une nouvelle génération d’outils d’aide à la décision s’appliquant à
la prévention des risques lors de la prescription des antibiotiques : combinaison des technologies web
sémantique et de l’aide multicritère à la décision, https://tel.archives-ouvertes.fr/tel-01684761/document
316
Cairn, February 2001, Les systems d’aide à la décision médicale, https://www.cairn.info/revue-les-
cahiers-du-numerique-2001-2-page-125.htm
317
PwC, june 2017, Why AI and robotics will define new health,
https://www.pwc.com/gx/en/industries/healthcare/publications/ai-robotics-new-health/ai-robotics-new-
health.pdf

106
 Adopt AI Study

Use case: Medical Decision-making318

Derm.AI Researchers from Fraunhofer Portugal AICOS (FhP-AICOS) have

developed a digital solution to make the process of referencing
skin lesions more efficient. The technology that results from
the Derm.AI project consists of two components: a mobile
application and Artificial Intelligence (AI) software.
“In recent years, clinicians at the National Health Service (SNS)
have been increasingly concerned with spotting skin cancer early.
However, in Portugal, with the reduced number of dermatology
specialists in the SNS, it is important to improve the referral
process, ensuring that the information collected is of good quality
and helping to prioritize the referred cases”, adds the researcher.

Improving operations
The ageing of the European population translates to a larger need for medical care
while keeping operations high quality and costs down.319 Even though replacing a
doctor or healthcare professional with a health bot is neither feasible nor desirable,
innovative solutions have started to emerge such as AI-powered virtual nurse
assistants (other dimensions related to support to decision-making are detailed in the
next section) to increase the productivity of the existing workforce. 320
Similarly, the digitalisation of back-office activities (e.g., administration) in health can
produce important financial and productivity gains. As with other sectors, automation
of administrative tasks would reduce the administrative burden for the workforce and
free up valuable resources for a budgetarily stressed healthcare system. 321
However, the possibilities opened by AI in healthcare are still not fully clear and are
often met with resistance from the ecosystem. Indeed, the analyses of the gain from
AI in healthcare typically set aside the implementation cost, while implementation of
AI systems is costly and work-intensive. Furthermore, these gains must be tempered
as the acquisition of AI systems needs to be assessed on a case per case basis, taking
into consideration the running costs and the benefits from other technologies. 322

Use Case: Health Operations323

Qventus Qventus is an AI-based software platform that solves

operational challenges that occur in the hospital. Delays or
cancellations of surgeries plague hospitals and can result in
worse clinical outcomes, ineffective use of healthcare
resources (e.g., theatres, anaesthetist time) and higher costs
per patient
The software detects unexpected orders and late-start risks
during the preoperative phase and optimises the block

318
https://www.aicos.fraunhofer.pt/en/news/archive/2022/dermai.html
319
EC, sin dato, Public Health, https://ec.europa.eu/health/health-workforce/overview_en
320
Harvard Business Review, May 2018. Promising AI applications in healthcare,
https://hbr.org/2018/05/10-promising-ai-applications-in-health-care
321
Harvard Business Review, May 2018. Promising AI applications in healthcare,
https://hbr.org/2018/05/10-promising-ai-applications-in-health-care
322
Wolf Justus et al (2022) The economic impact of artificial intelligence in Health Care: systematic review,
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7059082/
323
Transforming healthcare with AI The impact on the workforce and organisations

107
 Adopt AI Study

schedules in real-time. It helps hospital teams prioritise, for

example, by identifying high-priority actions and nudging the
teams to resolve issues.

Artificial Intelligence can also benefit the public procurement process itself. Public
procurement in healthcare can be a difficult and complex task that could be simplified
by an AI system with the automation of repetitive tasks, enabling better-decision
making, avoiding tender duplication, lowering costs and increasing transparency. 324
This is especially the case in health with important transaction costs that can arise
when the acquisition is surrounded by uncertainties about the product’s qualities,
asset specificity and frequency. Similarly, the acquisition of pharmaceuticals and
health goods requires checking safety and compliance.325
This complexity leads to a series of concrete and damaging consequences for the
health sector. For example, a group of UK hospital trusts collected and standardised
manufacturers and price data for generic products (e.g., exam gloves) usually bought
by individual trusts. The study identified that more transparent and centralised
procurement would represent cost-cutting opportunities ranging between 15 and 50%
compared to the current best prices paid by the National Health System. In addition,
the need to relax checks and procedures to meet the emergency caused by the COVID
outbreak led to an increase in the circulation of suboptimal drugs and even corruption
scandals. The inflexible nature of the safety rules did not allow for rapid adaptation
and transparent procedures that were needed in a time of emergency. 326
According to Pettersen Inger et al, automation of public procurement implies moving
to a new paradigm of “transactional contracts” (contracts being summarised as a
mutually profitable exchange). However, the authors warn that the approach is not
necessarily optimal in the context of healthcare where “relational exchanges” grant
more important to human and social controls — two important dimensions to control
public procurement in healthcare.327
More concretely, relational contracts are driven by efficiency criteria such as scientific,
costs, price, etc. Relational exchanges allow us to consider ethical and intersubjective
elements. In other words, transactional contracts using automated solutions can
generate important gains in efficiency-driven fields and areas. In other words, it is
unwise to generalise the use of an automated public procurement system. However,
these warnings coined by Petterson Inger et al. are based on one case study. Mapping
efforts taking into consideration qualitative and human factors are needed to confirm
these intuitions as well as to identify areas where transactional and relational logic is
the most suited.
Human and social controls based on “relational exchanges” are important security
factors in the acquisition of sensitive healthcare goods (pharmaceuticals, etc). The
number of activities that can be automated in public procurement of healthcare will
therefore be more limited.

324
EC, sin dato, Emerging technologies in public procurement, https://ec.europa.eu/growth/single-
market/public-procurement/digital-procurement/emerging-technologies-public-procurement_fr
325
EC, 2020, Expert panel on effective ways of investing in health (EXPH),
https://op.europa.eu/en/publication-detail/-/publication/0fa5efff-b138-11eb-8307-01aa75ed71a1/language-
en
326
Kohler Clare Jillian, September 2020, The urgent need for transparent and accountable procurement of
medicine and medical supplies in times of COVID 19 pandemic,
https://joppp.biomedcentral.com/articles/10.1186/s40545-020-00256-w
327
Pettersen Inger et al (2020) Public procurement performance and the challenge of service complexity,
https://www.researchgate.net/publication/342649253_Public_procurement_performance_and_the_challeng
e_of_service_complexity_-_the_case_of_pre-hospital_healthcare

108
 Adopt AI Study

Advanced and enhanced public procurement would greatly benefit from the
development of an inclusive health data ecosystem across the health value chain. This
involves the modernisation and connectivity of hospitals’ digital platforms and the
development of health data spaces.
Strengthening innovation
As explained in further detail in the subsequent section on the health AI value chain,
AI is being taken up by large pharmaceutical companies in the area of R+D 328. The
promise of AI stems from its ability to “both collect high-quality data from each
patient and connect it to data from large pools of patients for analysis with artificial
intelligence-based algorithms”329. Furthermore, large big pharma players such as
AstraZeneca and Bristol-Myers Squibb are entering into partnerships with AI
companies with objectives such as acceleration of drug discovery and designing
protocols for precision treatment. Other early applications include:
• Disease state and target understanding
• Lead selection and optimization
• Clinical dose and endpoint selection
• Therapeutic tailoring
• Portfolio management
However, per the CEO of Novartis Vas Narasimhan, “a lot of talks and very little in
terms of actual delivery of impact” 330 and the bulk of recent investment had focused
on data-related projects that will provide the foundation for future R+D and
applications.

Use Case: Strengthening Innovation

Astra Zeneca AstraZeneca teamed up with BenevolentAI for

Data and AI: collaboration on the drug discovery of two novel targets
Drug to enter its drug portfolio, for chronic kidney disease
Discovery (CKD) and idiopathic pulmonary fibrosis (IPF) 331.
BenevolentAI and AstraZeneca scientists collaborate
using the Platform and Knowledge Graph to interrogate
underlying disease mechanisms, frame and test
hypotheses and rapidly identify novel targets. New
knowledge generated through the disease programmes
is also fed back into the Platform, improving the quality
of future drug target predictions.

Improving population-health management

Beyond the frontline of healthcare delivery, AI can improve the overall health of
populations over time. With access to real-time data from sufficiently large
populations, AI models can be deployed for early detection of trends. By extension, AI
can be a tool for early prevention and give systems early warning systems valuable
time to distribute resources in the case of scenarios such as a surge in hospital

328
EIT Health & McKinsey (2020) Transforming healthcare with AI
329
EIT Health & McKinsey (2020) Transforming healthcare with AI
330
EIT Health & McKinsey (2020) Transforming healthcare with AI
331
BenevoltentAI (2022) “BenevolentAI achieves further Milestones in AI-enabled target identification with
AstraSeneca”: https://www.benevolent.com/news/benevolentai-achieves-further-milestones-in-ai-enabled-
target-identification-collaboration-with-astrazeneca

109
 Adopt AI Study

admissions, staff shortages or epidemiological events. These can also better inform
understanding of longer-term events such as demographic patterns and immunity.

Use Case: Population-health management332

Risk prediction Mount Sinai’s Department of Population Health has

for hospital been using machine-learning algorithms to mine data
emergency that identifies patients who are at risk of an unplanned
admissions admission among the system’s 500,000 patient
population health programme and develop predictive
modelling features333.
This shifts the working patterns of practitioners from
reactive care to proactive care. To adequately address
the risk, practitioners and social workers need to
understand how the model identified the patient and
which factors may need to be addressed to mitigate the
risk

2.4.2 Value chain analysis

The Health Value Chain can be defined as “the chain of suppliers, producers,
distributors, healthcare providers and dispensers that provide clinicians with the drugs
and supplies they need to care for patients and patients.”334
Highly complex, the health value chain includes external players which can quickly
become influential through the provision of innovative technologies. This is the case of
Apple’s wearables that serve as tools to measure health signals used by insurance
companies to incentivise their customers to remain healthy through a system of
rewards.335 Furthermore, each link that composes the health value chain has
undergone important changes over the years. This can be attributed to the increased
use of digital solutions and the growing importance given to achieving cost
efficiency.336
The importance of digital solutions in the modern health care industry is better
understood through the representation of an R&D and a Digital Value chain, plugged
into the traditional value chain. These two “side value chains” are where the potential

332
EIT Health & McKinsey (2020) Transforming healthcare with AI
333
EIT Health & McKinsey (2020) Transforming healthcare with AI
334
TraceLink, sin dato, The future of the healthcare value chain, https://www.tracelink.com/agile-supply-
chain/healthcare-value-
chain#:~:text=What%20is%20a%20healthcare%20value,care%20for%20patients%20and%20patients.
335
Another example is a joint venture between Amazon, J.P Morgan Chase and Berkshire Hathaway, which
established an independent health care company, Haven, for their more than one million employees. The
objective of the establishment of the company was to improve health care services and cost efficiency for
the employees. Another example is the Amazon-introduced line of private label over the counter medicines
and selling point for medical supplies to doctors, dentists and health institutions. Deloitte (2019), 2019
Global health care outlook: Shaping the future, https://www2.deloitte.com/global/en/pages/life-sciences-
and-healthcare/articles/global-health-care-sector-outlook.html.
336
Sarah Collins (2015), Analyzing value cgain business models: medical devices industry,
https://marketrealist.com/2015/11/analyzing-value-chain-business-models-medical-device-industry/

110
 Adopt AI Study

for Artificial Intelligence arises, such as automated robotics for lab experiments,
virtual twinning or digital experiments. 337 338 339
Another key element of the European Health Value Chain is the central role played by
the State. For example, in 2019 the government schemes and compulsory schemes
and saving accounts represented 79.7% of all health expenditures in Europe. 340 This
importance of public authorities has a decisive influence on both the competitive
dynamics and the role that public procurement can play in influencing the AI market in
healthcare. 341
The value chain links will be analysed together to identify synergies and centre the
analysis on potential state intervention and potential for AI solutions. An additional
sub-section on the impact of Covid will be added as the pandemic had a decisive
influence on the data and digital value chain of the health sector.
Figure 56 Health sector value chain

Source: Four elements of the digital value chain were taken from Bain & Company342 and further elaborated by CARSA,
in 2019. The T&L function will not be analysed in this value chain section. It is an important enabling function but it is not
central to the delivery of health value for this study.

The value chain analysis has four sub-sections:

 R&D and production
 Purchasing and intermediaries

337
Healthcare Industry BW (2016), Industry 4.0 in the medical technology and pharmaceutical industry
sectors, https://www.gesundheitsindustrie-bw.de/en/article/dossier/industry-40-in-the-medical-technology-
and-pharmaceutical-industry-sectors
338
Rhenu Bhuller (2018), Beyond the pill pharma takes a stake in digital health,
https://pharmaboardroom.com/articles/beyond-the-pill-pharma-takes-a-stake-in-digital-health/
339
Bain & Company (2015), Getting the Dose Right: A Digital Prescription for the Pharma Industry,
https://www.bain.com/insights/a-digital-prescription-for-the-pharma-industry/
340
Eurostat, sin dato, Analysis of current healthcare expenditures, https://ec.europa.eu/eurostat/statistics-
explained/index.php?title=File:Analysis_of_current_healthcare_expenditure,_2019_(%25).png.
Comparatively, in 2011, public authorities, government and mandatory schemes represented 48% in the US
and 56% in China EMERGO, 2021, Europe – Overview of Medical Device industry and healthcare statistics,
https://www.emergobyul.com/resources/europe-overview-medical-device-industry-and-healthcare-statistics
341
European Observatory on Health Systems and Policies Series, 2005, Purchasing to improve health
systems performance, https://www.euro.who.int/__data/assets/pdf_file/0004/98428/E86300.pdf
342
Ibid

111
 Adopt AI Study

 Health providers and end-users

 COVID impact

Research & Development and Production

R&D and production stages are closely intertwined as these tasks are usually
undertaken by the same group of companies. In turn, these stages can be segmented
into a series of 3 key players: Medtech, pharmaceuticals and public research bodies
(public research bodies will be analysed concerning their contribution to MedTech and
pharmaceuticals).
“MedTech, or medical technology, is every product, service, or solution using medical
technology to improve people’s health by preventing, diagnosing, monitoring, and
treating disease.”343 An important link in the modern health value chain, Medtech
includes a range of different players, from SMEs to tech giants and research centres.
MedTech is an innovation-driven field with strategic and economic importance for the
EU. The EU MedTech market represents around €140 billion, 760,000 employees and
27.6% of the world market (second after the USA). Furthermore, the sector
represents an economic surplus of around €8.7 billion for the EU and 344 95% of
European MedTech companies are SMEs.345
In addition to this economic importance, the European MedTech market is a dynamic
and innovative field. In 2020 only, 14,200 patent applications were filled at the
European Patent Office (EPO) compared to 8,500 applications filled for the
pharmaceutical sector. These figures mean that Medtech represents 38% of all patent
applications at the EPO.346
With a share of 70% of all MedTech purchases347 in Europe, public procurement can
play a key role in supporting the uptake of AI-driven MedTech. So far, the uptake of
AI-driven MedTechs and solutions is concentrated in large hospitals and large cities.
This represents an unexploited potential for larger adoption through the harmonisation
of capacities.348In many cases, the lack of utility of the AI solutions comes from a lack
of complementary solutions to enable the use/uptake of AI MedTechs (e.g., digital
medical records require new integration software and platforms). 349Risk-capital from

343
The scope of MedTech corresponds to the scope defiend by the Medical Defice Directive. See Obelis
group, sin dato, Scope under the Medical Device Directive, https://www.obelis.net/industries/medical-
device-directive-scope-under-the-medical-device-directive/ and AKRNconsulting, 01/06/2020, What is
MedTech, https://akrnconsulting.com/what-is-
medtech/#:~:text=MedTech%2C%20or%20medical%20technology%2C%20is,%2C%20monitoring%2C%2
0and%20treating%20disease.
344
MedTechEurope, 21/06/2021, Facts and figures, https://www.medtecheurope.org/about-the-
industry/facts-figures/
345
This further stresses the importance taken by R&D in medtech: “The average global R&D investment rate
(R&D spend as a percentage of sales) is estimated to be around 8% in the medical technology sector.
Products typically have a lifecycle of only 18-24 months before an improved product becomes available.”
MedTech expenditure in Europe represented €265 per capita in 2020. MedTechEurope, 2021, The European
Medical Technology Industry in Figures, https://www.medtecheurope.org/wp-
content/uploads/2021/06/medtech-europe-facts-and-figures-2021.pdf
346
MedTechEurope, 2021, The European Medical Technology Industry in Figures,
https://www.medtecheurope.org/wp-content/uploads/2021/06/medtech-europe-facts-and-figures-2021.pdf
MedtechEurope, sin dato, Value-based procurement, https://www.medtecheurope.org/access-to-
347

medical-technology/value-based-procurement/
348
Brooking institute (2022) Why is adoption in Health Care lagging?
https://www.brookings.edu/research/why-is-ai-adoption-in-health-care-lagging/
349
Brooking institute (2022) Why is adoption in Health Care lagging?
https://www.brookings.edu/research/why-is-ai-adoption-in-health-care-lagging/

112
 Adopt AI Study

public bodies is needed to target the development of complementary solutions in an

emerging market.350
Furthermore, pharmaceutical companies are diverse and range from SMEs (they
account for 65% of all European pharmaceutical companies351 352) to chemical giants
synthesising molecules or academics contributing to research & development. In total,
in 2019, the European pharmaceutical sector represented a production of €275 billion
a positive trade balance of €140,000 million and a total workforce of 795,000
employees.353 The European pharmaceutical market represents around 23.9% of the
world market with an expected CAGR growth of 4.5% over the 2021/2025 period. 354
While the range of applications for AI solutions in pharmaceuticals is very wide, its
most potent and important impact is on speeding up the drug discovery process. 355 To
contribute to the greater uptake of AI solutions and generate economic gains in drug-
discovery, better integration, new standards for molecule representation and solutions
for data continuity in cognitive factories are needed.356
Through its system of tax incentives, a network of research bodies, and subsidies, the
State already represents a crucial source of funding in Pharmaceuticals, paving the
way for the reallocation of these fundings to support the development of AI
solutions.357
AI solutions require standardised molecule representation for data exchange (within
the company or across value chains). This is often prevented by the lack of
computable and standardised molecule representation thus preventing the use of new
AI solutions for digital twinning, digital trials, etc. This barrier requires the
development of large public-private partnerships bringing together pharmaceuticals,
academics and chemical companies.358

350
JRC (2021) How can Europe become a global leader in AI in health?,
https://knowledge4policy.ec.europa.eu/sites/default/files/JRC123420_AI_health_Policy%20Brief_FINAL_0.
pdf
351
Based on Eurostat figures. SMEs are companies under 50 employees and large companies are above 50.
Figures for number of companies were available for 2018. However, the total number of pharmaceutical
manufacturers was only available for 2016. The final rate is therefore only an approximation based on
existing data. Eurostat, sin dato, Industry by employement size,
https://appsso.eurostat.ec.europa.eu/nui/submitViewTableAction.do
352
SMEs are extremely important for pharmaceuticals as 42% of the medicines developed by SMEs address
orphan disease with a success rate of medicine development between 75% and 68% for all drug developed
by SMEs. European Medicine Agency, 11/05/2016, Supporting Innovative SMEs as major drivers of new
pharmaceutical developments, https://www.ema.europa.eu/en/news/supporting-innovative-smes-major-
drivers-new-pharmaceutical-developments
353
EFPIA (2020) The Pharmaceutical Industry in figures,
https://www.efpia.eu/media/554521/efpia_pharmafigures_2020_web.pdf
354
Statista (2021) Pharmaceutical industry in Europe – Statistics & Facts,
https://www.statista.com/topics/8631/pharmaceutical-industry-in-europe/#dossierKeyfigures
355
National Library of medicine (2021) Artificial Intelligence in Drug Discovery,
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7577280/ and Biospace, 14/03/2022, AI in
Pharmaceuticals, https://www.biospace.com/article/ai-in-the-pharmaceutical-market-size-to-worth-around-
usd-9-24-bn-by-
2030/#:~:text=The%20AI%20in%20pharma%20market%20segmented%20by%20based%20on%20applic
ation,is%20accounted%20highest%20market%20share.
356
These points are detailed in CARSA et al. (2021), Study on technological and economic analysis of
industry agreements in current and future value chains, https://op.europa.eu/en/publication-detail/-
/publication/8c021023-53ee-11ec-91ac-01aa75ed71a1/language-en
357
European Parliament (2021) European Pharmaceutical Research and development,
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/697197/EPRS_STU(2021)697197_EN.pdf
358
These points are more detailed in CARSA et all, 02/12/2021, Ibid. The report provides a framework
based on 3 industry agreements to efficiently tackle these challenges.

113
 Adopt AI Study

The uptake of AI solutions in pharmaceuticals is often impeded by the lack of data

quality, quantity and interoperability. News standards are required to harmonise
systems and foster digital continuity in cognitive factories (e.g., integration of
production systems with R&D function for automated return on experience to speed-
up drug-discovery process).359
Purchasers and intermediaries
The number and type of organisation (e.g., local government, central health fund)
involved in the intermediation and purchasing of health supplies varies for each
European country. However, the European market is characterised by the domination
of the purchasing and intermediation stages by public players. 360
According to the European observatory on health systems, competition is absent at
these stages of the value chain. Even though private actors exist, they are surrounded
by legislative frameworks and rules to monitor and frame their activities.
Consequently, private bodies simply do not enter into direct competition with public
authorities that dominate and organise intermediation and purchasing activities. 361
Consequently, the purchase of health supplies is mainly done by public authorities. In
Europe, public procurement represents 70% of all MedTech purchases. 362 The payment
of pharmaceuticals by compulsory health insurance systems and national health
services (ambulatory care only) represented a total of €135,485 million in 2018.363 If
we divide this by the total production figures for the pharmaceutical sector, this means
that the State represents around 49% of all pharmaceutical payments in Europe364.
The domination of public bodies in intermediations stages means that uptake and
adoption of AI solutions will only be achieved if the state plays the leading role in
investing, purchasing and implementing innovative solutions. The state can also play
the role of an enabler in the development of new solutions. This requires the
development of horizontally integrated platforms and data spaces for the exchange of
standardised information.
Providers and end-users
As for other links in the health value chain, public authorities again play a central role
in the provision and usage stages. The state is the main source of funding for patients
regarding medical services and goods: “massive predominance of public funding in
inpatient curative care: even if part of the total health expenditure is always funded by
private insurance and out-of-pocket payments, almost the entire amount of inpatient
health expenditure is publicly financed.”365 Similarly, the majority of hospitals and
hospital beds are managed by public organisations while private bodies have to

359
These points are more detailed in CARSA et al. (2021) Ibid. The report provides a framework based on 3
industry agreements to efficiently tackle these challenges.
360
European Observatory on Health Systems and Policies (2005) Purchasing to improve health systems
performance, https://www.euro.who.int/__data/assets/pdf_file/0004/98428/E86300.pdf
361
For example, in Germany, private health insurance schemes have to adopt a non-for-profit
status.European Obervatory on Health Systems and Policies (2005) Purchasing to improve health systems
performance, https://www.euro.who.int/__data/assets/pdf_file/0004/98428/E86300.pdf
MedtechEurope, sin dato, Value-based procurement, https://www.medtecheurope.org/access-to-
362

medical-technology/value-based-procurement/
363
EFPIA, 2020, The Pharmaceutical Industry in figures,
https://www.efpia.eu/media/554521/efpia_pharmafigures_2020_web.pdf
364
In Europe, in 2020, medtechs represented 7.6% of all health expenditures, pharmaceuticals 14.9% and
inpatient/outpatient care accounted for 77.6%. MedTechEurope, 2021, The European Medical Technology
Industry in Figures, https://www.medtecheurope.org/wp-content/uploads/2021/06/medtech-europe-facts-
and-figures-2021.pdf
365
European Hospital and Healthcare Federation (2018) Hospitals in Europea Healthcare data,
https://www.hope.be/wp-content/uploads/2018/07/2018_Hospitals-in-EU-28-Synthesis-final-for-
publication-002.pdf

114
 Adopt AI Study

respect painstakingly defined rules and regulations.366 As for other segments of the
value chain, this gives the state and public authorities a driving role in the
development of the IT architecture and infrastructure to enable the development of
new AI solutions. The development of a health data space would indeed have to start
at the service delivery and usage stages. Artificial intelligence is as good as the data it
is fed with. Consequently, the development of AI solutions for health requires
increasing the quantity, quality, interoperability and availability of health data. 367 This
involves generalising the use of standardised Electronic Digital Health Records (for
patient data). This stage represents great opportunities and potential for the
development of new AI solutions at almost all stages of the value chain. IT platforms
to manage public hospitals are another precious source of health data for
procurement, use of materials, needs, etc. The heterogeneity of the platform
landscape is partially addressed by initiatives from public hospitals to develop common
solutions. However, more efforts are needed (especially to develop truly European
coordination and standardisation).
The impact of COVID: data spaces, cross-border exchanges and new
opportunities for AI
COVID highlighted the vulnerabilities of the European health sector caused by
globalisation. Indeed, with the closure of borders, many key supplies were in critically
short supply for both basic needs (masks) and even the production of pharmaceutics
and MedTech. 368However, despite its tragic costs in terms of human life, the COVID
outbreak also accelerated a series of positive trends in the digital space: cross-border
exchange of information, tracing apps, vaccine development and the development of
health data spaces.
In April 2020, the European Commission launched the Covid-19 Data Sharing Platform
for rapid cross-borders data exchanges related to COVID-19. The initiative could build
on prior initiatives that developed the groundwork for rapid deployment of IT
architectures and standardisation of data. 369 In other words, the crisis contributed to
accelerating the deployment of the data and IT architecture necessary for the
development and uptake of new AI solutions. However, more efforts are necessary to
build on this legacy. Standardisation and interoperability of data sets and platforms
remain important limitations for cross-border exchanges. Furthermore, the suspension
of the French Health Data Hub initiative in January 2022, showed that ethical and legal
concerns are still a central obstacle to the development of sustainable health data
spaces in Europe. 370 To turn these temporary gains into long-terms infrastructure,
more involvement is necessary from all stakeholders which are part of the value chain,
mainly from the end-user side. A governance structure designed by the stakeholder’s
ecosystem is necessary to turn data spaces into successful and sustainable initiatives
leading to the development of new AI solutions.

366
WHO, sin dato, Hospital beds, https://gateway.euro.who.int/en/indicators/hlthres_28-beds-in-publicly-
owned-hospitals-total/ For just France in 2019, 45% of all hospitals were public institution and 22% were
not-for-profit private organisations. Gouvernement français, sin dato, Vue d’ensenble,
https://drees.solidarites-sante.gouv.fr/sites/default/files/2021-07/Vue%20d%27ensemble.pdf
367
These points are more detailed in CARSA et al (2021) Ibid.
368
MedTechEurope (2021) The European Medical Technology Industry in Figures,
https://www.medtecheurope.org/wp-content/uploads/2021/06/medtech-europe-facts-and-figures-2021.pdf
369
EC, April 2020, Coronavirus: Commission launches data sharing platform for researchers,
https://ec.europa.eu/commission/presscorner/detail/en/IP_20_680
370
L’usine Digitale (2021) Où en est le Health Data Hub ?,

115
 Adopt AI Study

Figure 57 Summary of key intervention areas for public authorities and public procurement along the health value chain.

Source: CARSA, 2022.

2.4.3 Main drivers and barriers

The table below presents a PESTEL analysis of the uptake of AI technologies in public
procurement in the e-government sector.

116
 Adopt AI Study

Table 6 PESTEL analysis of the uptake of AI technologies in public procurement in the Healthcare sector

Criteria Drivers Barriers

 The successful management of healthcare  The adoption of AI systems and technologies by

systems is a key political issue. With more the public healthcare system will need to open
public policies promoting the use and adoption tendering processes which might extend the
of AI in healthcare, uptake is increasing. adoption time significantly.
Political
 Due to its importance, strategic public  Institutional reticence to reform healthcare
procurement can drive demand for systems due to its difficulty and cost.
pharmaceuticals and healthcare technologies for
the benefit of the citizen.

 By examining data patterns, AI technologies  It is crucial to determine the best place/process

can help healthcare organizations make the to implement AI solutions and technologies to
most of their data, assets and resources, decrease, and not increase, the overall cost of
increasing efficiency and improving the healthcare systems.
Economic performance of clinical and operational
workflows, processes, and financial operations.
 The adoption of AI systems requires the support
of external expertise to assess each situation
 AI is accelerating operations to save and propose, produce and implement the most
productivity hours, so AI has the potential to suitable AI systems for each case.
save costs.

 Optimisation of diagnosis and treatments  A misconception among medical practitioners

through AI solutions has the potential to that AI will replace doctors in the coming years.
improve people’s life quality. AI has a high The doctors and practitioners believe that skills
potential to improve the speed and accuracy of such as empathy and persuasion are human
Social diagnosis and detection. skills, and thus, technologies cannot completely
rule out the presence of a doctor.
 AI helps streamline procedures, automate
functions, instantly share data and organize  Lack of understanding of the capabilities of AI
operations, all of which help relieve medical technologies in the public healthcare sector.
professionals of juggling too many tasks and Responsible AI principles within an organisation

117
 Adopt AI Study

increase the quality of the work environment can guide how to operate and exploit AI
for overstretched healthcare professionals. systems fully, responsibly and ethically using
effective internal communication and training.

 The need to develop more simple technologies  Doubts about the capabilities of AI solutions in
such as data storage and computing power that terms of accurately diagnosing patient
allow data collection, storage, movement and conditions.
transformation that will afterwards support
intelligent decision-making (AI).
 Lack of required talent, as AI adoption in
healthcare, needs staff with interdisciplinary
 The need to develop common and high-quality knowledge both from technological and medical
Health Data Spaces for the centralisation of disciplines. Specific investment in educating and
data and development of AI applications. reskilling healthcare workers is required to truly
take advantage of the implementation of AI
 Improvement in sensitive data protection and solutions.
treatment, as well as cybersecurity aspects.
 Difficulties working with unstructured data,
such as medical imaging, represent a large
Technological share of relevant data in healthcare. This
means that the AI system still needs to be
complemented by the experience of human
doctors.
 Lack of interoperability is a limiting factor to be
considered in terms of AI adoption by the
healthcare sector. With hospitals, specialists
and clinics working with diverse Electronical
Medical Record platforms, it is challenging for
one entity to be able to access a patient’s full
record as systems are not compatible and
interoperable. This limits the information that
AI can see and causes incomplete analysis of
the medical record.

Environmental  Alongside quality workwear clothing for  The energy use and emissions associated with
healthcare workers, sanitary PPE, and limiting training AI models have indicated the wider

118
 Adopt AI Study

human contact, AI can automate processes that consequences of this rapid development. These
may have required human labour. These emissions are largely associated with the
automated processes reduce human action, and upfront development and training of AI
the waste involved with ensuring environments algorithms while the tuning and adaptation are
are clean for healthcare standards. potentially less costly.
 In terms of sustainability, savings can be seen
in energy and time. Lower energy use links
directly to a lower environmental impact from
electricity generation.

 Accelerate the development of policy  No shared official standards in the industry for
frameworks to build trust and foster the how AI can be used by organisations nor how
adoption of AI in healthcare. its performance should be evaluated, leading to
uncertainty on the legitimacy of the use of AI
 Build and maintain a balanced regulatory by hospitals.
environment, based on existing applicable
Legal regulations, that enables and stimulates future  Lack of rules of accountability in the use of AI
technological innovation and evolution. for decision-making. As AI technology replaces
parts of the decision-making process
 Speed up the regulation of AI in the market traditionally carried out by humans, there is no
through a shared official definition in the
regulation on how to include non-human actors
market of what AI technology is.
in the legal accountability system.
Source: Authors’ elaboration based on the challenges and the value chain analysis.

119
 Adopt AI Study

Figure 58 Strengths, Weaknesses, Opportunities and Threats of the uptake and public procurement of AI-technologies in
the healthcare sector

S
Increase workplace productivity - Rather than spending hours of manpower on
menial, repeatable tasks, employees can configure AI to manage it instead. AI allows to
manage a multitude of tasks more efficiently and improving performance of clinical and
operational workflows, processes, and financial operations.
Better quality of healthcare – Adoption of AI in healthcare has the potential to improve
life quality of all citizens making high quality care more accessible and affordable, as well
as improving working conditions of health professionals.
Great applicability range – AI can be used and adopted by a variety of industries,
including healthcare. The type and sophistication of the AI required depends on the tasks
to be performed and the range of potential applications in healthcare is broad, having
examples such as early and accurate diagnosis, development of medicines, Streamlining

W
Patient Experience or Mining and Managing Medical Data.

Artificial intelligence remains inhuman - skills such as empathy and persuasion are
human skills that technologies cannot replicate. Therefore, AI is a tool that can
communicate, but it can’t communicate emotionally. And so, although it can use

O
information, it won’t be able to grasp or react to the complexities of human emotion.

Combing AI with newer forms of tech – The combination of AI with “lower”

technologies provides the opportunity of crating new AI based technologies capable of
responding to numerous necessities alongside the healthcare value chain.
Less strain on employees – A large amount of tasks, such as administrative issues, can
be handled thanks to the adoption of AI technologies, which lessens the workload of
medical professionals. Therefore, the sensitive adoption of AI enables lowering the stress
levels of professionals and improve their quality of life

T
Job loss – Part of the society believes that the adoption of AI will lead to job loss, as
machines will replace humans.
Trustworthy AI -The need for doctors to interact with a machine for critical decision-
making poses an issue of trust, particularly in a field such as healthcare, where face-to-
face interaction between patients and doctors is so important.

Source: Authors’ elaboration based on the challenges, value chain and PESTEL analysis.

2.4.4 Case study: Developing the next generation of bionic prostheses

for low-limb amputees (Belgium)
Key Actors
Innoviris: The Brussels Region´s Innovation Agency
Axiles Bionics: A Spin-off from the Vrije Universiteit Brussel
Type of policy
This use case of AI in health was driven by the Innovative Starters Award (ISA)
programme carried out by Innoviris to provide funding for ambitious start-ups and
scale-ups. In the words of Barbara Trachte, the State Secretary for Economic
Transition and Scientific Research: “The Innovative Starters Awards is a spearhead
programme of Innoviris to spotlight and financially support start-ups and scale-ups to
further develop their innovative ambitions. I am pleased that ecological and social

120
 Adopt AI Study

criteria were used this year. After all, Brussels-based companies have many ideas for
acting positively towards the planet and its inhabitants371. The winner of the 2021
edition, Axiles Bionics, will receive €500,000 in funding to further realise their
strategic innovation plan over the next 3 years.
Goals
The goal is to create a cloud-based data collection platform with information from
every prosthesis in use. Currently, nothing is connecting the human brain with a
prosthesis which makes mobility more difficult. Incorporating another form of
intelligence through technology, the current disconnect between what happens
naturally and how a prosthesis should react to stimuli can be diminished 372.
Description
Axiles Bionics, co-funded by Innoviris, is seeking to develop an “innovative technical
solution that can be applied to any medical device that would benefit from patient-
specific self-learning capabilities, such as prostheses, orthoses and wheelchairs,
among others”. This will be the company´s first application and its first flagship
product, demonstrating the necessity of the initial public investment. In addition to the
biomechanics of prostheses used in everyday mobility, the solution also includes
reinforcement learning whereby the devices will be able to learn autonomously using
the data from the cloud-based platform mentioned above.
Data can be submitted to the cloud every night by recharging and connecting the
prosthesis while the amputee is sleeping. This can then be used to improve the daily
functioning of the prosthesis through user-specific data which are sent back when the
person is awake and active373.
Successes and key lessons
 Competitions with social considerations produce solutions with social
utility. For the 2021 edition of the ISA, a competition that had been running
since 2010, there were social and ecological criteria in its assessment 374. This
was the first time that these had been included and were introduced as part of
the 2021-2027 Regional Innovation Plan. While previous winners of the award
had a significant social impact, the inclusion of more altruistic criteria helps to
create incentive structures and produce solutions akin to Axiles Bionics, a
project with significant social impact that this kind of project has for people with
reduced mobility and the improvement of their daily life. In terms of key
lessons to be replicated, although it seems obvious, explicit criteria in the very
popular funding mechanism of research calls help condition the results.
 Regional clustering with universities and public funding agencies to
maximize potential. Formed from the Vrije Universiteit Brussel and funded by
the Brussels region innovation agency, the story of Axiles Bionics is grounded in
the local AI ecosystem. Using networks, local expertise and support beyond
finance to maximise outputs and progress is essential and a model that other
regional authorities can look to replicate. In the words of Pierre Cherelle (the
CEO), “We are very grateful for the continued trust and support of both
Innoviris, in our team and our vision,". Nevertheless, with the funding totalling
500,000 euros over 3 years, this type of procurement model can be easily

371
Innoviris.brussels (2021) "Axiles Bionics and Tapio win an innovative Starters Award 2021”
https://innoviris.brussels/news/axiles-bionics-and-tapio-win-innovative-starters-award-2021
372
Innoviris.brussels (2021) "Piere Cherelle Axiles bionics” https://innoviris.brussels/stories/pierre-cherelle-
axiles-bionics
373
Ibid
374
Innoviris.brussels (2021) "Axiles Bionics and Tapio win an innovative Starters Award 2021”
https://innoviris.brussels/news/axiles-bionics-and-tapio-win-innovative-starters-award-2021

121
 Adopt AI Study

exported to the entirety of the EU and help foster local AI ecosystems and
clusters.

2.4.5 Economic analysis

A large part of public sector activity in the health sector has been in the area of
research and innovation with the belief that more funding will unlock better uptake of
AI in health. Subsequently, the first model estimates the relationship between
healthcare research and policy modes. As discussed above, another touted benefit of
AI in health is its ability to improve the “low-hanging fruit”375 of the administrative
side of the healthcare system which is reflected in the second model.
Figure 59 Health & AI – Member state classification

N.B Cyprus is categorized as a Facilitator and Malta as a planner

To measure research in AI, model 1 uses the OECD´s AI Research database and takes
the number of publications per member state in 2021 by “healthcare Institutions”376.
The hypothesis would be that those countries already using AI have a more
sophisticated AI ecosystem including research output on AI in health. Similarly, those
countries funding programmes in AI in health should produce more than those not.
The second indicator is taken from the DESI´s eHealth section in the form of “seeking
health information” – a measure of the digitalization of a country’s health system 377.
This selection draws on the theory that AI improves policy delivery, public services
and citizen-government interaction378. In healthcare administration, can “enhance

375
EIT Health & McKinsey (2020) Transforming healthcare with AI
OECD.AI (2022), visualisations powered by JSI using data from MAG, version of 31/12/2021, accessed on
376

27/6/2022, www.oecd.ai
377
Eurostat (2022), Table isoc_bde15cua: Internet use and activities
378
AI Watch (2022) Road to the adoption of Artificial Intelligence by the Public Sector: A Handbook for
Policymakers, Public Administrations and Relevant Stakeholders, EUR 31054 EN, Publications Office of the
European Union, Luxembourg, 2022, ISBN 978-92-76-52131-0, doi:10.2760/693531, JRC129100.

122
 Adopt AI Study

existing service interfaces through applications that are more usable, and thus
decrease barriers to use”. The second model looks to test that theory.
Figure 60 Regression table - Health

AI Research by Seeking Health

Health Institutions Information

User 224.37 70.24

Funder -87.17 -3.05

Planner -204.23 -6.32

R² 0.22 0.07

Observations 27 27

 Research: User countries are correlated with the largest research output by
health institutions with 87 more publications per year on AI than funder
countries and 204 more than planner countries.
 Information: Similarly, of the 3 categories, planners have the lowest
percentage of citizens seeking health information online at around 63%, 3%
less than funder countries and nearly 6.5% than developers.

2.5 Education
In the words of educationalist and academic Anthony Seldon, AI in education is not “a
passing innovation like the introduction of photocopying or smart boards, but a once-
in-five-hundred-year revolution”379. AI is set to transform education for learners,
teachers and the ecosystem as a whole. Broadly speaking, the consensus is that AI
will have a differentiated effect on different areas of education both supporting
existing capabilities and creating cognitive capabilities that would not be possible
without technology and reducing the importance of some human cognitive capabilities
or making them obsolete380. From improving special needs education through the
detection of conditions such as dyslexia to the generation and assessment of tests, the
potential use cases in schools and early learning are highly varied if not yet currently
especially common.
However, more than a tool to improve the provision of education, AI is also a
prevalent subject of learning around the world as governments try to educate citizens
from all age-groups Thanks to its relationship with labour and employment, the
predicted structural changes caused by AI to jobs through increased automation will
have a drastic impact on education. Traditional educational institutions, employers and

379
Baker, Toby and Laurie Sue Smith (2019) “Educ-AI-tion Rebooted ? Exploring the future of artificial
intelligence in schools and colleges.”
380
European Commission, Joint Research Centre, Tuomi, I.,(2019) The impact of Artificial Intelligence on
learning, teaching, and education, Punie, Y.(editor), Vuorikari, R.(editor), Cabrera, M.(editor), Publications
Office

123
 Adopt AI Study

individuals will need to ensure what they are learning remains relevant to avoid being
left behind.
Even in the nearer term, adult learners must be provided with the means to prosper in
an ever-changing world as other previous skills become less useful. AI as a subject is
increasingly common with online learning and Massive Online Open Courses (MOOCs)
are seen as an effective strategy for the dissemination of awareness and
understanding of AI, even for policymakers themselves. In the words of an EIT report,
“Policymakers and employers must also create pathways for non-near AI talent to
eventually upskill to AI talent, for instance creating a learning pathway from business
operations analyst into a data scientist, and then finally into an AI worker”.381
The policy landscape: Education and AI
As with the other sectors, supporting the growth of AI in education requires a
comprehensive picture of the current activity in the research sphere and broader
market of AIEd382. Featuring every published National AI strategy, education or
“human capital” is a significant priority for the EU´s member states. Nevertheless, the
majority of policies are focused on learning about AI rather than learning with AI with
a particular focus on adult learning and upskilling (see figure below). However, this
uniform signal of intent has not translated into meaningful policy platforms particularly
procurement of AI technologies for use in the education system. While there is
considerable variety in national approaches and the nuances of the policies
themselves, it is possible to construct a broad typology of policies both in what role
they perform and who they target. Indicative of the focus on upskilling and adult
education that defines AIEd for many, the largest focus on policies is on industry and
professional training also in the form of a “course” (which is longer than a “workshop”
and has more restricted access than a MOOC). These can be relatively open like
Slovakia´s Kinit.sk or have a sector-specific focus such as Slovakia´s AI courses for
the public sector or Belgium´s EluciDATA lab which is marketed towards individuals
already working in the technology sector.
The educational initiatives on AI can be broken down broadly into 5 categories:
• Workshop. A short-term introductory session which will teach the learner the
basics of AI or demonstrate a particular use case
• Course. A medium to long/term learning programme typically takes several
weeks after which the learner will receive a qualification/certificate.
• MOOC. Similar to a course except with free and universal access
• Portal/Platform. The infrastructure upon which a user can host AI education
content or interactions between EdAI and Edtech companies.
• Centre. A multidisciplinary setting which combines many of the categories
above and provides research, short- and long-term learning opportunities.

381
EIT Health (2021) Analysis of the educational activities available in the area of Artificial Intelligence:
https://eit.europa.eu/sites/default/files/analysis_of_the_educational_activities_available_in_the_area_of_ai
_crosskic_deliverable_1.pdf
382
Baker, Toby and Laurie Sue Smith. “Educ-AI-tion Rebooted ? Exploring the future of artificial intelligence
in schools and colleges.” (2019)

124
 Adopt AI Study

Figure 62: Breakdown per policy type Figure 61: Policies per education level

Among the member states, Belgium leads the way with 10 policies which range from
project grants for edtech companies in the SmartEducation@schools initiative to the
“i-learn” portal and courses. Choosing the multidimensional approach of investment in
centres, France and Germany have established the interdisciplinary Institutes of AI
and the AI campus respectively. Without the financial clout for such large investments,
smaller member states such as Malta are also innovating. The country has helped set
up the AI Family Challenge and the AI Olympiad, which hones in on schools and
specifically Early Childhood and Care (ECEC) while most states are focusing on later
stages of education. As Nesta notes, this is still quite far away from the paradigm shift
in education and there are no systemic applications of AI to education although many
satisfy the EIT´s list below:

The EIT´s 10 actionable suggestions to become a leader in the European

AI education landscape383

 Focus on European applied topics like language and culture in the courses.

 Complement the hard skill education in AI with AI-related soft skills.

 Market AI programs specifically to a female audience to foster diversity.

 Strengthen collaboration with universities and facilitate the offering of accredited

courses.

 Create the first European open platform for online courses.

 Offer students AI courses free of charge and early in education.

 Help professionals in specific industries solve current problems with applied AI

workshops.

 Educate and enable managers to create pragmatic AI governance.

 Facilitate courses for decision-makers to understand the potential of AI.

 Strengthen courses for policymakers to understand the ethics of AI.

383
EIT Health (2021) Analysis of the educational activities available in the area of Artificial Intelligence:
https://eit.europa.eu/sites/default/files/analysis_of_the_educational_activities_available_in_the_area_of_ai
_crosskic_deliverable_1.pdf

125
 Adopt AI Study

2.5.1 Challenges and corresponding AI solutions/applications in the

education sector
As with all economic sectors, education is being redefined and profoundly transformed
by societal and technological challenges that have materialised over the past decades.
A UNESCO paper on AI in education divides these into 4 principal focus areas where AI
application can be found384:
 Lifelong learning (Student facing)
 Learning and assessment (Student-facing)
 Empowering teachers and enhancing teaching (teacher-facing)
 Education management and delivery (system-facing)
These different challenges will be briefly introduced as well as the AI techniques to
address them. However, one can already highlight that most of these techniques are
at a very early stage in terms of Technology Readiness Level (TRL). In reality, despite
some lighthouse projects, AI in education has few concrete examples of
implementation and most solutions are still at the R&D stage.
Figure 63 Summary of challenges and AI solutions for education

Public Ensure Improve learning Prepare teachers Increase efficiency in

sector relevant skills outcomes for the digital age education systems
challenge in the
population

AI value Lifelong Learning and Empowering Education

driver learning assessment teachers and management and
enhancing teachers delivery

AI AI-driven Intelligent tutoring AI-driven Educational chatbots

Solutions lifelong systems discussion forum
Predictive algorithms
learning monitoring
Dialogue-based tutoring
companions
systems AI-powered
AI-enabled teaching assistants
Exploratory learning
continuous
environments
assessment
Automated writing
AI-enabled
evaluation
record of
lifelong AI- supported reading
learning and language learning
achievements
Smart robots
Teachable agents
Education virtual and
augmented reality
Learning network
orchestrators

Lifelong learning
In the current age of rapid technological change, lifelong learning has become a
requirement for education systems to ensure their citizens remain competitive.

384
UNESCO, 2021, AI and Education: Guidance for policy-makers,
https://unesdoc.unesco.org/ark:/48223/pf0000376709

126
 Adopt AI Study

With the fast rate of obsolescence of knowledge and skills, teachers and students need
to constantly acquire new abilities and update their knowledge. The skills demanded
from the workforce are constantly changing, with more and more emphasis being
placed on STEM fields. “For example, 54 per cent of American workers who responded
to a Pew Research Centre survey in 2016 indicated that they thought they would need
further upskilling during their careers, and 39 per cent of large-company executives
surveyed by Deloitte expressed difficulty in finding the talent their companies
required. Many of the skills shortages lie in STEM fields.”385
Over the years, the interest in AI and Lifelong Learning has converged, with the hope
that new AI techniques could simplify and increase the impact of LLL on workers’
skillset. 386 In total, 3 main types of applications of AI for LLL can be identified: 387
Even though LLL and e-learning are dynamic and high-upside markets,388 the use of AI
in lifelong learning remains, so far, very limited. Indeed, most technologies in the field
have not reached the level of technological maturity needed for large-scale
implementation. Many projects are still underway, not just to develop these
techniques, but also to understand their potential and to assess their risks.389 In
addition, important ethical concerns are being raised regarding the risks of “Lifelong
Learning” turning into “Lifelong Tracking” with the use of AI techniques.390 391
In addition, criticisms were voiced against the very concept of LLL itself. Indeed, some
advocate that the replacement of “Lifelong Education” with “Lifelong Learning”
indicates the utter domination of economic-driven interest and the corporate world in
this sphere. This involves a series of risks such as the confinement of developed
materials to economically-driven topics and the exclusion of low-income or
unemployed people from most interesting initiatives. 392

385
UNEVOC – UNESCO, 2021, Understanding the impact of artificial intelligence on skills development,
https://unevoc.unesco.org/pub/understanding_the_impact_of_ai_on_skills_development.pdf
386
Rebecca Eynon et al, January 2021, Methodology, legend, and rhetoric: the constructions of AI by
academia, industry, and policy groups for lifelong learning,
https://journals.sagepub.com/doi/pdf/10.1177/0162243920906475
387
UNESCO, 2021, AI and Education: Guidance for policy-makers,
https://unesdoc.unesco.org/ark:/48223/pf0000376709
388
GlobeNewsWire, 21/05/2021, The e-learning market size is expected to grow at a CAGR of over 13%
during 2020-2026, https://www.globenewswire.com/news-release/2021/05/21/2233971/0/en/The-e-
learning-market-size-is-expected-to-grow-at-a-CAGR-of-over-13-during-2020-2026.html
389
For example: OxfordInternetInstitute, sin dato, Understanding the potential of AI in Lifelong learning: A
critical perspective, https://www.oii.ox.ac.uk/research/projects/understanding-the-potential-of-ai-in-
lifelong-learning-a-critical-perspective/
390
Solidar Foundation, 2020, Position paper,
https://www.solidar.org/system/downloads/attachments/000/001/142/original/SOLIDAR_Foundation_AI_Po
sition_Paper.pdf?1592215657
391
Regarding the ethical risks for AI in education, the reference is the “International conference on artificial
intelligence and education” (Beijing, 16/05/2019). Participants adopted the “Beijing Consensus on Artificial
Intelligence”. Declares a humanistic approach for usage of AI in education:
The Consensus details the policy recommendations on AI in education in five areas:
1. AI for education management and delivery;
2. AI to empower teaching and teachers;
3. AI for learning and learning assessment;
4. Development of values and skills for life and work in the AI era; and
5. AI for offering lifelong learning opportunities for all.

It also elaborates recommendations corresponding to four crosscutting issues:

1. Promoting equitable and inclusive use of AI in education;
2. Gender-equitable AI and AI for gender equality;
3. Ensuring ethical, transparent and auditable use of education data and algorithms; and
4. Monitoring, evaluation and research.
UNESCO, 2019, The Beijing Consensus on Artificial Intelligence and Education,
https://unesdoc.unesco.org/ark:/48223/pf0000368303
392
Poquet Oleksandra et al, 2021, Developing Capabilities : Lifelong learning in the Age of AI,
https://www.researchgate.net/publication/351700209_Developing_capabilities_Lifelong_learning_in_the_ag
e_of_AI

127
 Adopt AI Study

Generally, three main stakeholders are involved in AI for LLL: governments, industrial
players and academia. However, academia-designed-solutions are only rarely applied
in a real-life setting and tend to be complex and confined to the academic world and
are yet to become a mainstay of the business world.
Policy-makers could support the development of AI in LLL in three ways:
 The low maturity of the technology in the field requires more investment in
R&D. The technology still requires investments in R&D projects and would
benefit from support funds and research projects from public authorities in the
three fields: learning companions, continuous assessment and certification.
 To develop a more ethical approach, the “Ethical Committee” could be
developed to involve third parties, including citizens, governments, academia
and industry-players. These committees would be involved in the design and
monitoring of AI in LLL to avoid lifelong tracking.
 Public authorities can increase the uptake of market solutions and speed-up
development of new AI in LLL by procuring and pilot-testing these techniques
on their Lifelong Learning platforms (e.g., the “EAD” platform (Éducation à
Distance”) managed by the unemployment office in the south of Belgium).

Lifelong Learning

Type Description Maturity

 So far, only research

AI-driven  Development of an AI
projects at an early TRL
lifelong companion based on the
stage exist in that field.
learning skills, needs and
companions experience of the learner

 This type of application
AI-enabled  This type of application
already exists for basic
continuous already exists for basic
assessment (e.g., using
assessment assessment (e.g., using
scripted answers to
scripted answers to
answer basic questions).
answer basic questions).
 As for other AI solutions
AI-enabled  With the increase in
in LLL, this remains an
record of workers engaging in LLL,
R&D project with no
lifelong the need to make these
concrete examples.
learning learning achievements
achievements valuable in the work

market or with
traditional educational
institutions is
symmetrically rising. An
AI-enabled record of LLL
would track all
achievements,
certificates and courses
taken to develop a
personalised portfolio
that could be shared with
an employer.
Learning and assessment
The area which has received the most attention thus far, for better and for worse, is
AI In learning and education. Particularly pervasive thanks to the pandemic, AI-

128
 Adopt AI Study

enabled education seeks to provide access to “high-quality, personalized and

ubiquitous” learning393.
New technologies, global competition and even the rise of AI itself mean that
curriculum must be constantly improved to integrate new technological-driven
knowledge. Indeed, education with AI does not simply mean using AI for educational
purposes. It also means preparing workers to learn about AI and also use AI in their
future work.394
With the rapid rise in the demand for new skills, education institutions, students and
workers face difficulties in maintaining employability. Similarly, employers have
difficulties finding the correct profile for the tasks at hand. This mismatch is already
well-known. AI can help resolve these difficulties by helping to raise awareness about
employability and helping individuals to develop a career path: 395
“Helping to develop self-awareness about choices and develop career paths
 AI experimental solution for eye tracking and scanning that can assess
personality
 AI games that let people express and identify their knacks (traits and abilities)
 AI-powered Chatbots that advise people about their career path and exploration
options and resources
Learning and skills building
 AI educational solutions that adapt and personalize the learning experience
 AI-powered augmented reality solution that builds interactive visual material
 AI virtual learning assistant
 Speech-to-text AI learning solutions
Qualification assessment and matchmaking:
 AI virtual assistants for profile building, resume review and ATS compatibility
 AI-powered neuroscience games to match candidates to the best-fit jobs
 AI search engines that sort and categorize jobs from a wide range of job sites
 AI chatbots for job recommendations
 AI matchmaking platforms that use different forms of personal and qualification
assessments”
Unfortunately, AI for employability is dominated by foreign companies. The US
companies represent 52% of innovations in the domain, followed by the group made
of Israel, Canada and the UK (25%), followed by Europe (11%) and China and India
(around 8% combined). Combined, African companies represent 4%. 396
Nonetheless, this remains a promising field for further research. AI for higher
employability would reduce friction in the work market and recruitment processes.
Furthermore, research in the development of AI techniques is, in itself, a way of
creating new jobs 397 Chatbots and other AI systems can half the matchmaking
process time for recruitment. 74% of candidates dropping from a candidacy process

393
UNESCO, 2021, AI and Education: Guidance for policy-makers,
https://unesdoc.unesco.org/ark:/48223/pf0000376709
394
Brookings (2018) The role of AI in education and the changing US workforce,
https://www.brookings.edu/research/the-role-of-ai-in-education-and-the-changing-u-s-workforce/
395
Marwan Akram (2020) Impact of artificial intelligence on education for employement: learning and
employability framework, https://fount.aucegypt.edu/cgi/viewcontent.cgi?article=1839&context=etds
396
Ibid
397
Ibid

129
 Adopt AI Study

made that decision out of “frustration” because of a complex or unclear procedure —

something that could be greatly simplified thanks to AI techniques. 398
To help support these techniques, it would be recommended that:
 The curriculum should be updated to include skills that are important in new
economies. Legislation and policy-makers should support the development of
partnerships between schools and their ecosystem to support workforce
development partnerships (involving educators, internships, etc). 399
 The development of gamified AI to improve self and market awareness among
young people can help to decrease the mismatch between job offers and
demand. 400

Student-facing AI

Type Description Maturity

 Most mature field with

Intelligent  Provides step-by-step
some concrete
tutoring tutorials individualised to
applications cases. Also
systems each student on
generated interest from
structured topics (e.g.,
the private world with
IT and mathematics).
ongoing investments.
However, this solution is
Sometimes implemented
based on ignoring other
in Learning Management
learning assumptions
Systems (e.g., Khan
(e.g., Productive
Academy).
Failures). Furthermore,
this approach prescribes
 60 commercial ITS
available, the main ones
content instead of
are Alef, ALEKS, Byjus,
encouraging students’
Mathia, Qubena, Riiid,
agency.
and Squirrel AI.
 There is only limited
evidence of these
solutions’ effectiveness.
 Low maturity.
Dialogue-  AI uses natural
based tutoring languages to simulate a
 Research project with a
systems heavy focus on
spoken tutor-student
application in ICT.
dialogue. Encourages
Examples: Watson tutor
students to co-create
and AutoTutor.
explanations to have a
deeper understanding.
 Only research projects
Exploratory  Students are encouraged
exist.
learning to explore their learning
environment environment to build
 Examples are ECHOES,
Fractions lab and Betty’s
their path and
Brain.
curriculum. The role of
AI is to provide
automated guidance and
avoid cognitive overload

398
Ibid
399
Brookings (2018) The role of AI in education and the changing US workforce,
https://www.brookings.edu/research/the-role-of-ai-in-education-and-the-changing-u-s-workforce/
400
Marwan Akram (2020) Impact of artificial intelligence on education for employement: learning and
employability framework, https://fount.aucegypt.edu/cgi/viewcontent.cgi?article=1839&context=etds

130
 Adopt AI Study

leading to students
giving-up.
 Already used in an
Automated  Automated techniques to
education context with
writing provide feedback on
programs: WriteToLearn,
evaluations writings to help a
e-Rater and Turnitin.
student improve writing
or to facilitate (and
automate) scoring of
students’ works.
 Varying level of
AI-supported  Includes speech
maturity. Genuine AI
reading and recognition used to
solutions are still in the
language compare students’
early research stage.
learning productions with a
Some simpler versions
recorded sample of
are already in use, eg:
native speakers. Also
AI teacher, Amazing
includes automatic
English, Babbel, and
translation to read
Duolingo.
learning materials in
other languages.
 Already some examples
Smart robots  Speech-enabled
of applications in Asia.
for social humanoid robots created
Examples: Nao and
inclusion for learners on the
Pepper (kindergarten
autism spectrum provide
robot in Singapore) used
mechanical interaction
to introduce kids to
more reliably than
STEM and programming.
human ones. Or
telepresence robots for
sick or absent students.
 Only at a research stage,
Teachable  AI acts to simulate
the “Betty” project is the
agents another student that can
main existing example.
be taught by the learner.
This helps and
encourages the active
acquisition of knowledge.
 Already been used for
Educational  VR and AR-assisted
some specific learning
virtual and learning to increase
activities in geology or
augmented learners’ engagement.
biology (e.g., simulating
reality
being inside a human
womb or volcano).
 Examples: Blippar, Eon
reality, Google
Education, Neobear, and
VRMonkey.
 No concrete examples,
AI-enabled  Enhancing learning
only a research project
collaborative outcomes through the
at this stage.
learning creation of a
collaborative
environment. Help to
coordinate activities
between learners and
between learners and
teachers.

131
 Adopt AI Study

Empowering teachers and enhancing teaching

The same considerations of needing to keep the skills and knowledge base in the
population no only applies but is even more acute with Europe´s teachers and
educators. Most EU countries face the dual challenge of an ageing work force and a
general shortage of teachers. For example, in Italy, 53% of teachers are aged 50 or
older.401
Artificial Intelligence can help tackle these obstacles in two different ways. Firstly, AI
can drive the apps and systems to personalise and automatically adapt training offers
(by creating learning pathways and automatically adapting the classes and courses to
the student). 402 Secondly, AI can automate repetitive work thus sparing time for
teachers to focus on key tasks.
Together, these solutions represent an important gain in teaching times. According to
the University College London, teachers in England work, on average, 47 hours a
week, and less than half that time is dedicated to teaching tasks. 403 These findings
match the results from the study carried out by McKinsey:

Source: McKinsey 404

According to McKinsey: “Our current research suggests that 20 to 40 per cent of

current teacher hours are spent on activities that could be automated using existing
technology. That translates into approximately 13 hours per week that teachers could
redirect toward activities that lead to higher student outcomes and higher teacher
satisfaction. In short, our research suggests that existing technology can help teachers
reallocate 20 to 40 per cent of their time to activities that support student learning.”405

401
European Parliament (2020) Teaching careers in the EU,
https://www.europarl.europa.eu/RegData/etudes/BRIE/2019/642220/EPRS_BRI(2019)642220_EN.pdf
402
Some concerns are raised about the potentiality of AI to really empower the workforce/student. See for
example: Akhras Fabio et al, June 2011, First workshop on Artificial intelligence in Education to support the
social inclusion of communities,
https://www.researchgate.net/publication/251401277_First_Workshop_on_Artificial_Intelligence_in_Educati
on_to_Support_the_Social_Inclusion_of_Communities_AIEDSIC
403
Financial Times (2020) How AI eases teachers’ heavy workloads, https://www.ft.com/content/da3a9bac-
3e0a-11ea-b84f-a62c46f39bc2
404
McKinsey (2020) How Artificial Intelligence will impact K-12 teachers,
https://www.mckinsey.com/industries/education/our-insights/how-artificial-intelligence-will-impact-k-12-
teachers
405
McKinsey (2020) How Artificial Intelligence will impact K-12 teachers,
https://www.mckinsey.com/industries/education/our-insights/how-artificial-intelligence-will-impact-k-12-
teachers

132
 Adopt AI Study

Even though the exact time spent on each task might differ for European countries,
the general lessons' conclusions and logic still apply to the European context. Policy-
makers can support the development of these solutions in the following way:
 Investments should be targeted on solutions that are mature and easy to
implement to accustom and upskill the teacher workforce. Teachers' and school
leaders’ capacity to harness new technologies should be strengthened. 406
 Development and implementation of these techniques should be done
inclusively, involving all key stakeholders: school directors/leaders, technology
companies, governments, teachers and learners. 407

Empowering teachers and enhancing teaching

AI-driven  Monitor discussion  Successful pilot projects

discussion forums (where students conducted in the US face
forum can respond to tasks, important ethical
monitoring and ask questions about criticisms. The Assistant
materials, classes, etc). Jill Watson developed at
AI can answer simple Georgia Tech in the US
questions (about was used to send
schedules for example), assignments deadlines
triage questions and and other basic
identify the relevant information by email.
teacher or even use However, the program
sentiment analysis to was criticised for
identify non-productive emulating to be a human
emotions (e.g., by delaying answers and
aggressivity). even using humour.

AI-Human  Instead of replacing  Research project. Mostly

“dual teacher” humans to carry specific developed in China for
model tasks, this paradigm remote areas. Existing
relies on developing a programmes use videos
mix human-AI approach. to have experienced
teachers giving classes
 The AI could help the to remote locations
human teacher with
where they are assisted
many tasks, including
by local and less
providing specialist
experienced teachers.
expertise or professional
The Le Waijiao AI
development resources,
classroom is developed
collaborating with
to replace the long-
colleagues, both within
distance teacher with an
and outside the
AI to support the local
particular setting,
teacher.
monitoring the students’
performance, and
tracking progress over
time.

406
McKinsey (2020) How Artificial Intelligence will impact K-12 teachers,
https://www.mckinsey.com/industries/education/our-insights/how-artificial-intelligence-will-impact-k-12-
teachers
407
McKinsey (2020) How Artificial Intelligence will impact K-12 teachers,
https://www.mckinsey.com/industries/education/our-insights/how-artificial-intelligence-will-impact-k-12-
teachers

133
 Adopt AI Study

AI-Powered  As mentioned, many  Some research on these

teaching technologies are has been undertaken,
assistants designed to relieve but many technical and
teachers of time- ethical issues need to be
consuming activities overcome before they
such as taking can be harnessed in real
attendance, marking settings
assignments and
answering the same
questions over and over
again

Education management and delivery

In addition to supporting students and teachers, AI holds great promise when applied
to education systems on a broader level. At the school-level, AI can be used for
expediting organizational tasks and “learning analytics”:
Simplification and automation of administrative tasks (e.g., naturally written language
recognition and automatic grading for basic tasks 408; sending automated reminders for
students about homework and other information).
Forecasting and Efficiencies (e.g., enhanced planning for future administrative and
directorial tasks, avoiding waste and generating savings).
Real-time data: dashboards and systems to track and provide educators with
information about students (e.g., identification of students at risk of dropping-out).
Some school districts in the US already use this type of system.409
Furthermore, AI and big data can offer valuable insights at the system-wide level
where data from multiple educational institutions can be aggregated and analysed to
improve educational policy itself.

Education management and delivery 410

Educational • Chatbots are online • With varying levels of

Chatbots computer programs that use maturity, most solutions are
cloud-based services and AI at a research or pilot stage.
techniques to hold The most promising
simulated conversations examples (are basic
with people. The human chatbots replying using pre-
user types or speaks a programmed scripts);
question, and the chatbot advanced chatbots using
responds, providing natural language
information or undertaking recognition and machine
a simple task. learning to generate unique
responses.

408
“Century Tech’s system marks each student’s work automatically, collecting the data for teachers so they
can identify who understood the lesson and who struggled. Funded by angel and social impact investors, it
charges £5 a year per pupil, and invests revenues into research.” Financial Times, 09/03/2020, How AI
eases teachers’ heavy workloads, https://www.ft.com/content/da3a9bac-3e0a-11ea-b84f-a62c46f39bc2
409
EDSEMBLI, sin dato, How can AI help educators redirect time to what matters most?
https://www.edsembli.com/ai-and-workload-management-for-teachers/
410
Most of these solutions are already mapped in the section about “Changing demography”. Also see:
UNESCO, 2021, AI and Education: Guidance for policy-makers,
https://unesdoc.unesco.org/ark:/48223/pf0000376709

134
 Adopt AI Study

2.5.2 Education value chain analysis

While not as well-suited as with other sectors, the value chain approach can still help
to structure and guide the analysis. Indeed, it gives an idea of the different steps
needed to deliver education services. The figure below provides an elaboration of a
value chain and digital value-chain representation for the education sector.
Figure 64 Education value chain

Source: Authors’ elaboration.

This representation clarifies the different steps involved in the delivery of education
services. However, from the perspective of implementing AI solutions, these steps are
too intertwined to be analysed separately. Consequently, this graph should be seen as
an overview of the education sector and the digital functionalities that exist in the
sector. This section will investigate the three following points:
 The COVID-19 outbreak's impact on the digitalisation of the educational sector;
 AI in the EdTech industry;
 AI for the delivery of education services.

The COVID-19 outbreak's impact on the digitalisation of the educational

sector
With the closure of schools and the impossibility to deliver organise in a physical
format, the COVID crisis acted as an accelerator for the digitalisation of the
educational system, contributing to a year-to-year increase of 21% in the EdTech
market in 2021.411
While this increase is substantial, many elements highlight that these changes were
not as profound as originally expected or even hoped for. In reality, once able to “go
back to normal”, most schools abandoned the use of digital tools to revert to original

411
Forbes (2022) Five trends the EdTech industry should pay attention to in 2022,
https://www.forbes.com/sites/forbesbusinesscouncil/2022/03/10/five-trends-the-edtech-industry-should-
pay-attention-to-in-2022/?sh=17fa884a5a4a

135
 Adopt AI Study

teaching methods. Overall, the desire to maintain a mixed-teaching methodology

incorporating digital solutions seems to be lacking throughout Europe. 412
More profoundly, even though digital solutions were key to maintaining learning
activities, they fell short of the initial objective of maintaining a similar level of quality.
An impact study from the Netherlands concluded that the use of Digital schooling
during the crisis led to lower learning gain and increased inequalities. On average,
Dutch high-school students lost between 0.06-0.2 points in reading and between 0.13-
0.33 in mathematics. The greatest learning losses were observed in lower-income and
lower-education households. 413
In reality, most countries already launched ex-post remedial plans to compensate for
the learning losses. Many hopes surrounded the use of AI during the crisis but also in
the context of these remedial plans. However, concrete achievements and the use of
AI are very limited. In reality, the majority of measures were focused on setting-up
emergency solutions to reach all students. Neither curricula nor teaching methods
were adjusted to incorporate AI or other advanced digital techniques. Similarly, only a
few efforts were made in monitoring the acquisition of learnings and measuring
results.414
In the same order of ideas, in the UK, 86% of the technologies introduced in the
primary school system during the crisis were learning platforms used to send and
received pupils’ work. In high-school, 89% of the digital systems implemented served
to enable live remote sessions and learning remotely.415 Comparatively, the use of
advanced technologies and AI systems was very modest, with only 16% of high
schools and 5% of the primary school mentioning making use of AI416: “Use of other
technologies was significantly lower. Assistive technology was the most commonly
used, followed by learner analytics and secondary schools were significantly more
likely to use these technologies compared to primary schools. A minority of schools
used any artificial intelligence (AI), virtual reality (VR) or augmented reality (AR)
technologies. Use of accessibility features built into mainstream devices and software
was particularly low in primary schools.”417
Despite optimistic claims, the main achievement of the crisis regarding educational AI
was to increase positive attitudes toward AI technologies in education, but with limited
impact in terms of actual implementation and use. 418
Nonetheless, several measures could help societies to reap the benefits of the COVID
crisis in the education sector:
 The COVID crisis acted as a large-scale experiment for testing digital systems in
educational institutions. However, there is a lack of impact studies on the
matter. Research efforts are needed to map digital strategies; identify what

412
Vegas Emiliana (2022) Education technology post-COVID-19: a missed opportunity?,
https://www.brookings.edu/blog/education-plus-development/2022/03/11/education-technology-post-
covid-19-a-missed-opportunity/
413
Haelermans Carla et al, (2022) Sharp increase in inequality in education in times of the COVID-19
pandemic, https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0261114
414
WorldBank (2022) Remote Learning During the Global School Lockdown: Multi.country lessons,
https://documents1.worldbank.org/curated/en/668741627975171644/pdf/Remote-Learning-During-the-
Global-School-Lockdown-Multi-Country-Lessons.pdf
415
UK Department of Education (2021) Education Technology (EdTech) Survey 2020-21,
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/105781
7/Education_Technology_EdTech_Survey.pdf
416
However, this figure is not detailed enough in the report. It is therefore unclear if these were genuine AI-
systems or not. Furthermore, AI applications can include chatbot to answer to student’s questions about
schedules.
417
UK Department of Education (2021) Education Technology (EdTech) Survey 2020-21,
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/105781
7/Education_Technology_EdTech_Survey.pdf
418
Pantelimon Florin-Valeriu et al (2021), The evolution of AI-driven Educational Systems during the
COVID-19 Pandemic, https://www.mdpi.com/2071-1050/13/23/13501/pdf?version=1638844186

136
 Adopt AI Study

exact solutions were used; identify best-practices; identify clear failures and
mistakes to avoid. Based on these elements, a roadmap for digitalisation and
implementation of AI in the education sector could be developed. 419
 The “back to normal” scenario seems to highlight that the changes made during
the COVID crisis were superficial (no changes in curriculum, measures were
mostly focused on organising classes using videos, etc) with no explorations of
the possibility to develop mixed-methodologies. Identifying the reasons behind
this fact could help to develop a clear and substantiated understanding of what
are the exact barriers and solutions for the uptake of AI in education.420
AI and the European EdTech industry
The expected benefits for the education sector from the wider use of EdTech are
widely acknowledged and accepted. For example, in the UK, 88% of headteachers and
84% of teachers agree that these technologies already deliver benefits and will
continue to do so, especially in terms of reducing workload and saving time to help
teachers focus on key tasks.421
Simultaneously, many optimistic assessments and statements highlight the market
potential for EdTech. 422 The World Bank, for example, expects Education AI to reach
$6 billion by 2024.423 Marketsandmarkets assess the global AI in the education market
to reach USD 3.68 billion by 2023 at a CAGR of 47%. 424 Similarly, the e-learning
market size is expected to rapidly grow at a CAGR of 13% during the 2020-2026
period.425
However, EdTech is a complex market that includes different technologies and
solutions including blockchain for administrative documents (e.g., digitalised degree
ensuring traceability), big data, gamification, platforms, etc. 426
In reality, the importance of Artificial Intelligence in the existing business model of
EdTech companies is much more limited than sometimes advertised. 427 “Answering
[the question of implementing AI in education] and similar questions is challenging
due to the unstructured market, lack of evidence, and topic-specific complexity.
Moreover, the market is growing much more slowly than other markets are with the

419
For example, the initiative in Peru to ask daily reports from teacher using the digital platform led to
increasing the workload and causing burnout in the educational personnel. WorldBank, 30/03/2022, Remote
Learning During the Global School Lockdown: Multi.country lessons,
https://documents1.worldbank.org/curated/en/668741627975171644/pdf/Remote-Learning-During-the-
Global-School-Lockdown-Multi-Country-Lessons.pdf
420
Vegas Emiliana (2022), Education technology post-COVID-19: a missed opportunity?,
https://www.brookings.edu/blog/education-plus-development/2022/03/11/education-technology-post-
covid-19-a-missed-opportunity/
421
UK Department of Education, (2021), Education Technology (EdTech) Survey 2020-21,
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/105781
7/Education_Technology_EdTech_Survey.pdf
422
“Investors generally continue to be enticed by education technology companies such as Century, with
venture capital funding for the sector surpassing $8bn in 2018, according to Holon IQ, a market intelligence
agency. Think and Learn, the Indian company behind learning app Byju, was recently valued at $8.2bn
following fresh funding of $200m from US private equity group General Atlantic.” Financial Times,
09/03/2020, How AI eases teachers’ heavy workloads, https://www.ft.com/content/da3a9bac-3e0a-11ea-
b84f-a62c46f39bc2
423
UNESCO (2021) AI and Education: Guidance for policy-makers,
https://unesdoc.unesco.org/ark:/48223/pf0000376709
424
Marketsandmarkets, May 2018, Ai in Education market, https://www.marketsandmarkets.com/Market-
Reports/ai-in-education-market-200371366.html
425
GlobeNewsWire, 21/05/2021, The e-learning market size is expected to grow at a CAGR of over 13%
during 2020-2026, https://www.globenewswire.com/news-release/2021/05/21/2233971/0/en/The-e-
learning-market-size-is-expected-to-grow-at-a-CAGR-of-over-13-during-2020-2026.html
426
Which can lead to complete overestimation of the market, such as Globant estimating the market to
reach 7 trillion dollars by 2027. Globant, 2021, The Next Wave of EdTech, https://reports.globant.com/wp-
content/uploads/2021/12/Whitepaper_EdTech-1.pdf
427
Nazaretsky et al, 2021, An instrument for measuring teacher’s trust in AI-Based Educational
Technology, https://dl.acm.org/doi/abs/10.1145/3506860.3506866

137
 Adopt AI Study

dynamics of digital transformation” thanks to the diversity of actors involved in the

market.” 428
Concrete use cases of AI applications in education exist such as the
DreamBoxLearning program for mathematics developed by the Netflix group in the US
to enable personalised learning using AI and adaptative intelligent learning systems.
There is also the Knewton solution to analyse user data to establish success prognoses
and identify learning profiles; the use of IBM’s Watson supercomputer to provider 365
days of feedback to students at Deakin University in Australia; the use of Degree
Compass by Austin Peay State University to establish probabilities for students to pass
courses; the eAdvisor solution used by Arizona State University. 429
However, most of these are flagship initiatives with little progress in terms of
replication or large-scale implementation. In reality, there are very few examples of
successful implementation of AI solutions in education settings. Many cases that are
advertised as AI solutions, once investigated, turn-out to not be data-driven AIs. 430 431
In addition, in a European context, it is extremely difficult to identify similar flagships
and initiatives leading to more difficulties to encourage replication. Moreover, Europe
tends to lag in the development of AI solutions and is also showing a lack of
investment in the Education Technology market compared to China and the US.
Figure 65 Comparison of venture capitals in Education Technology for China, the US and Europe

Source: FT.

In reality, even if most EdTech companies offer some AI services, most of them are
limited and lack a data-led business model. While there is evidence of the production

428
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.
429
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.
430
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.
431
Financial Times, 09/03/2020, How AI eases teachers’ heavy workloads,
https://www.ft.com/content/da3a9bac-3e0a-11ea-b84f-a62c46f39bc2

138
 Adopt AI Study

of interactive whiteboards, educational software as a service, hosting of educational

data, etc, there is a lack of genuinely mature products already available on the
market. 432
Educational institutions are in a make-or-acquire dilemma situation. Institutions have
to decide whether to develop the solutions they envision or acquire them.
Furthermore, EdTech companies can act as competitors with education institutions as
EdTech companies tend to become education service providers for AI solutions and
systems.433The market is thus at a very early stage with low maturity, low demand
and limited offer. One of these powerful barriers is the lack of acceptance of AI
solutions in an educational setting.434 435
Figure 66 Drivers and barriers to the uptake of AI in education

Source: Renz André et al 436

Policy-makers can undertake several actions to encourage the development of the

educational AI market:
 Tax incentives and other fiscal measures to encourage greater investment of
venture capital in the development of educational AI.

432
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.
433
Nazaretsky et al, 2021, An instrument for measuring teacher’s trust in AI-Based Educational Technology,
https://dl.acm.org/doi/abs/10.1145/3506860.3506866
434
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.
435
To highlight this reluctance from the educational institutions, we can mention the position taken by The
“European Trade Union Committee for Education” (ETUCE). The organisation released a position paper about
the EU regulation of Artificial Intelligence. According to ETUCE, the classification of AI tools in education
requires stricter regulations to mitigate risks. Furthermore, the union demands to forbid AI tools aimed at
replacing or de-professionalising teachers and education personnel. Furthermore, ETUCES claims that
transparency should be ensured and solutions developed in partnership with education personnel as co-
creators of AI tools in education. Education social partners should therefore be engaged in the upcoming
“Artificial intelligence Board”. Finally, greater AI and technology literacy of the education personnel must be
ensure. ETUCE, 15/06/2021, Artificial Intelligence in education must not undermine the professional role of
teachers, https://www.csee-etuce.org/en/news/education-policy/4474-artificial-intelligence-in-education-
must-not-undermine-the-professional-role-of-teachers
436
Renz André et al, 24/02/2020, Prerequisites for artificial intelligence in further education : identification
of drivers, barriers, and business models of educational technology companies,
https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-020-00193-
3#:~:text=While%20the%20desire%20for%20flexibility,solutions%20at%20the%20same%20time.

139
 Adopt AI Study

 Listing, mapping and assessing existing European flagships initiatives to

encourage replication, identify best practices and disseminate results.
 Tenders and granting research funds and support to European companies active
in educational AI to complement the lack of market dynamism in the field.
 Development of large pilot projects for the procurement and implementation of
existing solutions as well as identifying key research needs and future
pathways.

2.5.3 Main drivers and barriers

The table below presents a PESTEL analysis of the uptake of AI technologies in public
procurement in the e-government sector.

140
 Adopt AI Study

Table 7 PESTEL analysis of the uptake of AI technologies in public procurement in the education sector

Criteria Drivers Barriers

 There is a huge opportunity for a policy that  A profound change in the education system
supports innovative ecosystems to realize the requires a large amount of political capital due
opportunities of AI in the field of education, as to the difficulty and time needed to implement
educational institutions are aware of the solutions on a system-wide basis.
necessity of AI adoption.
Political  The adoption of AI in education has the
potential to use data to facilitate and lower the
effort on the assessment of the educational
performance of regions, countries and Europe
for the creation of reports and strategic plans
for the future.

 The use of AI can help school leaders to predict  A major challenge in the adoption of AI by
issues before they arise. AI solutions can be educational institutions lies in planning and
used to lower energy costs, improve the strategizing the school’s budget. As some AI
management of staff optimise resource solutions and AI-powered technologies can have
planning, etc. Education institutions can make a considerable cost, understanding which AI-
advantage of AI to increase efficiencies and based learning tools are available and how they
reduce the number of unforeseen events. can benefit the students.
Economic
 AI applications can reduce the administrative
tasks of educators while freeing up time from
repetitive tasks can have an impact on educator
recruitment and retention. Additionally, it
increases the efficiency, as well as the
dedication of teachers to educational and
student-oriented actions.

Social  AI tools can help make impactful decisions in  Learning complex skills like developing
the classroom, which will enhance the learning perspectives, socializing in professional norms,

141
 Adopt AI Study

process of students and their educational feeling motivated, and mastering expertise
levels. Aspects such as determining which requires the guidance of a human teacher.
portions of the content need to be covered in
greater depth and which resources are the best
 Potential to increase the digital divide as the
adoption of AI requires minimum technological
to use for a given unit and why along with
characteristics (stable and reliable
providing the opportunity to cover more
electrification, internet connection, hardware,
material will potentially improve each student’s
etc.) which might be a barrier in less developed
performance.
areas, regions or countries.
 Personalization is one of the biggest
contributions that AI can provide to students
 There is an important need to train and upskill
all key stakeholders in AI including school
through a personalized approach to learning
directors/leaders, technology companies,
programs based on their own unique
governments, teachers and learners. Teachers
experiences and preferences. AI can adapt to
will need to be trained in new teaching
each student’s level of knowledge, speed of
methodologies and tools, students will need to
learning and desired goals so they are getting
get familiar with AI technologies and make safe
the most out of their education.
use of them; school leaders will have to
 AI tools can be successfully trained to help any understand the benefits AI technologies will
group of students with special needs, thus provide to the educational centres to invest in
ensuring access to education for students with them and technology providers will have to
special needs. The adoption of innovative AI understand the needs of the educational
technologies can open up new ways of environment and create efficient and
interaction with students with learning trustworthy solutions.
disabilities.
 Universal 24/7 access to learning. Accessibility
for all students, anytime and anywhere to
learning can be guaranteed through AI-powered
tools. It provides the possibility for each
student to learn at their own pace, and 24/7
access makes it easier for students to explore
what works for them without waiting on an
educator.

142
 Adopt AI Study

 Potential to build large-scale data sets about  Interoperability and the need for a complete
the performance of students in a particular tool that includes several functionalities and
category across different regions, countries or covers all the requirements of a classroom
the world. rather than various tools that are individually
used for a singular purpose. The time it takes to
 AI allows the design and creation of Smart transition between tools for different uses, and
Content such as digital lessons (Digital learning
the static nature or lack of flexibility of some
interfaces with customization options, digital
tools concerning how they are and can be used
textbooks, study guides, bite-sized lessons,
are considered a barrier in the educational
etc.), information visualisation (New ways of
processes.
perceiving information, such as visualization,
simulation, web-based study environments,  Having highly reliable data is very sporadic,
etc.) and learning content updates. Textbooks unevenly distributed, distorted, and limited in
can be digitized, and new learning interfaces many other ways. Additionally, for the
are being created to help students of all description of the learning process, other data
academic grades and ages. such as the physical and emotional health of
Technological students, their socio-economic status, family
situation or governmental policies can affect
learning progress.
 Technological linearity. To embed AI in
Education, it is required that the technologies
developed should be introduced early in
conjunction with other digital tools so that its
more widely accepted.
 Low Technological Readiness Level. Most
technologies in the educational field have not
reached the level of technological maturity
needed for large-scale implementation. There is
a need to understand their potential and assess
their risks, considering important ethical
concerns and biases.

Environmental  AI solutions can be used in educational  The positive environmental impact can be
institutions to lower energy consumption. shadowed by the negative environmental

143
 Adopt AI Study

Accurate forecasts generated using machine impact, mainly in terms of energy consumption,
learning based on historical patterns can of AI solutions, as well as other aspects such as
ensure, for example, that heat and light are the implementation of large server farms.
provided according to the actual demand.
 The creation of Smart Leaning Materials can
significantly reduce the number of physical
materials used in educational institutions. Along
the same line, de minimisation of physical
material can translate into waste reduction.

 The widespread adoption of AI solutions in  Attributing responsibilities. There is still

education and other domains is a driver for the uncertainty about who will take responsibility
adoption of legislative adjustments required to for the negative consequences that might occur
properly and effectively address data handling from the usage of AI technologies, for example,
issues. if a bias is detected in an AI-based system that
makes decisions about enrolment or grading.
Legal
 Biases. If there is a bias in the initial dataset,
machine learning algorithms will eventually
incorporate it and proceed to make biased
decisions. Therefore, guaranteeing aspects such
as that the admission or grading process will be
fair, is complicated.
Source: Authors’ elaboration based on the challenges and the value chain analysis.

144
 Adopt AI Study

Figure 67 Strengths, Weaknesses, Opportunities and Threats of the uptake and public procurement of AI-technologies in
the education sector

S
Educational institution management -AI solutions have the power to optimize the
overall management of educational institutions in areas such as resource planning, energy
consumption, staff management, etc. Adoption if Artificial Intelligence, together with other
technologies such as IoT, are able to substantially increase efficiencies and reduce the
number of unforeseen events.
Better engagement – AI adoption in education generates a positive impact in student
engagement. A personal approach (through individualized schedules, custom tasks,
interaction with digital technologies, Smart Contents and personal recommendations)
helps students feel special, increasing their engagement and raising interest in studies.

W Slow on the uptake related to legal and ethical aspects - Regulations and laws aren’t
keeping up with the rapid development of the technology.
Skilled teachers and students – Students and teachers should have adequate
understanding and skills in the usage of AI technologies in order to make the most of the
AI enhanced education. Although the adoption of digital skills appear to be commonplace,
AI based solutions will require readjustments of people skills for the teachers and the

O
students.

Universal AI assisted education – The general concern of educational institutions for

improving and enhancing their learning processes through the application of digital
technologies, including AI tools, presents the opportunity for the design and development
of universal or standard AI powered educational platforms and technologies. Moreover, a
standard platform would ensure educational interoperability at region, country and
European level.
Student experience – Adoption of AI in educational institutions has the potential to
personalise and, therefore, enhance the experience of each student. Student experience
will be improved through a personalized approach to the learning program based on each
student’s characteristics, combined with the assistance that AI can provide to teachers
when making impactful decisions.
Policy creation – Being education a common aspect for all European countries, there is a
huge opportunity for the creation of policies at region, country or European level in order
to overcome the specific necessities and generate solutions and guidelines to support AI

T
adoption for the creation of a innovative educational ecosystem.

Digital gap– Although in European context digital gaps are not as noticeable as in a
Worldwide context, there are still areas, regions or countries that are currently
overcoming the digitalisation barriers.
Privacy concerns - Collection, storage, disclosure, processing, and dissemination of
personal data is required to power AI based solutions in education. Additionally, as it
comes to the education sector, most of the data concerns to underaged people.

Source: Authors’ elaboration based on the challenges, value chain and PESTEL analysis.

2.5.4 Case study: Educating the world on AI: The elements of AI MOOC
(Finland)

Key Actors
 University of Helsinki
 MinnaLearn (formerly Reaktor)
 The Finnish Ministry of Economic Affairs and Employment

145
 Adopt AI Study

Type of policy:
The initiative was funded from the budget of the Ministry of Economic Affairs and
Employment and is estimated to have had a budget of EUR 1,679,000 (VAT 0%). This
policy is associated with the areas of digital transformation, artificial intelligence, new
skills, the transformation of work and sustainable development which Finland’s
Presidency was spearheading during its Presidency of the Council of the European
Union with the help of the Ministry of Economic Affairs and Employment.
Goals
The initiative had three aims: Finland, as the then outgoing Presidency, wanted to
invest in skills, increase EU citizens’ practical understanding of the opportunities and
nature of artificial intelligence and reinforce the digital leadership of the EU and thus
increase its competitiveness.
A competitive and socially inclusive EU is one of the main priorities of the Finnish
Presidency. Europe's competitiveness draws on skilled people and, increasingly, on the
use of artificial intelligence and digital solutions.
Project description
Elements of AI is a Massive Open Online Course (MOOC) that is aimed at anyone
interested in learning about artificial Intelligence. With the revolution that AI is
destined to bring about, this MOOC helps teach the basics of the technological area in
an accessible and well-designed manner. With its genesis in a shared building at the
University of Helsinki, what started as a digitalized version of an entry-level computer
science course at a single University now boasts over 750,000 users in 170 countries.
Aimed at the entire population rather than a narrow band of those with high
technological aptitude or interest, the course doesn’t require basic programming skills
or knowledge of advanced mathematics.
The core idea was to start by teaching 1% of the Finnish population about AI and build
incrementally. However, with the arrival of Finland´s Presidency of the Council of the
European Union, the MOOC was seen as the perfect vehicle for increasing the digital
skills of the entire EU. Increasing inclusion and competitiveness of the EU were some
of the main priorities of the Finnish Presidency and knowledge and uptake of AI are
seen as a way to achieve these ends. With the investment made by the original
partners and the Finnish Government, the MOOC was translated into every official EU
language and more. In 2019, the course won the Massachusetts Institute of
Technology´s global grand prize in the Inclusive Innovation Challenge 437.
Successes and Key lessons
 Inclusion and accessibility are building blocks for success. In the initial
ideation period for the MOOC, which was an adapted University Course, the
consortium took an initial version to be tested in high schools and reveal the
overly technical parts. This base helped to create a final product that would be
highly accessible regardless of age, reflected by the fact that over 25% of
students are over the age of 45 438. Furthermore, conscious of the considerable
gender gap that exists in computer science439 and STEM subjects440, the team
designed the course with a female as the principal user. This exercise allowed
the team to think inclusively about key components such as the language used
in the course, the UX and the UI. Subsequently, the course has seen a gender

437
Datarooms rating (2022) Tech for Good prosperity workers: https://dataroom-
rating.us/mitinclusiveinnovation/
438
Minnalearn (2023) Elements of AI https://www.reaktor.com/?page_id=102694
439
Texas Tech Today (2021) “Why is Computer Science Unpopular Among Women”
https://today.ttu.edu/posts/2021/09/Stories/why-is-computer-science-unpopular-among-women
440
American Association of University Women (2022) “The STEM gap: Women and Girls in STEM”
https://www.aauw.org/resources/research/the-stem-gap/

146
 Adopt AI Study

split of 40% women which, although it is not full parity, outperforms its
education sphere.
 Strategic dissemination is vital to scaling up. A huge part of the success and
popularity of MOOCs is the dissemination actions around them. Making use of a
synergistic consortium which saw private consultancy MinnaLearn take the lead
on dissemination, the strategy leveraged Finland´s strengths to great effect.
With the Finnish President in attendance at the first graduation of the course,
the team also used the various social media accounts of Finnish Ministries to
challenge their Swedish counterparts to take the AI challenge. Generating
attention around what was then still a national initiative, the same process was
carried out with other member states´ public sectors such as Germany. This
scaling up both paved the way for and formed the base of the strategy behind
the Finnish Presidency´s handling of the MOOC which would turbocharge the
expansion worldwide.

2.5.5 Economic analysis

Building on the literature and the inventory of policies, the quantitative analysis is
based on two sets of regression models focusing on the concentration of talent in a
selection of sectors and the penetration of AI in higher education. Each set
encompasses multiple indicators from reliable sources. This is undoubtedly
unrepresentative of the overall education sector with the omission of schools and
ECEC although both the lack of policies in the areas and married with the absence of
indicators means that they are not included in the analysis. On a broad level, User
countries are much more clustered with much larger variation in facilitator and planner
countries, especially vocationally.
Indicators and variables
Each set of models relies on a particular database and set of indicators. For the labour
aspect, the models use the OECD´s talent concentration dataset which gives both
member-state and sector-level data441. More specifically, the observations are the
percentage of LinkedIn members with AI skills or who perform an AI occupation (e.g.
machine learning engineer) – per country and industry over the last year. The models
focus on the following sectors: Education, software and IT services, finance, hardware
and networking and manufacturing
Alternatively, for higher education, the models use data from the 2021 edition of the
AI Watch Index and its societal dimension. Three categories are split into Bachelor and
Master, and the indicators measure the AI programmes, places and intensity. The
first442 indicates the intensity with which AI is included in official curricula, as a proxy
of the supply of AI capacities or the Proportion of programmes with AI content
compared with the total number of programmes. The second 443, places, provides an
estimation of the potential future workforce trained with AI skills in a specific AI

441
OECD.AI (2022), visualisations powered by JSI using data from LinkedIn, accessed on
22/6/2022, www.oecd.ai
Righi, R., López-Cobo, M., Alaveras, G., Samoili, S., Cardona, M.., Vázquez-Prada Baillet, M., Ziemba,
442

L.W., and De Prato, G., Academic offer of advanced digital skills in 2019–20. International comparison.
Focus on Artificial Intelligence, High Performance Computing, Cybersecurity and Data Science, EUR 30351
EN, Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76- 21541-9,
doi:10.2760/225355, JRC121680. https://publications.jrc.ec.europa.eu/repository/handle/JRC121680
443
Gómez Losada, Á., López-Cobo, M., Samoili, S., Alaveras, G., VázquezPrada Baillet, M., Cardona, M.,
Righi, R., Ziemba, L., and De Prato, G., Estimation of supply and demand of tertiary education places in
advanced digital profiles in the EU. Focus on Artificial Intelligence, High Performance Computing,
Cybersecurity and Data Science, EUR30377EN, Publications Office of the European Union, Luxembourg,
2020, ISBN 978-92-76-22281-1, doi:10.2760/559530, JRC121683.
https://publications.jrc.ec.europa.eu/repository/handle/JRC121683

147
 Adopt AI Study

domain. Definition Number of available places in university programmes with AI

content by AI domain (ML, AI ethics, Robotics, Computer vision…). The final
indicators444 measure the size of the potential future workforce trained with AI skills,
the Proportion of available places in university programmes with AI content in the total
number of places in university programmes.
Vocation

Talent Concentration Education Software and IT Finance Manufacturing Hardware

Services and
Networking

Intercept 4.46 3.33 0.95 0.69 1.45

Industry-focused AI 2.82 1.35 0.28 0.41 0.71

courses

R² 0.11 0.09 0.05 0.09 0.08

Observations 27 27 27 27 27

 Countries with programmes dedicated to vocational and industry-focused

policies on AI education have higher levels of AI talent concentration in
each of the 5 sectors than countries without any programmes.
 The gap is largest in education and software & IT services where there is
2.82% and 1.35% more talent in countries with adult learning AI courses.
 In finance and manufacturing, where even for countries with programmes the
talent concentration is less than 1%, the gap is small (0.28% and 0.41 less
respectively).

Higher education

AI in Higher BProgrammes MProgrammes BPlaces MPlaces BIntensity MIntensity

Education

Intercept 5.86 9.44 18586.00 8720.41 5.10 6.52

Higher 0.21 0.47 17696.11 17550.48 0.69 0.59

Education AI
Support

R² 0.01 0.01 0.05 0.16 0.01 0.01

Observations 23 27 23 26 23 26

444
Gómez Losada, Á., López-Cobo, M., Samoili, S., Alaveras, G., VázquezPrada Baillet, M., Cardona, M.,
Righi, R., Ziemba, L., and De Prato, G., Estimation of supply and demand of tertiary education places in
advanced digital profiles in the EU. Focus on Artificial Intelligence, High Performance Computing,
Cybersecurity and Data Science, EUR30377EN, Publications Office of the European Union, Luxembourg,
2020, ISBN 978-92-76-22281-1, doi:10.2760/559530, JRC121683.
https://publications.jrc.ec.europa.eu/repository/handle/JRC121683

148
 Adopt AI Study

 The models confirm the hypothesis that countries with policies that support
AI in higher education have a higher number of programmes, places
and intensity at both the bachelor and master levels.
 However, with very low R² scores for the “programmes” and “intensity”
models, the relationship between the two variables is very weak and
unreliable. Nevertheless, the direction is still positive in favour of countries
that have higher education AI policies in place.
 The most reliable models are the “places” where countries with policies are
correlated with over 17500 more places, in both bachelor and master degrees
containing AI learning, than those countries without.

2.6 Conclusions
While each sector is unique in its organization and dynamics, some common themes
would likely apply across most areas beyond those addressed in this chapter. Artificial
Intelligence has established a foothold in each of the 4 key sectors but has not yet left
a distinguishable mark – at least in the everyday lives of ordinary citizens. In other
words, the public sector has seen widespread surface-level integration of simpler AI
solutions and some more advanced applications but the transformative change which
AI has the potential to usher in is yet to materialize. Bringing about this change in the
status quo and overcoming the substantial barriers to adoption and implementation
described in the chapter will require a more emboldened public sector in both
procurement and development as well as at the national and European levels.
In all 4 sectors, the utility of AI applications for increasing efficiency is well-
established. Appropriately referred to as low-hanging fruit, these applications address
simpler administrative and “back-office” tasks as they are relatively undisruptive,
uncontroversial and cheaper than other solutions. An emerging paradigm that will
likely continue is the growth of AI-assisted hybrid cases where a public actor, whether
that be an administrator, teacher or healthcare professional, is supported by AI
technology in a manner that might seem more advanced than other past technological
assistance but not necessarily unfamiliar to the citizenry.
There are commonalities too between the 4 sectors with the more advanced AI cases.
Undoubtedly, proof of concept exists for many hugely promising applications that have
captured the collective imagination such as autonomous vehicles, AI-enhanced
education or patient-care. Nevertheless, these are still limited to small pockets of their
respective sectors and yet to become widespread. Furthermore, whether it’s a
hospital, a school or a transport hub, algorithmic prediction of real-time trends can
help any public entity better deploy its resources. However, once again, the difficulty
is to scale these proven applications into a smart and interconnected health or
transport system, a considerable and multifaceted challenge which is yet to be
conquered.
All 4 sectors showed that part of this challenge is inherently technical with a lack of
data infrastructure and general interoperability combined with a public sector which is
for the most part unfamiliar with the use and potential of AI. To this effect, what
happens in the education sector is critical to the progression of other sectors as
displayed by the need for public sector AI experts highlighted in the research of the
other 3 sectors. Moreover, at a time of rampant inflation and stretched public
finances, the considerable immediate set-up costs of adopting AI are an obstacle
throughout the government despite the promise of savings in the future. Finally, given
the considerable complexities and risks of AI, the socio-political dimension of AI
reticence due to complexity and bias will need to be addressed if the technology is
going to fundamentally change and improve society. This ties into some of the
fundamental threats to the successful roll-out of AI throughout the public sector; a

149
 Adopt AI Study

lack of trust, fears of trends like replacement in the labour market and concerns over
privacy and data. A common theme throughout the chapter was that increased
participation of the citizenry fosters greater understanding and increases the chances
that the public will buy-in to AI projects.
Perhaps unsurprisingly, within the European Union, regardless of the individual sector,
a group of larger and wealthier countries are leading the way in AI public policy.
Countries such as Belgium, France and Germany already have a well-rounded
repertoire of policy interventions including regulations, facilitating mechanisms,
funding regimes and crucially development of AI uses and solutions in-house.
Meanwhile, countries such as Croatia and Romania, which are yet to even publish an
AI Strategy as of late 2022, are risking being left behind and would greatly benefit
from increased public sector focus on AI and assistance from the European AI
ecosystem.
Ultimately, the pace of development, use and scaling of AI applications will likely need
to be turbocharged by a more entrepreneurial and less risk-averse approach from
national governments and the EU if Europe is going to achieve its broader policy goals
and the twin-transition.
For eGovernment, the main factors affecting the propensity of AI adoption were the
technological and financial capacities of the institutions looking to use the technology
in addition to specific, at times niche, specializations. With a few exceptions in local
settings such as Plovidic in Bulgaria and Katernini in Greece, the main users of AI
Govtech were those with high levels of prior digitalization such as the Nordics and
Western Europe. On the implementation side, political impetus and an
acknowledgement of the need for AI adoption ensured that there were few difficulties,
legally and culturally, in establishing the technology in public sectors while a lack of
skilled administrators in AI has held back progress. These trends were reflected in the
case study of Mercé which took place in Barcelona, a city with a highly skilled
workforce and innovative and well-funded public sector. The project itself, which was
dependent on the participation of individual citizens submitting data, is emblematic of
the necessary widespread public support and participation in AI projects such as this.
In terms of measuring outcomes, the models showed that across all three indicators,
countries using AI had better rates of use of eGovernment services and availability of
information online.
For the mobility sector, uptake and use have been quite limited with most of the
progress taking place in countries with highly developed transport industries although
a majority of countries had passed regulations in the sector or were supporting it
through funding or facilitating policies. AI in mobility is a proven solution to some of
the current environmental issues and has a demonstrable track record of being
economically viable. The mobility value chain was divided into the four transportation
modes that exist: road, railway, maritime and air with the current lack of effort for
cross-modal exchanges and fertilisation holding the sector back. Solutions developed
for driverless cars could, for example, benefit the railway industry. Specific
recommendations have been developed for each mode, including the uptake of smart
road infrastructure to increase the uptake of driverless-cars, investing in smart
maintenance and smart trackways using AI for railways, developing a common digital
approach for the uptake of AI systems across European ports (which would give
Europe a unique competitive advantage in the world) and encouraging the uptake of
smart maintenance in aerospace. Nevertheless, the sector´s growth is being stifled by
a lack of interoperability and harmonized rules not to mention very high access costs
to the technology. However, the Mobility Data Space in Germany is a shining example
of the potential for mobility AI once these barriers are overcome. With over 200
members, in a country with a highly sophisticated automotive industry, a collaboration
of data has been the emphasis with the project also helping allay societal concerns
such as safety and the environment.

150
 Adopt AI Study

For AI and health, the use cases are incredibly varied and touch every aspect of this
diverse ecosystem, something which can likely be attributed to the pervasiveness of
the policy of bottom-up funding programmes and the heterogeneity of actors in the
sector. The health value-chain was also divided into 3 segments: R&D, purchasers and
intermediaries, providers and end-users. Public authorities have a crucial role to play
in health as they are considerably the largest acquirers of health products and services
in Europe. The uptake and harmonisation of AI-enabled digital platforms for public
hospitals could be a boon for European health providers but the current landscape is
fragmented between different and sometimes incompatible platforms within nations
and across borders. The harmonisation of exchange systems across borders and within
national states could open many new opportunities for AI systems in health
procurement, improving administration and treatment delivery. The relatively low
uptake of AI-driven MedTech solutions highlights the role that public authorities could
play in boosting the market (demand side). This is especially true as public authorities
represent around 70% of all MedTech purchases in Europe. However, the acquisition
of advanced systems should be carried out while keeping in mind the need to
harmonise systems across European hospitals and healthcare. Furthermore, true AI-
enhanced MedTech is rare and more R&D and large-scale pilots remain necessary to
help the market reach full maturity (offer side). Spurred on by necessity amid the
COVID-19 pandemic, data sharing and cross-border exchange are increasingly
common while a lack of trust and doubts over the lack of familiarity and human
components are holding this back. The case study of Innoviris and Axiles Bionics is
emblematic of these trends with an open funding mechanism creating a
transformative AI solution for what is still a non-central area of healthcare in
prostheses.
In the education sector, uses of AI for delivery are still yet to become mainstream
albeit despite a surge in its teaching as a subject through MOOCs, specializations and
a mainstream subject at all levels of education. The promise of optimizing the
experience of teachers and learners alike has been held back by valid legal and ethical
concerns as well as a lack of familiarity of stakeholders with the technology. The main
conclusion from the education value chain analysis was the relative immaturity of this
sector for a larger uptake of AI solutions. The COVID-19 outbreak highlighted this by
providing a worldwide large-scale experiment but truly AI-driven systems were only
rarely applied or implemented during the outbreak. More research is necessary at the
European level to identify the AI tools that were implemented in education and
determine which were successful or lacking. This would be the first step to addressing
the two main barriers to a higher procurement of AI systems in education: a lack of
existing AI solutions and willingness from teachers to use AI solutions. Across the
board, those countries which have enacted national policies of educating people on AI
had higher concentrations of talent in five sectors (IT, finance etc). The “Elements of
AI” course is a worthy example of how an educational tool can be used to bridge the
digital divide and reduce unequal access to education about AI while also
demonstrating how understanding AI can lead to more innovation in developing uses
and solutions.
Finally, below is a short list of some cross-cutting themes from the best practices
of the case studies in the report which could be beneficial for any public authority
wishing to create policy in AI:
Inclusion breeds trust
Inclusion was a considerable part of both the Mercé and Elements of AI projects, the
case studies in the eGovernment and education sections. With the former, the
inclusion of the voice and feedback (and data) of residents ensured that the solution
was specifically catered to addressing the everyday needs and concerns of the local
population. Similarly, the Elements of AI MOOC was designed for accessibility from the
very start of the project, testing the project in a school and using inclusive language
and design for the interface. Involving the very people that are set to supposedly

151
 Adopt AI Study

benefit from the technology from the start ensured that these projects were not
expensive and unusable technological white elephants. Furthermore, participation
breeds understanding which together builds trust – something which is paramount for
public participation and AI technology to reach its potential.
Measuring success successfully
Similarly, Mercè also promoted the use of a new set of objectives and indicators for
urban planning documents which could be accessed and scrutinized by ordinary
citizens. As opposed to simply orthodox retrospective analysis, the Mercè project
allows citizens to use real-time data for forecasting future trends. Meanwhile, in the
health section as part of its award procedure, the Innovative Staters Award from
Innoviris explicitly used social and environmental criteria in 2021 and awarded funding
to winners with clear altruistic benefits. Beyond the technology itself, if the processes
in the sphere of monitoring and evaluation have a focus beyond the simply economic,
AI solutions should be constructed to also benefit humanity in other ways.
(Creative) collaboration
A very necessary element of the creation and operation of the Mobility Data Space in
Germany, the only of its kind in Europe, was the collaboration between multiple types
of stakeholders. The project had no fewer than six categories of actors including the
founder, managing body, government bodies, co-partners, clusters and of course the
data users and providers. This ground-breaking dataspace should be a model for
others, both in the mobility space, other geographies and other sectors. The project
demonstrated the diversity and balance of stakeholders needed to harness the power
of data. Furthermore, the meteoric growth of the Elements of AI MOOC was partly
driven by the positively opportunistic ways in which the organizers sought to
collaborate. Initially calling on counterparts in government departments of other
member states over social media, the project gained international status by
harnessing the Finnish Presidency of the Council of the European Union to vault the
status of the project with impressive results. Such is the complexity of AI procurement
and adoption that well-conceived cooperation between the private and public sectors
and different levels of government is paramount.

152
 Adopt AI Study

3 Chapter 3: Consultation of stakeholders’ analytical,

comparative report
3.1 Introduction
This chapter presents a comparative analysis of stakeholder consultations undertaken
as part of this study. The consultations included:
 Four sectoral workshops, dedicated to exploring the challenges and measures in
the uptake of AI by the public sector in the health, e-government,
mobility/transport and education sectors.
 The policy workshop dedicated to identifying and exploring policy solutions and
recommendations for the uptake of AI in the public sector.
 An online survey, dedicated to assessing the challenges and policy
recommendations in the uptake of AI in the public sector.
The more detailed methodologies for these consultations can be found in the annex.
The chapter proceeds by providing analyses of the workshops, the analysis of the
online survey, and the comparative analysis of all stakeholder consultations in the
conclusion.

3.2 Health workshop

The health workshop took place on 26 September (10:00-12:30 CET) via Microsoft
Teams, utilising the Mural board to conduct interactive exercises.
The workshop was attended by 19 participants in total, represented by 3 business
associations, 6 AI providers, 2 NGOs, 3 national public bodies, and 2 research
organisations.
The rest of this section provides an analysis of the results of the exercises undertaken
during the workshop and the discussions that followed.
3.2.1 Exercise 1: Challenges affecting the uptake of AI by the public
sector
In this exercise, participants were asked to write down challenges affecting the uptake
of AI by the public sector concerning procurement process, data, sectoral and AI-
specific technology, and organisational capacity challenges for industry and public
sector, and indicate which ones are the most significant.
Procurement process challenges
For industry, the most significant procurement process challenges are the lack of
acceptability of real-world data and evidence in regulatory assessment, the
lack of integration into the medical protocol, unclear/inconsistent pricing and
reimbursement guidelines, and the lack of experts in the European Medical Device
Regulation445, which is very relevant to procurement issues of AI in healthcare. Lack of
expertise and limited knowledge was also raised concerning the policy setters.
Other procurement process challenges affecting the industry are the lack of
openness and transparency in the procurement process. Lack of transparency in AI
predictive models due to weak clinical trials affects the ability to build trust among
stakeholders. There is also a lack of funding which is allocated to the procurement of
digital health solutions, including AI products, and reimbursement challenges. There is

445
More information available at:
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32017R0745

153
 Adopt AI Study

also the need to increase support for SMEs, including poor upfront engagement of
key stakeholders within the receiving organisation, eg. In showing value, addressing
concerns on risks, etc. Another major group of challenges relates to the lack of
clarity on regulation and the need for sectoral-specific ethical guidelines since there
is insufficient understanding of sector-specific applicability. There is an overall issue
with the definition of what is AI, and what is and is not in the scope. There is also a
problem of potentially differing regulations when dealing with multinationals and a lack
of regulatory spaces for experimenting with AI and its impact.
For the public sector, the most significant procurement process challenges are
inadequate national IT and human capital strategies. The public sector is not
sufficiently competitive to hire IT specialists, partly due to the lack of funding for IT
projects. There are also insufficient financial and human resources inside hospitals,
including IT specialists and data managers.
There is also a lack of regulation, contributing to the uncertainty as to legal regime
and implications of AI projects, especially in terms of who holds risks and the blurring
of responsibilities between several public actors, such as governments, national
health funds, regional health authorities, regional IT bodies, and hospitals. Other
challenges raised are variable procurement skills on AI projects within many public
sector organisations, lack of buy-in at Executive levels and a clear value
proposition/business case to organisations. Relatedly, there is a challenge of how to
promote the adoption of standards when there is not an immediate tangible benefit for
hospitals and healthcare providers.
During the discussion, one AI provider pointed to the lack of clarity as to who bears
the legal risks if something goes wrong. Clarification on legal responsibility is needed
so that health professionals have more confidence in accessing AI products.
Another AI provider underlined the challenge of dealing with multiple different people
with contradictory views, which creates complexity in the procurement process.
Several stakeholders remarked on the lack of acceptability of real-world data and
evidence in regulatory assessment and the need for interoperability. It is important to
allow citizens to extract all the data from healthcare institutions, and hospitals to
transfer to another. This involves getting regulators to agree on the quality of real-
world data.
Data challenges
For industry, the most significant data challenges are the lack of interoperability
and uniform data quality, which creates unclear, unharmonised rules for cross-
border data flows. Data quality is variable due to fragmented data architectures and
variable data standards (lack of common standards), patchy/incomplete data.
Significant effort must be made in cleaning data to make it fit for purpose
Consequently, the generation, sharing and access to representative datasets for
target populations, and access to high-quality datasets are challenging.
Another set of challenges concerns regulation, in particular, differentiated GDPR
implementation, concerns that regulation may require sharing of patient data at a
level which could compromise patient confidentiality or violate patient consent.
There is also a problem with the uncertainty of “data ownership”.
Finally, there are challenges concerning open AI Models to open black boxes and build
trust, bias in management and noise in the data, and ensuring a lack of bias in data,
and AI-derived from data.
For the public sector, the most significant data challenges concern data quality,
data interoperability, and the lack of standards and leadership. There is a need to
invest significant time upfront in stock-taking on data to be used and also understand
user concerns on data of a specific type. This is compounded by insufficient
resources. Another challenge is the fragmentation of IT systems between

154
 Adopt AI Study

providers and the lack of unified health data space. Consequently, access to data
remains difficult.
During the discussion, an AI provider noted that there is a lot of fragmentation in the
public sector, and fragmentation of budgets with projects developing in parallel. “Data
ownership” with the ability to create a synchronous view of the data for all healthcare
establishments – the federated data platform (FDP) of the NHS is an effort in that
direction.
A national public body noted that the national database in France has a single national
database, the quality of which is fairly good. The question is the resources needed to
exploit that data. Interoperability between the providers and lack of resources to
exploit the data are key issues. A lot can already be done with the existing data, but
there is a lack of resources to exploit the data. It is difficult and cumbersome to access
the data. It is harder to train and keep data specialists in healthcare establishments.
Another AI provider commented on data quality and that they are putting in huge
efforts to make data ‘regulatory grade’. A key challenge is to generate a reward
system to ensure investment in data quality. On patient confidentiality, there is a
dichotomy of views on sharing of medical data in the public sphere – privacy vs.
usefulness.
Sectoral and AI-specific technology challenges
For industry, one set of significant sectoral and AI-specific technology challenges is
around the issues of trust in the application of AI solutions in medical decision-
making. Building trust and education among practitioners and patients is needed.
Users and patients need to receive a better explanation of what AI solutions do.
There are also potential fears from health practitioners that AI solutions might replace
practitioners. Positioning AI as support for experts rather than a replacement for
experts is important. Thus, AI models for health should be transparent, openly
accessible and not black boxes. Health professionals should know how a prediction is
made. Relatedly, ensuring patient safety and privacy as ethical issues are
important.
Another set of challenges is related to the problem of skills. There is a lack of deep
tech talent in the health sector, and a gap in the skills required. Professional training is
needed to overcome it and to empower healthcare professionals to make informed
decisions concerning AI solutions. Finally, there are challenges concerning bias
concerning gender data, lack of diversity in data science teams, or potential for
discrimination based on AI results, for example in insurance policies. Thus, it is
important to engage in bias management and ensure that society’s current biases are
not hard-wired in AI solutions.
For the public sector, significant challenges concern the lack of training of health
professionals in the usage and adoption of AI solutions. There is also a lack of a clear
narrative on the value of AI for public services. This leads to misperceptions
around the role of AI, eg as a decision-support system. There is also a challenge
that the data used and produced can lead to biases, for example in health payment
decisions.
During the discussion, points around trust were prominent as there is a fear of AI
replacing healthcare professionals. One business association commented that there is
enough training for AI development, but a lack of training offered for the adoption of
AI. Need to convince professional training organisations to push for the adoption of AI
in organisations. Healthcare professionals also need the training to understand the
benefits of AI to adopt AI in the workplace.
A national public body pointed out that AI is a buzzword, but there is little evidence of
AI improving healthcare for patients and professionals in a real-world setting. The
buzz around AI is detrimental to AI itself as it loses its meaning. Better to focus the
use of AI on specific aspects, rather than aiming to have it everywhere.

155
 Adopt AI Study

An AI provider commented that healthcare professionals don’t see the iterative aspect
of developing an AI product, they see it as a finished product. They need to be
involved in the development and refinement of AI rather than having expectations.
A national public body opined that achieving AI in clinical settings is difficult. It just
signed a contract with Microsoft to establish a platform already working on two use
cases, extended soon to four. It is solving technical questions on how to use the data.
The scheme of the platform developed is a joint effort between clinicians and IT
professionals that would work best.
Several comments were made about the disconnect between the health sector and the
IT sector and expectations from the former.
Organisational capacity challenges
For industry, significant organisational capacity challenge revolves around the lack
of a skilled workforce. There is a need to increase education in the digital skills of
health practitioners and, conversely, data scientists in life science since currently there
is a lack of cross-sectoral and cross-organizational competition of qualified experts. It
is important to increase understanding in top management because currently there
is a lack of buy-in and little early engagement with clinical leaders, partly stemming
from the lack of understanding of digital/AI/data at senior levels and resulting in the
lack of integration of AI solutions into medical practice.
Other challenges included difficult access to healthcare markets for companies beyond
their country borders and big differences in pay between healthcare and provider
sectors.
For the public sector, the most significant challenges also revolve around skills and
expertise. There is a lack of IT staff and resources inside hospitals, and a lack of
hybrid profiles, e.g. clinical and data/AI experts. Thus, linking" HC professionals with
IT specialists are demanding, and time-consuming but essential. Also, leadership on
the topic is unevenly distributed among top management.
It is difficult to "divert" healthcare professionals from day-to-day clinical work to the
implementation of AI solutions. Overall, IT is still perceived as adding unnecessary
workload. In addition. "basic" software is often outdated (i.e Windows 98). There is
also the lack of a business case for AI solutions and value to patients.
During the discussion, several points were made concerning the barriers at the
managerial and technical levels, lack of skills, and limited interoperability. A national
public body commented on the need for hybrid profiles. The blending of expertise is
needed and it’s still very rare today.
One AI provider thought there is a need to do a better job of making the value case
for AI to secure engagement from health professionals.
A national public body suggested using cloud technology for hospitals to share data. A
European strategy with a unified cloud would be useful.
3.2.2 Exercise 2: Measures to increase the adoption of AI
In this exercise, participants were asked to write down measures, the most promising
ideas or steps in addressing the challenges discussed above, and indicate which ones
are the most significant.
Awareness of measures, the most promising ideas or steps
Industry solutions included having a broader adoption of federated learning (e.g.
through partnerships), creating an interoperability platform connected to input
layers and cognitive services, increasing the sharing of datasets from clinical

156
 Adopt AI Study

trials, generating and sharing AI principles and policies, and having controlled
sharing of data, such as Vivli446.
Public authority solutions included establishing European Health Data Space447,
measures in the New European Innovation Agenda448, in particular training 1
million deep tech talents, support for innovation procurements and regulatory
sandboxes. At the national level, potential solutions suggested were national funding
programs, the creation of national digital agencies, and digital healthcare acts.
Other measures suggested by participants were standardisation,
industry/academia/public partnerships, health data donor’s initiatives and
collaboration on the development of career pathways.
During the discussion, an AI provider suggested that partnership work with academic
groups, synergies of expertise, setting up projects with the NHS 449, and embedding
data scientists within the NHS. Hybrid roles help to bridge the gap between the IT and
healthcare sectors.
A national public body noted that education is very relevant, but there are not enough
PhD programmes at the EU level that combine medical and data science. Should
identify the leading hospitals in Europe with AI development strategies, they have the
highest possibility to involve people and access large data banks. In each country,
identifying one or two large hospitals and clustering them would be helpful. Should
identify hospitals that start defining the structure.
Another national public body commented that in France, the AI industry has not done
a good job of making the business case for clinicians. There is a digital ecosystem that
remains apart from healthcare operations, and that is not a viable solution because
many IT start-ups will not find clients. Little attention has been paid in policy to the
workings of healthcare facilities, therefore nothing will happen in terms of AI adoption.
it is important to understand how healthcare establishments work.
Suggestions for other measures
Other suggestions clustered around education, such as promoting degrees in
Computer Science and related disciplines as the first step to having AI professionals in
industry and academia, developing interdisciplinary PhD courses on AI/Healthcare, and
having education and awareness programs including a campaign to foster patients'
availability to donate data (patient's associations).
Steps related to regulation were mentioned, such as improving clarity on the scope
of regulation and avoiding duplication, establishing healthcare-specific sandboxes, and
addressing different scenarios. Acting on medical device approval for AI solutions in
healthcare and definition of accountability. Develop a version of GDPR for AI models.
Another set of suggestions revolves around public initiatives, such as having a public
register for CE-marked AI products in healthcare, optimising funding in support of AI
in the public sector to promote the spreading and scaling of reusable solutions,
reinforcing and advancing existing initiatives on open data and interoperability, and
promoting AI in the public sector in support of sustainability while developing

446
Vivli is an independent, non-profit organization that has developed a global data-sharing and analytics
platform. It’s focus is on sharing individual participant-level data from completed clinical trials to serve the
international research community, acting as a neutral broker between data contributor, data user and the
wider data sharing community. Further information is available at: https://vivli.org/
447
Further information on the European Health Space available at: https://health.ec.europa.eu/ehealth-
digital-health-and-care/european-health-data-space_en
448
Further information available at: https://research-and-innovation.ec.europa.eu/strategy/support-policy-
making/shaping-eu-research-and-innovation-policy/new-european-innovation-agenda_en
449
National Health Service (NHS). Further information available at: https://www.nhs.uk/

157
 Adopt AI Study

sustainable AI, in compliance with environmental principles, and leveraging civic

engagement and participation.
3.2.3 Exercise 3: Examples of cooperation between the public sector
and AI providers and success stories
In this exercise, participants were asked to write down cooperation best practices and
success stories, and indicate which ones are the most significant.
Cooperation best practices
Best practices include co-creation, federated learning on patient-level data across
cancer registries, collaboration with AI in health via conferences, work with
universities/schools (especially those with AI and health Masters), and public-private
partnerships. In terms of actions to be taken, participants stressed the collection and
addressing every stakeholder's viewpoint, involvement of bio-ethics committees,
understanding the goals, drivers and motivation of parties, fostering openness,
transparency, developing relationships over time and leveraging health and AI
communities networks, and linking hospitals with startups.
Success stories
Several success stories were mentioned, including EIT AI Community450, which
matches AI start-ups and healthcare organisations to solve specific health challenges,
the ATHENA451 partnership, the Alan Turing Institute – Roche strategic
partnership452, and the Salus Coop453.
During the discussion, an AI provider pointed to the challenge of data sharing with
legal implications. Also, the healthcare sector might find it challenging to articulate its
AI needs, there is complexity.
A business association commented that it has already supported a thousand start-ups
in Europe - 144 start-ups identified – and it is working closely with healthcare
providers to identify AI challenges.
A national public body noted that government, healthcare providers and the AI
industry all need to get together to make their priorities clear and articulate a plan.
Communities of practice should establish a dialogue between IT, legal and healthcare
stakeholders.
An AI provider stressed that the key to partnerships is to address bioethical questions
from the start.

450
More information available at: https://ai.eitcommunity.eu/#page-top
451
ATHENA (Augmenting THerapeutic Effectiveness through Novel Analytics) is a collaborative network
which brings together a unique, multidisciplinary and complementary partnership of academia, hospitals
and industry who explore and use the concept of machine learning for the realization of predictive analytics
in oncology. Further information available at: https://portal-uat.athenafederation.org/
452
The goal of this partnership is to establish a world-leading collaboration in advanced analytics between
Roche and the Turing, focused on enabling the transformative benefits of personalised healthcare to
become a reality for patients around the world. Publication of methods and algorithms will follow the
principles of open science to ensure that they are reproducible and interoperable. Further information
available at: https://www.turing.ac.uk/research/research-projects/alan-turing-institute-roche-strategic-
partnership
453
Salus is a cooperative that provides a platform for individual users to store their health data and control
the use of this data. Through their platform, Salus Coop aims to facilitate secure sharing of health data that
enable citizens to control their own health records while incentivizing data sharing to accelerate health
research innovation. Further information available at: https://datacollaboratives.org/cases/salus-coop.html

158
 Adopt AI Study

3.3 E-government workshop

The e-government workshop took place on the 27th of September (10:00-12:30 CET)
via Microsoft Teams, utilising the Mural board to conduct interactive exercises.
The workshop was attended by 18 participants in total, represented by 1 business
association, 2 AI providers, 7 national public bodies, 4 research organisations, and 2
international organisations.
The rest of this section provides an analysis of the results of the exercises undertaken
during the workshop and the discussions that followed.
3.3.1 Exercise 1: Challenges affecting the uptake of AI by the public
sector
In this exercise, participants were asked to write down challenges affecting the uptake
of AI by the public sector concerning procurement process, data, sectoral and AI-
specific technology, and organisational capacity challenges for industry and public
sector, and indicate which ones are the most significant.
Procurement process challenges
For industry, the most significant procurement process challenges clustered around
the unique features of public sector clients. This included the lack of scaling of AI
services beyond the specific context they are procured, as a research organisation
noted, AI companies often favour the private sector where they can sell one-size-fits-
all products, then approach public organisations as an afterthought rather than dealing
with the complexity of the public sector from the start. Additionally, there are often
unrealistic expectations from clients in terms of what the technology or service is
capable of doing and the promotion and selling of AI products that are not yet
required by the public sector. Finally, it was noted that many AI suppliers are startups
that could potentially disappear during a project lifecycle.
Additionally, industry stakeholders also identified that the market itself is fragmented
due to a lack of centralised processes and principles, often leading to disconnections
across governments, and between tech suppliers and buyers. Furthermore, the
procurement processes themselves are often lengthy and costly for the industry
leading to a lack of
For the public sector, the greatest procurement challenge identified was the lack of
clarity on the purpose behind the adoption of AI services. This was described by
multiple stakeholders as “AI for the sake of AI” and was discussed as being driven
by an overarching lack of awareness and understanding of how and why AI
services would be of benefit. It was of particular note that, at times, technology can
be experimental or in the early phases of development and there can be an incomplete
understanding of how to best integrate such services. One AI provider noted that
more work is needed to be done to ensure a common awareness of what AI can
deliver, whilst an International organisation detailed how the lack of clear use cases,
good practices, clarity and guidance on AI for the government was hindering
procurement efforts. Furthermore, a different International organisation discussed how
the lack of a problem-solving approach to assess technology needs was a key
challenge in ensuring the appropriate procurement of AI products.
Like other sectors, limited internal resources, the outsourcing of IT
infrastructure, and the lack of internal expertise and skills within the public
sector were noted to hinder the writing of technical procurement documents and the
managing of procurement processes. Should AI be successfully procured, it was noted
that a lack of flexibility in the execution of an AI project and that engagement
with the private sector was unnecessarily burdensome and at times difficult to
understand. Related to this it was highlighted that at times, several suppliers may be
required due to the focusses of their AI services being highly specific, in turn, this
increases the administrative and financial burden of engaging with such suppliers. It is

159
 Adopt AI Study

also difficult to fully understand the costs and benefits due to the nature of certain AI
technologies.
Legal issues were also outlined, primarily around the issue of data and IP
ownership to ensure that the public sector can avoid vendor lock-in as well as the
possible legal issues surrounding the data process where AI development within the
private sector lacks transparency.
Data challenges
For industry, the most significant challenge identified was the concern over how to
best determine whether source data itself was biased, whether it was sufficiently
cleaned, and how to ensure transparency and oversight when providing data
access. This was closely followed by questions about “data ownership”, data
management, and the lack of clarity as to where ownership resides in partnerships
between the private and public sectors.
It was also mentioned how the history of outsourcing within public services hinders
the data that is available meaning that it can be difficult to properly access the
necessary data that AI services would require.
A final group of challenges clustered around cultural issues. For industrial
stakeholders themselves, it was noted that there is often a lack of incentives to share
crucial data, particularly when sharing would require a significant administrative and
financial cost. It was acknowledged that there is at times a need to better sensitize
the private sector to the data culture and needs of the public sector. Finally, it was
noted that there is often a lack of trust between the private sector and citizens who
would prefer that private companies not involve themselves with the data of citizens.
For the public sector, two key challenges were identified. The first key challenge
focussed on the strictness of GDPR and the lack of training within the public
sector to ensure GDPR compliance. When data is available, there is a lack of
understanding within the system over what public sector data can be used.
The second key challenge focussed on the lack of the quantity and quality of training
data needed for AI services. Similar to industry stakeholders, public sector participants
noted that there was a clear lack of clarity around “data ownership”, as well as
around questions of data classification and confidentiality. In situations where
public sector organisations are permitted to share data, it was noted that there
remains a distinct unwillingness to share. One research organisation noted that
vagueness as to what data can be shared between the public and private sectors was
partially driving this unwillingness.
A few challenges were noted that clustered around skills and strategy. At a strategic
level, it was noted that only a few countries have a public sector data strategy,
making it challenging when working with countries where no such strategy and
associated processes exist, particularly when wanting to access private data and share
data between the public and private sectors. Consequently, it was viewed that most
governments fundamentally lack the required data governance structures and
processes needed to fully engage with AI providers and integrate AI services and
technologies. As one research organisation highlighted, a clear barrier here can often
be the different interpretations of data sharing and data protection found across
different administrations. Finally, one international organisation detailed the
importance of data leadership was discussed, and how the lack of clear data
leadership around best practices and good data usage was limiting data sharing,
particularly within lower levels of an organisation.
At a skills level, it was noted that there is often a lack of internal skills required to
assess the quality of data for AI projects, especially as this often needed to be done
on a case-by-case basis.
At a technical level, public sector participants noted that not all services are
digitised, which itself leads to a range of problems when discussing AI services.

160
 Adopt AI Study

When services are appropriately digitised, there is still a limited amount of

interoperable data structures that would facilitate data sharing and reduce the
associated costs of processing and cleaning data. Finally, it was noted that the public
sector often lacks the knowledge needed around bias mitigation. One AI provider
noted that the technology was out there to be used, but those fundamental issues of
interoperability and data standards were inhibiting progress.
Sectoral and AI-specific technology challenges
For industry, key challenges clustered around trust and transparency, produced by
“AI-washing”, for example, portraying AI chatbots as “human”, or not providing
accessible documentation.
Other challenges are a willingness to cut corners to go to market, selling AI
systems developed in one context to another one, adapting models to future
changes in data environments, and fulfilling requirements for accountability.
For the public sector, key challenges also orbit around the issue of trust, such as
the lack of trustworthy certification, and the lack of trust in results. Accountability is
also a challenge, especially related to algorithms, and due to a lack of ethical
frameworks. Due to public fear, an important challenge is an explainability. Thus, it
is important to underline that the use of AI is to assist, rather than replace, human
decision-making. Relatedly, it is important to avoid bias and have more benchmarking
and impact assessments of AI algorithms.
During the discussion, the AI provider commented that transparency and ethical
issues are well-known and covered extensively in the literature. Trust aspects require
action to be coordinated across the private and public sector side.
An international organisation suggested more and more governments are embracing
data intelligence, also to reinforce accountability. There is a learning curve that comes
with it, but intentions are there.
Organisational capacity challenges
For industry, the key challenges are the lack of co-responsibility about ethical
issues, reselling AI without adaptation and establishing a portfolio of projects
across public sector clients. Similar issues relate to stakeholder governance and
incentivisation along use cases, how to build needed competence/capacity in
government clients, lack of capacity working with public sector organisations and
difficulty in replicating private sector development to government.
For the public sector, there is a lack of knowledge and expertise in AI
possibilities in the organisation, including at the political level. This includes the lack
of strategic and tactical management. Often, the attitude of marketing quickly wins
rather than paving the way for profound service development. The public sector
sometimes forces to apply AI even if it is not necessary to solve their problems. There
is a challenge of retaining AI expertise in the public sector as in demand from the
private sector and the need for wider recruitment/talent reforms.
There is also a lack of resources to deal with data, infrastructure, and ethical
responsibilities of AI results, including the lack of resources to sustain implementation
beyond a pilot. There is a strong reliance on external expertise due to a lack of
internal AI experts. Also, it is difficult to gain a long-term perspective because
priorities change with political cycles. Teams are often created for an 'AI project' with
time-bound funding rather than teams funded to make the best choices about how to
solve a given problem.
During the discussion, a national public body stressed that it’s not just about AI skills,
but also about dealing with ethical challenges.

161
 Adopt AI Study

A research organisation noted that the use of AI regularly brings up capacity issues,
awareness, and competence issues. There is a large spectrum of new competencies
public servants would have to acquire.
Another research organisation commented that AI can be used to reduce costs and
increase efficiencies. However, this requires public organisations to have the required
capacity to do this in the first place. Then there are hiring costs associated with the
implementation of AI, for instance, recruiting data scientists.
3.3.2 Exercise 2: Measures to increase the adoption of AI
In this exercise, participants were asked to write down measures, the most promising
ideas or steps in addressing the challenges discussed above, and indicate which ones
are the most significant.
Awareness of measures, the most promising ideas or steps
Industry solutions include identifying areas where the same solution can be
used by many public administrations thereby reducing costs for individual
administrations. Other solutions would be establishing ethical risk management tools
in companies' enterprise architecture management (EAM), creating prototypes and
experiments with synthetic data, and feasibility studies, and sharing success
stories. It would be also a good idea to build networks for AI research and innovation
inviting public administrations and establishing an AI hub with an open-source
solution.
Public authority solutions include establishing AI and data competency centres
within public organizations, cooperation between the public sector and the R&D
sector, and innovation partnerships between public administrations and industry.
Another set of solutions included algorithmic assessment and transparency
standards, codes of practice, and the creation of national and EU frameworks and
strategies to build competency within public sector organisations.
Suggestions for other measures
Suggestions for other measures to promote public-private partnerships, build a
data/AI sandbox for member states, and build a community of practice among AI
leaders/practitioners. Provide a guide on problem identification and tech suitability
and how to build multi-disciplinary and diverse AI teams in governments, a practical
playbook on the challenges that surface and how countries have responded to them.
Introduce certification scheme(s) for AI.
Support organisation-level creation of data/AI strategies, including training
resources on AI for civil servants (not only technical). Changing EU funding
resources to accommodate more on implementation as well, including supporting the
creation of national/provincial government AI CoE hubs. Finally, it was suggested to
lead by example - show what AI solutions are used in EU agencies/institutions, help to
share success stories and provide an AI Capacity Assessment Tool/Research for
public administrations.
During the discussion, an AI provider commented that what accelerates the adoption
of AI is data-driven culture, it is about the intelligent use of data. Privacy-preserving
setup also encourages greater use of AI. There are communities of practice
developing, for sharing ideas on how to best integrate AI solutions.
An international organisation noted that AI culture needs to develop at the top civil
servant level in parallel to the lower and private levels. In the UK, there are digital
leaders among civil servants.
According to a national public body, there are a lot of testing and certification aspects
that can be quite heavy in the public sector. There is an issue defining AI and maturity
isn’t there yet.

162
 Adopt AI Study

One AI provider thought one solution is a general AI technology, like a chatbox

provided to municipalities, and the same setup can be reused across many
municipalities.
3.3.3 Exercise 3: Examples of cooperation between the public sector
and AI providers and success stories
In this exercise, participants were asked to write down cooperation best practices and
success stories, and indicate which ones are the most significant.
Cooperation best practices
Some of the best practices for the industry include data marketplaces with
prespecified ecosystem rules, joint shaping of an in-house AI Center of Excellence,
setting-up long-term upskilling workforce paths
Public sector best practices were the use of Universities and research centres for
feasibility and prototypes (trust) and industry for operational projects. Building strong,
competent teams internally for public sector bodies which initiate collaborations and
procurement. Chatbot in the Norwegian tax authority is a good example. 'GovTech'
funding streams/catalyst efforts to set public sector challenges and fund private sector
solutions. Rather than outsourced contracts, baking knowledge and skills transfer into
the procurement process to build public sector capacity from private sector suppliers.
Investments into augmenting competencies. Sharing AI knowledge between public
sector agencies of different countries in workshops and seminars
Success stories
Several success stories were mentioned, including chatbot / conversational AI in
Norway municipalities, a data observatory and Public Procurement Innovation
Directive in Chile, the Canadian government’s AI Source List for the promotion of
innovative procurement454, an annual update of the land cover map with AI in Walloon
public procurement455, a Programme in Data Science and Artificial Intelligence in
Public Administration in Portugal456, CitizenLab in Belgium457, AI registers in
Amsterdam, Helsinki and Nantes.
In the discussion AI provider commented that projects are a safe space for
governments to explore and discuss technologies, and for AI companies to engage
with current knowledge and potential public sector clients. Innovation projects involve
academia and the private and public sectors.
A national public body said that collaboration between civil servants and AI developers
works well from personal experience, with validity ensured at every step. Technical
specifications developed in collaboration with private companies being well guided to
coach civil servants to use the AI product. Establishing trust across the private and
public sectors is paramount.

454
Public Services and Procurement Canada (PSPC), together with the Treasury Board of Canada Secretariat
(TBS), held a procurement process to establish a list of suppliers who can provide the Government of
Canada with responsible and effective AI services, solutions and products. Federal Government
departments and agencies across Canada can use these pre-qualified suppliers to launch a streamlined
procurement process to obtain AI solutions. More information available at:
https://www.canada.ca/en/government/system/digital-government/digital-government-
innovations/responsible-use-ai/list-interested-artificial-intelligence-ai-suppliers.html
455
More information available at: https://www.eurisy.eu/stories/the-public-service-of-wallonia/
456
More information available at:
https://www.fct.pt/media/docs/Brochura_ResearchinDataScienceandAIappliedtoPA.pdf
457
CitizenLab is an online project and collaboration platform. More information available at:
https://www.citizenlab.co/about

163
 Adopt AI Study

3.4 Mobility (transport) workshop

The mobility (transport) workshop took place on the 3rd of October (10:00-12:30 CET)
via Microsoft Teams, utilising the Mural board to conduct interactive exercises.
The workshop was attended by 12 participants in total, represented by 6 AI providers,
2 local public bodies, 1 national public body, 2 research organisations and 1
international organisation.
The rest of this section provides an analysis of the results of the exercises undertaken
during the workshop and the discussions that followed.
3.4.1 Exercise 1: Challenges affecting the uptake of AI by the public
sector
In this exercise, participants were asked to write down challenges affecting the uptake
of AI by the public sector concerning procurement process, data, sectoral and AI-
specific technology, and organisational capacity challenges for industry and public
sector, and indicate which ones were the most significant.
Procurement process challenges
For industry, the continual emergence of start-ups and other new actors within the
sector often requires new supplier strategies and processes to be established resulting
in delays with procurement processes. Similarly, it was noted that there were limited
levels of international coordination with procurement processes.
For the public sector, the key procurement process challenge identified was the
lack of experience with Innovation Procurement. Often the procurement process
itself focuses on simply purchasing an AI service rather than being viewed as a longer-
term partnership between key stakeholders. The lack of a standardised legal
vocabulary related to AI, limited collaboration between public Data
Protection Offices (DPO) and the limited involvement of lawyers, DPOs, and
procurement experts at the start of AI-service procurement processes were all
identified as significant procurement process challenges. both identified as significant
challenges.
Data challenges
Two common challenges facing both industry and the public sector were identified, a
lack of clarity around “data ownership” and the lack of data management and
sharing infrastructure. Issues of “data ownership” are particularly prominent due to
the complex nature of an integrated supply chain where it is challenging to
specifically define and identify data owners and holders as data moves across multiple
actors and systems. It was noted that there is a need for the standardising of
“data ownership” within contracts to help mitigate this.
For the public sector, the most significant data challenge concerns obscurities and
obstacles in data transfer. It was noted that the transfer of both personal and non-
personal data had faced challenges due to a complex legal environment particularly
due to a lack of coherence between the EU and the US. It was noted that there
was a distinct EU dimension to this, in that even between municipalities and regions
there is a hesitancy to share data due to the complexity of the environment. It was
discussed that the lack of use cases within the public sector, more specifically, a
lack of readily available and accessible use cases, plays a key role in this, often
meaning that questions and conversations around the realities and benefits of data-
sharing remain abstracted. Finally, a lack of incentives for private actors to share
data, and the difficulties in determining the return on investment were noted as
inhibiting public-private initiatives and often resulting in investment decisions being
made in silos.

164
 Adopt AI Study

A final challenge was identified as the lack of investment into data integration
infrastructure driven by a lack of legal and financial clarity that leads to barriers
in the availability of data.
Sectoral and AI-specific technology challenges
Industry stakeholders identified two challenges, the first being the ‘Right to
Rectification and Erasure’ under GDPR, the second being a lack of effective
transparency of the algorithmic systems often underpinning most AI services and
technologies.
For the public sector, no single challenge was perceived as being greater than
another, but like industry, there was a clear and sustained focus on the risk of data
bias and a lack of transparency within algorithmic systems. Again, GDPR issues
were mentioned with a particular focus on the lack of clarity as to whether the data
processor role applies to those undertaking data analytics at a different point of the
supply chain. It was discussed that there is a need for more innovation projects that
combine datasets, for example, combining mobility and socio-economic data sets to
better understand patterns of travel and transportation use and need. Relatedly, it
was noted that as more projects of that kind are done, and as more systems and
databases are integrated into Mobility as a Service (MaaS) projects, there is a need to
better engage with ethical issues and questions of fairness. The lack of
internationally equivalent ethical principles was identified as proving a challenge
when looking across contexts, though it was noted that universal principles are
challenging if not impossible to establish given the relative cultural and political
context in different countries.
A local public body outlined that there was a need for the mobility sector to move
beyond semantic standards when it comes to MaaS and also looks at fostering public
trust. They further highlighted the lack of Open Standards for Linked Organisations
(OSLO) for MaaS, and that more was needed, for example, around issues of consent,
handling metadata, and accessing APIs.
Organisational capacity challenges
For the public sector, a key organisation challenge was identified as being the lack
of competence, experience, and skills, within the organisational workforce to
appropriately identify and make use of AI services and systems. Furthermore, it was
noted that often public sector bodies lack the necessary technical knowledge
when it comes to innovation procurement and that this is often outsourced.
Integrating AI services and technologies into existing city systems (for
example by allowing AI to take control of a system) was also discussed as a significant
challenge that is compounded by the lack of skills and expertise within the system to
appropriately identify where AI services are needed.
During the discussion it was outlined by a research organisation and a local public
body that the use of AI systems should always be a means but never an end in
itself, and that using such services can create short-term challenges if done abruptly
and without focus. Furthermore, it was noted that if skills and competency gaps aren’t
properly addressed that public sector bodies will struggle to keep pace with
technological developments.
3.4.2 Exercise 2: Measures to increase the adoption of AI
In this exercise, participants were asked to write down measures, the most promising
ideas or steps in addressing the challenges discussed above, and indicate which ones
are the most significant.
Awareness of measures, the most promising ideas or steps
Public authority solutions change the procurement process to include both pre-
commercial procurement and innovation procurement projects to better identify

165
 Adopt AI Study

and adopt developing technologies. Establishing an EU AI register was also noted as

one solution to help centralise use cases and foster trusted partnerships.
In terms of infrastructure changes, participants identified the need for data spaces,
identifying systems such as the federated data infrastructure Gaia-X458, and the
Swedish Logistics Data Lab mobility data space 459, as good working examples. There
was also discussion of the need for more open data initiatives with AI Sweden’s My
AI460 and VLOCA461, the Flemish Open City Architecture project identified as two good
examples in this space. It was also noted that there is a need to build more trust
among citizens with the Flemish Data Utility Company462 project being a good
example. Finally, it was discussed that there is a need to build more communities of
practice to help with knowledge sharing.
Suggestions for other measures
Other suggestions focussed on the need for greater flexibility with subsidies,
particularly for innovation procurement projects, a need for greater accuracy when
discussing AI services and data marked by moving away from rhetoric such as “data is
the new gold” and instead providing clear and accessible explainers to public sector
bodies regarding the prospect and value of AI services. Finally, in terms of wider
involvement with AI service projects, it was noted that the conversation needed to
include a wider range of voices both in terms of citizen participation, as well as
education level. This would move the conversation away from people holding high
levels of qualifications and help to stop the conversations from feeling exclusive.
3.4.3 Exercise 3: Examples of cooperation between the public sector
and AI providers and success stories
In this exercise, participants were asked to write down cooperation best practices and
success stories, and indicate which ones are the most significant.
Cooperation best practices
Best practices in cooperation focussed on ensuring clear arrangements and conditions
were in place for any AI procurement activity. Early and clear communication about
data access, data needs, and system challenges can help smooth the process. Greater
cooperation along the supply chain will also help with this by, for example, pre-
commercial procurement processes involving companies at an early stage to ensure
the service or technology is developing towards a clear goal.

3.5 Education workshop

The education workshop took place on 4 October (10:00-12:30 CET) via Microsoft
Teams, utilising the Mural board to conduct interactive exercises.
The workshop was attended by 12 participants in total, representing 4 public bodies (3
national, 1 regional), 2 NGOs, 2 European Digital Innovation Hubs and 4 AI providers.
The rest of this section provides an analysis of the results of the exercises undertaken
during the workshop and the discussions that followed.

458
Gaia-X is a federated and secure data infrastructure that is working to establish a data space ecosystem.
Further information is available at: https://gaia-x.eu/what-is-gaia-x/
459
Further information available at: https://ibg-sweden.se/2021/11/17/new-logistics-data-lab-creates-
better-opportunities-for-attaining-climate-goals/
460
Further information available at: https://my.ai.se
461
Further information available at: https://www.ugent.be/mict/en/research/projects/2020/vloca-flemish-
open-city-architecture
462
Further information available at: https://www.vlaanderen.be/digitaal-vlaanderen/het-vlaams-
datanutsbedrijf/the-flemish-data-utility-company

166
 Adopt AI Study

3.5.1 Exercise 1: Challenges affecting the uptake of AI by the public

sector
In this exercise, participants were asked to write down challenges affecting the uptake
of AI by the public sector concerning procurement process, data, sectoral and AI-
specific technology, and organisational capacity challenges for industry and public
sector, and indicate which ones are the most significant.
Procurement process challenges
For industry, the most important challenges are a general lack of legislative
frameworks, guidelines, and benchmarks for integrating AI solutions into public
service provision. There is a growing demand for the public sector to formulate more
clearly what kinds of solutions would be relevant, and what kind of data is going to be
made available to train and run these operations. AI is at the top of the pyramid,
technology SMEs need data sources, data collection, data engineering etc. before
AI applications can be implemented. If there is no available data, the chances of
starting an AI project are low. A further problem is that SMEs competing for public
contracts do not have a clear idea of the challenges facing work processes in the
public sector. They struggle to demonstrate a clear “return on investment” since
some operations seem opaque. There also seem to be concerns about “procurement
theatre”, long-winded public procedures and press releases, instead of focusing on
the best solution and operationalising it fast.
For the public sector, the most significant challenges are a lack of digital skills
and expertise among public buyers to understand the technology market and its
latest products. This leads to technical specifications that are not always at the
cutting edge of technological research. AI providers want performance expectations
to be more clearly expressed. A related issue is the absence of unified guidance
and quality standards to assess AI solutions. There should be national legislation on
the use of high-risk AI, especially when the target audience of AI solutions for
education is minors. If, however, a specific AI solution is rolled out at a large scale,
the “freedom of education” of each learner should be respected, for example, by
always providing at least one alternative.
During the discussion, The most frequent comment was that there is still a lack of
knowledge about AI and its potential uses in education. This is partly because school
leaders and educators, often trained before the digital revolution in education, lack
digital skills and knowledge to use AI solutions, “to make safe use of AI solutions”.
The safety of minors is a great concern, but solutions exist and educational
stakeholders should receive more information and training about them.
The case for AI in education is yet to be made. As the representative of an EU-wide
educational NGO put it, “What is AI good for in the institutions?”. Education and
training institutions care about raising the quality of education, and “AI might not be
the solution.” Such challenges regarding the educational content, quality and
relevance are not necessarily addressed through technologically supported efficiency
gains.
Data challenges
For industry representatives, the most important challenges concern the quality
of data, sensitive data and “data ownership”. Different data models and a strong
belief in “old-fashioned structured data” lead to scalability problems in national
education systems. The approach of “data minimalism” is preferred. Since the
education sector’s main data users and data producers are minors, access to data is
often very complicated. Industry representatives asked for clearer definitions of
“sensitive data” and ethical frameworks outlining what can be done with it. The
most important challenge for the industry is “data ownership”: if the data is produced
by minors and used in a state-funded education system, are the data owners the
learners, their parents, the educators, the state institution, the education ministry or

167
 Adopt AI Study

the private AI provider? What are the “data ownership” options for public-private
partnerships?
From the public sector perspective, the biggest challenges are the lack of data
interoperability and the uneven implementation of EU regulation (esp. GDPR)
across Member States. The interoperability challenge affects data transfers and data
security. One proposal was to create better data spaces accompanied by
“national/supranational data governance frameworks”. The second challenge concerns
ethical issues in data collection (esp. for minors) and accountability for AI
providers. A more unified interpretation of the GDPR would offer clarity and a safer
space for all education stakeholders.
During the discussion, AI providers stressed that ownership of and access to data is a
big issue. Access to data can be constrained by the GDPR, yet some data is
strategically important to have for the sector. Private companies process the data and
the data owner (e.g. educational institution) then has to buy it back.
Common challenges include a lack of interoperability and a lack of infrastructure,
which might also stem from the limited digital literacy of some users.
Sectoral and AI-specific technology challenges
For industry representatives, the main challenges revolve around collecting and
processing data from underage users. At the moment, it is difficult to feed AI
models with sufficient data (big data). Only "big tech" has enough data for certain
solutions. Smaller providers would appreciate “open data sets” to train their AI.
Regardless of company size, the problem of underage consent and transparency is
striking. “Do students know what is done with their data?” was the most popular
comment overall. A second sector-specific challenge from the perspective of AI
providers is the entrenched infrastructure, hierarchy and bureaucracy of
educational institutions. Providers of AI solutions do not always know who to talk to
sell their solutions.
Among the public sector notes, the most popular statement was “First, you have to
know the goal, second: the data that we have, third: algorithm selection. Too often
started with AI selection”. This overlaps with the abovementioned comments about
the lack of a guiding framework, which sets out aims and basic principles for
introducing AI in education. It is also an important point about public procurement
procedures – which should have transparent goals – and public buyers, who need to
be able to offer providers reliable data.
Another point discussed in detail was equity. Equality is a core principle of the EU and
education systems are supposed to promote it, but AI-powered tools are not
accessible to all learners and educational institutions. This basic problem of availability
and access to high-tech solutions, coupled with algorithms automatically adapting to
learner profiles, may enlarge pre-existing social differences. Transparency about what
data should be collected (e.g. are data on socio-economic background acceptable?)
should
The final discussion concerned the question: “What are the guiding values of
universities?”. Should they be focused on efficient operations and effective
preparation for the labour market, or rather allow some space for flexible,
serendipitous learning guided by learners’ interests? Some participants agreed that
the efficiency gains from AI products (esp. for institutional governance) should not
override the process of free, inquisitive discovery.
During the discussion, the representative of a transnational education NGO said that
there were outstanding questions about AI being the most appropriate tool for staff
training, improving education and other values that are dear to workers in the
education sector.

168
 Adopt AI Study

For educators, the arrival of AI also poses a personal challenge (going beyond the
technical challenge), since they feel like their status and authority as sources of
knowledge and learning are under threat.
A representative of a higher education institution added that there are cultural
differences across Europe when it comes to the adoption of new technologies for
operations or teaching, with concerns about learner privacy more marked in some
countries than others.
The majority of participants agree that (views on) AI take up vary greatly between
educational institutions and the AI industry.
Organisational capacity challenges
For industry representatives, the biggest problem is that most developers have a
limited understanding of pedagogical principles and the education sector. There
is “no real customer orientation – tech comes first”. Sometimes this “tech first”
principle is reinforced by the perceived need of selling “trendy stuff”. The COVID-19
pandemic greatly accelerated the demand for good Edtech.
Nonetheless, regulation (comparatively strict for education, due to underage users)
could inhibit SME capabilities. Innovative start-ups are struggling to get long-term.
The interest in “open source” material for educational purposes generates further
uncertainty about the sustainability of the developer community.
From a public sector perspective, the most important request is a real discussion
on “what kind of future we want”. Teachers' beliefs on "what learning/teaching is" are
challenged by AI at a fundamental level. Teachers follow pedagogical traditions and
will need to improve their digital skills and literacy. This includes the interpretation
of AI output. “Do we need to change our curricula in addition to teaching style?” was a
question posed by several participants. More guidance on the didactical
use/integration of AI would be appreciated.
During the discussion, it was commented that a political discussion needs to precede
the technological discussion, as to what kind of teaching society wants, whether AI
can improve learning outcomes, and on a bigger scale what is the role of education in
the data economy. Most questions are political rather than of a technical nature.
Discussion on the European approach to learning and teaching and the place of digital
tools in that approach is important. Determining the right purposes for using AI
solutions is crucial to ensure the right level of adoption of AI technology.
There may be a misunderstanding of the challenges faced by the education sector
among AI providers, in that they see them from a business perspective. The overall
impression is a disconnect between AI providers and the education sector.
3.5.2 Exercise 2: Measures to increase the adoption of AI
In this exercise, participants were asked to write down measures, the most promising
ideas or steps in addressing the challenges discussed above, and indicate which ones
are the most significant.
Awareness of measures, the most promising ideas or steps
Solutions provided by the industry include public-private partnerships, (e.g.
Research partnerships between EdTech & universities), open innovation to increase
trust, and better dialogue with the education sector – for example by linking the
development of new tech products with training courses for educators.
Solutions provided by public authorities include better risk management with
clear policies and executable procedures - this will lead to higher consistency and
less uncertainty. Better and continuous professional training for educators who
have to keep up (peer learning is particularly good for the teaching community) is
needed. It is important to stimulate discussion about the role of teachers in a
changing digital society. These solutions should also be factored into budgeting and

169
 Adopt AI Study

the institutional and national level. Detailed standards for the development and
implementation of EdTech solutions, coupled with guidelines and benchmarks
would be much appreciated.
Suggestions for other measures
Most additional suggestions revolved around the themes of transparency and trust.
Participants requested “open data sets”, a large-scale “awareness-raising campaign”
and “Principles for fair use of AI in education” ensuring a “human-centric approach” in
education.
Measures to increase digital literacy were also considered very important. Peer
learning could be organised across Member States (e.g., for GDPR implementation) to
ensure that benchmarks & standards can be reached and upheld. This would allow
some degree of standardization of AI in Education.
During the discussion, the representative of an industry association recognised that
documentation around AI and its use, how it is developing etc. is important for the
education sector. There are good initiatives led by the government on those aspects.
Achieving a more centralised understanding and providing a framework for
understanding the technology (e.g. cloud infrastructure, esp. during the COVID-19
pandemic) is something that can be replicated for AI. There is a lack of trust in
technology within the education sector, which is sensed by the AI industry. It is
important to understand in what scenarios the AI industry and education sector can
work together. Public-private partnerships could help. At the moment there is a fear
among teachers that “robots will take over”.
The industry supports a practical approach to making sure regulations do not hamper
technological development. The EU’s draft AI regulations may be too constraining to
ensure AI take-up. AI in education is still in its infancy, having openly available large
datasets to build AI tools would be useful.
Partnership work can bring about better outcomes when training education sector
workers. Different education sectors often work in silos. Different providers need to
work together.
The first step is to establish communities of practice to make sure all stakeholder
groups are on the same page, i.e. education workers, policymakers, and AI
developers.
3.5.3 Exercise 3: Examples of cooperation between the public sector
and AI providers and success stories
In this exercise, participants were asked to write down cooperation best practices and
success stories, and indicate which ones are the most significant.
Cooperation best practices
Best practices highlighted by the public sector include the production of national AI
strategies (e.g. FR463), often with policies to increase the number of AI experts;
guidelines for Data Governance in education (NL464 & CH); Bodies that gather
practices and develop policies (CNIL,465 Hochschulforum Digitalisierung466, SURF,
JISC), and roadmaps for implementing AI in the education system.

463
More information available at: https://www.economie.gouv.fr/strategie-nationale-intelligence-
artificielle
464
More information available at:
https://www.nldigitalgovernment.nl/wp-content/uploads/sites/11/2019/04/data-agenda-government.pdf
465
More information available at: https://www.cnil.fr/
466
More information available at: https://hochschulforumdigitalisierung.de/

170
 Adopt AI Study

Success stories
Industry provided 3 examples from FI: The Association of Finnish Municipalities partly
funded the development of the “Annie Advisor”467 chatbot with a condition that there
would be an open-source version; my data skills transfer between Finnish DF & higher
education; Technology Industries Finland (employer union), relevant labour unions
and higher education: personalized learning paths to find a job or a better job via the
right training468.
The public sector provided examples of AMAI!-project (Flanders)469: cooperation
between public and industry, Smart Education @Schools projects (Flanders)470, City of
Tampere: optimising formal and corporate training to future Labour market demand
(including the use of synthetic data), Metropolia UAS (Helsinki): Curriculum to meet
SDGs horizontally at a course level471.
During the discussion, it was suggested that the government and industry have to
work together to further develop academic expertise and capacity in AI. Teachers and
students also need to be involved in any cooperation process.
The development of a skills portfolio in army training can be transferred to the
academic context. If all the players in the ecosystem know their value, cooperation
can work.
There is potential for bottom-up cooperation, from teachers’ needs to the development
of innovations. App developers can have a role to play here. The offer of open-source
tools can foster cooperation.
3.6 Policy workshop
The Policy workshop took place on 17 October (10:30-13:00 CET) via Microsoft
Teams, utilising the Mural board to conduct interactive exercises.
The workshop was attended by 29 participants in total, represented by 1 business
association, 5 AI providers, 9 national public bodies, 7 research organisations, 2
international organisations, 1 regional public body and 1 NGO.
The rest of this section provides an analysis of the results of the exercises undertaken
during the workshop and the discussions that followed.
3.6.1 Exercise 1: Identifying solutions to challenges affecting the
uptake of AI by the public sector
In this exercise, participants were asked to write down solutions to challenges
affecting the uptake of AI by the public sector concerning procurement process, data,
sectoral and AI-specific technology, and organisational capacity challenges.
Solutions to procurement process challenges
Participants were asked to provide solutions to four of the most important challenges
identified in the sectoral workshops:
Challenge 1: Lack of clarity on regulation (particularly around high-risk AI).
Solutions provided by participants:
 Provide clarity as to how EU legislation and regulatory frameworks interact
together and address legal misalignment and incoherence between Artificial

467
More information available at: https://www.annieadvisor.com/
468
More information available at: https://finland.fi/business-innovation/combining-finnish-educational-
expertise-and-artificial-intelligence/
469
More information available at: https://amai.vlaanderen/
470
More information available at: https://onderwijs.vlaanderen.be/nl/dien-je-idee-en-project-in-bij-smart-
education-schools-najaar-2022
471
More information available at: https://www.metropolia.fi/en/rdi/rdi-projects/sdg4biz-project

171
 Adopt AI Study

Intelligence Act, GDPR, Data Governance Act, Data Act, and AI Liability
Directive. Prepare a basis for guidance on the reconciliation of the regulatory
frameworks. Run “war games” to see how use cases may interplay across
regulatory frameworks.
 Provide a clear definition of what is AI and clarity on the scope of what AI
regulations cover to understand if this will impact any already ongoing activities
that we may not think of as AI.
 Introduce certification of AI on different levels and Production Planning.
 Introduce risk classification under AI Act and sectoral regulations.
 Ensure that all sectors engage with policymakers for the AI Act to ensure the
regulation meets the needs of all types of AI uses and that all key stakeholders
are working together to shape fit-for-purpose EU legislation for AI and data.
 Provide Impact Assessment of AI/ADM implementation in public sector services.
 Monitor developing AI regulations, principles and wider experience where
legislation has been enacted, for example, in China.
 Organise inter-organizational discussions on AI/data sharing regulation, for
example, between education ministries and the data protection commissioner.
 Promote awareness about possible exemptions from AI Act (e.g. in law
enforcement) for public bodies when it comes to implementation.
 Set up regulatory sandboxes on AI to promote public experimentation.
 Use of innovation-friendly procurement modes.
Challenge 2: Lack of centralised procurement strategy and lack of experience
with innovation procurement. Solutions provided by participants:
 Identify 'owners' for procurement strategy and their needs, to ensure that AI
solution can solve their problems.
 Reinforce ideas through other strategy documents and activities if no strategy
is on the horizon.
 Consider the procurement power of larger-scale organisations and assess what
the processes and needs are.
 Provide an inventory of best practices from similar entities.
 Encourage open source solutions, not to be locked in proprietary solutions from
the US or Chinese companies.
 Cybersecurity standards should be taken into account, given the critical nature
of infrastructure in many cases.
 Promote interdisciplinary collaboration between the public sector and other
stakeholders.
 Ensure procurement specialists are versed in the technology and flow.
Challenge 3: Lack of expertise among public buyers to articulate needs and
understand the purpose behind adoption. Solutions provided by participants:
 Avoid the "AI-first" approach and focus on clear problem framing, on the
specific improvements that an AI-based solution would bring.
 Introduce national points of contact for practical advice, Digital Innovation Hubs
for consulting, or competence centres for aiding public bodies to adopt/set up
fitting solutions.

172
 Adopt AI Study

 Educate and raise awareness of the potential and limitations of AI, and show
what AI interventions can achieve with real-world success stories. Education
should not only be in"digital literacy" but also on the basic theories
underpinning Computer Science and AI.
 Provide EU funds and support for implementing relevant training at the national
and local levels.
 Bring together procurement specialists and technical experts throughout the
design/delivery process.
Challenge 4: Lack of alignment between industry and public sector
expectations. Solutions provided by participants:
 Create regulatory sandboxes.
 Introduce “best practice” dialogues between industry, the public sector and
academia, and “roadshow” type events for building networks and shared goals
and understanding.
 Conduct internal research within public organisations to understand what
expectations and ideas on AI public employees have.
 Include the expectations of citizens and consumers because they are as
important as what public bodies want - the user journey should start with end
users of services.
Solutions to data challenges
Participants were asked to provide solutions to four of the most important challenges
identified in the sectoral workshops:
Challenge 1: Lack of clarity and harmonisation around cross-border data
flows. Solutions provided by participants:
 Formulate clear incentives and use cases for data sharing.
 Promote open application programming interfaces (APIs).
 Develop more federated data networks.
 The sector-specific data spaces, such as the European Health Data Space,
should bring clarity on data flows per sector.
 Clear rules on the level of de-identification connected to cross-border data
sharing, and clarity on applicability/use of GAIA-X/clouds.
Challenge 2: Lack of interoperability. Solutions provided by participants:
 Develop standards for data exchange including principles for metadata creation.
Develop “future-proof” standards that will easily adapt to new data
types/sources. It is important to use and align all the many existing standards
and not create even more new ones.
 Develop national strategies for data complemented by organisation/sector-
specific strategies for data that take a wide-angled approach to what’s involved
(leadership, skills, regulation, architecture, infrastructure).
 Make it a law that data format for anyone doing business in the EU should be
open so that anyone could build interoperable software. Or introduce incentives
for creating an interoperable system.
 Introduce compulsory documentation for semantic interoperability (data
dictionaries, ontologies, etc.).
Challenge 3: Lack of clarity on “data ownership”. Solutions provided by
participants:

173
 Adopt AI Study

 Define data control, access and other processes in relevant legislation - GDPR,
Data Governance Act and the Data Act.
 Increase the use of decentralized data architectures.
 Introduce dedicated system transparency conditions in agreements with
AI/ADM system contractors.
 Make it a special issue in certification schemes.
 Engage in public discussion, education and seminars around “ownership” and
usage of individuals' data to promote greater understanding by individuals of
how their data may be used.
 Encourage donation of data sets.
Challenge 4: Challenges surrounding the sharing of data between and across
sectors. Solutions provided by participants:
 Introduce guidelines as to what is and isn't legally possible when it comes to
sharing data cross-sectorally, and guidance in structuring data regarding
sensitivity.
 Consider GDPR compliance from the ground up.
 Encourage sharing of data between different data spaces, for example, in health
or agriculture.
 Promote common sharing, format and metadata standards.
 Promote cross-sectoral teamwork in the early stages of AI adoption for mutual
understanding of needs and concerns.
 Ensure that confidential data (business secrets) are end-to-end encrypted and
promote privacy-preserving techniques.
 Build the tools that help give users visibility and control over their data and
associated consents to see who and for what purpose it’s being used in practical
terms.
Solutions to technology challenges
Participants were asked to provide solutions to four of the most important challenges
identified in the sectoral workshops:
Challenge 1: Managing the potential for bias within data sources and data
science teams. Solutions provided by participants:
 Ensure ethics by design - interdisciplinary awareness/competence in
developing/procuring teams.
 Recognise that bias will exist in datasets and develop tools/promote education
to make understanding bias a standard part of any AI workflow. Actively
promote mitigating bias in developing AI as an area for research.
 Ensure representation in data science teams.
 Apply explainable algorithms used in AI/ADM systems in the public sector.
 Have clear requirements for training, testing and validation of data sets.
 Use diverse fictional cases to test AI outcomes. EU AI Act requirements should
include bias avoidance, this can be tested/audited before market placement as
well as ongoing for learning AI systems.
 Have consistent monitoring and evaluation of bias.

174
 Adopt AI Study

Challenge 2: Limited technological accountability and transparency. Solutions

provided by participants:
 Use of explainability tools along EU guidelines. Explainable AI (XAI)
requirements would help with accountability and transparency.
 Certification of AI on all assurance levels.
 Deeper user studies before designing interfaces.
 Clear requirements for training, testing and validation of datasets.
 Conformity assessment of transparency requirements, ideally via independent
third parties for high-risk areas.
Challenge 3: Misunderstandings of, and a lack of trust in, using AI solutions
(fear of the impact of new technology). Solutions provided by participants:
 Use storytelling to communicate the value and the worth, but also manage
unintended consequences and fears.
 Testing, inspection, and certification (e.g. by TIC Council companies and
notified bodies), especially in high-risk areas, will boost trust. Signal via tested,
certified “Made in Europe” label to enable trust needed for uptake.
 Clear regulation and rules.
 Greater education on AI at all levels of society. Promote education on the basics
of AI (regression/classification) to demystify it from primary school.
 Position the AI Act and AI Liability Directive as a way to build more trust in the
use of AI.
 Promote the adoption of ethical frameworks for AI.
 A position most AI systems will include a human in the loop to ensure safety.
There needs to be trusted not just in a single AI product, but also processes
and persons underpinning it.
 Making Algorithmic Impact Assessments standard practice and easy to find and
read.
Challenge 4: Difficulty in integrating new AI technologies/services into
existing systems (education/health/transport). Solutions provided by
participants:
 Aim for a longer cycle of integration because this happens with any new IT.
 Address security considerations by undertaking risk-adequate controlled
experiments via new solutions.
 Use "data ambassador" campaigns to better involve stakeholders.
 Work alongside existing systems as a support rather than immediately replacing
existing systems.
Solutions to organisational capacity challenges
Participants were asked to provide solutions to four of the most important challenges
identified in the sectoral workshops:
Challenge 1: Focus often on innovation and developing new solutions rather
than long-term implementation. Solutions provided by participants:
 Fund multi-disciplinary teams, not projects.
 Avoid “project silos” and reinvent the wheel each time, a more modular longer-
term approach is advisable. The overarching purpose should be not a “shiny

175
 Adopt AI Study

project” but a long-term solution for citizens, real-world solutions with a service
mentality.
 Promote, and invest in a facilitate (e.g. via legislation on data sharing) open-
source solutions and libraries.
 Make policy with input from representative groups of users.
 Develop reward/payment structures to promote long-term behaviour.
Challenge 2: Lack of central resources to help understand the role of AI and
whether AI services are the best solution. Solutions provided by participants:
 Incentivise, emphasise and prioritise service design over tech implementation.
Iterative design thinking should be emphasised.
 Better education in AI is needed to understand which techniques to use for a
particular case study, and more importantly, when not to use them and rely on
a simpler solution.
 Use central governance structures like business cases to challenge the
assumptions and be clear on the value of what’s being done.
 Establish cascading funding scheme for AI.
Challenge 3: Lack of skills, competencies, and capacity to understand and
utilise new AI technologies/services. Solutions provided by participants:
 Focus on social needs and less on technologies.
 Introduce digital Academy-type models for the public sector where AI is part of
a balanced curriculum about 21st-century digital era public service design and
delivery. Promote AI further education not just to understand AI basics but in
change management in terms of how to introduce solutions within and across
the organization.
 Ensure AI and tech are a key part of schooling for children.
 Build experimentation labs (controlled environments) for young adults.
 Advertise successful examples and best practices.
 Increase digital literacy through the establishment of communities of practice,
or peer learning groups across Member States.
Challenge 4: Lack of established relationships with private AI providers.
Solutions provided by participants:
 Public adoption should be safe, not become the testing ground for private AI
providers.
 Target SMEs and start-ups less known than big tech companies.
 Open calls targeted at University excellence centres and public interest
initiatives.
3.6.2 Exercise 2: Exploring solutions and opportunities for overcoming
barriers
In this exercise, participants were divided into three groups. For each group, the most
significant solutions from the previous exercise were selected. The participants in
groups discussed why each solution is important, who should implement it, what
should be the role of EC, who would benefit and how, and what could be the barriers
to their adoption.
Solution: Training for public buyers

176
 Adopt AI Study

This solution is important in raising the quality of public services, because of the
possible lock-in effects of (also wrong) purchasing decisions, and because smart
choices matter and should be taken based on competent assessments.
The solution should be implemented by professional further education bodies, ideally
offering certified further training in this domain, public institutions in the Member
States and National Institutes.
Europen Commission’s role should be in establishing and maintaining a pool of
resources (e.g. ISA2 for interoperability), producing guidelines for public AI
procurement, awareness-raising, best practice sharing (so mistakes aren’t made each
time again), quality benchmarks, providing a framework of legal and ethical
guidelines.
Beneficiaries of this solution will be everybody really, those taking decisions and those
benefitting (or not being harmed by) them at a later stage once solutions are
implemented. The buyers – they can do their job better. The users – work with a good
system. The sellers – interacting with someone with the right skills.
The potential barriers to this solution are legacy, lock-in effects from previous tools in
use (may need to be disrupted), issues of interoperability when each buyer uses a
different AI tech, and lack of funding to maintain and support advanced AI applications
where constant data collection is needed.
Solution: “data ownership” certification
This solution is important because certification forms a basis for trust in AI solutions,
increasing uptake and willingness to share data.
This solution should be implemented by independent third parties and conformity
assessment bodies.
European Commission’s role should be in ensuring that certified solutions are
chosen/recommended and provide guidelines, and frameworks for thinking about
these complex issues.
Beneficiaries of this solution will be citizens, who will know who is accountable,
governments, who can implement effective data governance, and AI developers, who
will have clarity on their freedom to operate.
The potential barriers to this solution are existing data collection systems in the public
sector in which the same data are collected by different agencies.
Solution: Ethics by design
This solution is important because it builds protections for individuals in AI solutions,
safeguards and prevention of early-on bias and discrimination, and raises awareness
of the risks and pitfalls of AI.
This solution should be implemented by development teams together with the owner.
European Commission’s role should be in providing training, encouraging research,
and providing ethical standards for AI technological development.
The potential barriers to this solution are acknowledging direct and indirect biases.
Solution: Regulatory sandboxes
This solution is important because there are multiple ways to create AI solutions, they
must be carefully classified and can be potentially disruptive to society.
This solution should be implemented by the European Commission together with
Member States, or by subcontracting to an EU company to avoid conflicts of interest.
European Commission’s role should be to facilitate the process and provide templates,
and "In-scope/Out-of-scope" guidance, to make sure a European set of solutions
emerges.

177
 Adopt AI Study

Beneficiaries of this solution will be start-ups in need of scaling innovation, users and
AI technology providers.
Potential barriers to this solution are the time it takes to set them up, the potential of
“dual-use” and user suspicion.
Solution: Standards for data exchange / Principles for metadata creation
This solution is important because it is a key to safe AI due to certification which lays
on standards, crucial to the end-to-end value of information exchange and creates a
consistent layer of data quality.
This solution should be implemented by producers of AI and some standard
organisations such as ISO or ECMA.
European Commission’s role should be in the production of AI certification schemes
and interacting with international organizations in creating data standards.
Beneficiaries of this solution will be users who would have confidence in AI and
producers, so they could focus their efforts.
Potential barriers are lobbying by private companies trying to push their solutions, the
complexity of the process, lack of know-how and skills and Change Management in
organizations tempted to go the "safest" way.
Solution: Story-telling to communicate the value
This solution is important in providing clear information for a broader society, building
trustworthy AI and avoiding repeating mistakes.
This solution should be implemented by local authorities, responsible for AI and GDPR
implementation.
European Commission’s role should be to establish the framework and support local
authorities.
Beneficiaries of this solution will be citizens and enterprises, public servants and
employees that need to apply and use AI technologies, and start-ups that are part of
the system.
Potential barriers are a lack of proper strategy, guidance for communication of AI, the
threat of a "technological elite" and information sensitivity.
Solution: Avoid “project silos” / Modular longer-term approach
This solution is important to reduce the gap between various stakeholders and
increase the longevity of AI solutions.
This solution should be implemented by various decision-makers across the landscape,
including Chief Data Officers with the bigger picture.
European Commission’s role should be to provide a policy-driven push for "free flow",
and require companies to provide open APIs to their services so that services can be
composed by third parties.
Beneficiaries of this solution should be EU SMEs providing advanced services based on
publicly available building blocks (APIs), and small projects within departments that
are not yet plugged into the overall picture (COVID showcased critical needs in this
regard).
Potential barriers are time and the lack of wide knowledge.

178
 Adopt AI Study

3.7 Online survey

The online survey was launched on the 22nd of June 2022 and closed on the 5th of
September 2022. More details on the methodology can be found in the Annex.
In total, the survey received 77 complete responses. Most responses came from
research organisations (18 responses), academic/independent experts (12 responses),
national public bodies (8 responses), European Digital Innovation Hubs (8 responses),
enterprises (8 responses), regional public bodies (7 responses) and local public bodies
(6 responses).
Figure 68 Respondents’ affiliation

Research organisation 18
Academic/Independent expert 12
National public body 8
European Digital Innovation Hub 8
Enterprise 8
Regional public body 7
Other 6
Local public body 6
European Institution 2
Public service operator/ Publicly owned… 1
Enterprise association 1

Source: Online survey, N=77.

Among those identifying as “Other” included an interest organization, several NGOs, a

technology consultant and a research organisation.
In terms of country distribution, most respondents came from Italy (8), Germany (7),
Spain (7), Sweden (6), Denmark (5) and Bulgaria (5).

179
 Adopt AI Study

Figure 69 Respondents’ country of origin

Italy 8
Germany 7
Spain 7
Sweden 6
Denmark 5
Bulgaria 5
Belgium 4
Poland 3
Austria 3
Romania 2
Portugal 2
Ireland 2
Greece 2
France 2
Croatia 2
Slovakia 1
Netherlands 1
Lithuania 1
Latvia 1
Hungary 1
Finland 1
Czech Republic 1

Source: Online survey, N=77.

Respondents outside the EU came from the United Kingdom (3), Canada (1), Mexico
(1), Montenegro (1), Scotland (1), Serbia (1), Switzerland (1), and the United States
of America (1).
Among the 8 enterprises responding to the online survey, 3 are micro-enterprises472, 2
are small enterprises473 and 3 are large enterprises474.
Most of the respondents belong to the education sector (29), followed by e-
government (22), health (19) and mobility/transport (16).

472
Staff headcount is less than 10; Turnover is less/equal EUR 2 million or Balance sheet total is less/equal
than EUR 2 million
473
Staff headcount is less than 50; Turnover is less/equal EUR 10 million or Balance sheet total is less/equal
EUR 10 million
474
Staff headcount is above 250; Turnover is above EUR 50 million or Balance sheet total is above EUR 43
million

180
 Adopt AI Study

Figure 70 Respondents’ sectors

Education 29
E-government 22
Health 19
Mobility/Transport 16
Finance/Economic affairs 9
Agriculture 9
Recreation, culture and religion 9
Environmental protection 6
Defence 6
Public order and safety 3
Social protection 2

Source: Online survey, N=77.

3.7.1 Sectoral comparison of challenges in the online survey

To compare the challenges between sectors in the online survey, the following steps
were made. First, the qualitative significance scale in the online survey was converted
into a numerical scale, ranging from 1 to 5 (from not significant to strongly
significant). Second, the average response was calculated, which allowed us to rank
the challenges according to their significance. Third, the same was done for each of
the sectors, which allows us to provide a comparison of challenges between the
sectors.
Procurement process challenges
The most significant procurement process challenges for all sectors are:
 Burdensome administrative requirements
 Lack of clarity from the public sector regarding their needs/demands
 An emphasis on price/cost-savings over service quality and non-financial
benefits
 Unclear regulatory requirements
 The complexity of the writing of technical specifications
 Unclear procurement processes
 Lack of a common language across organisations/sectors
 The need to demonstrate in-year cost savings
 Lack of ethical frameworks/ ethical oversight specific to AI
When comparing the significance of the challenges between the health, e-government,
transport/mobility and education sectors, there were slight sectoral differences in
rating. E-government and mobility/transport sector respondents rated “An emphasis
on price/cost-savings over service quality and non-financial benefits” and “The need to
demonstrate in-year cost savings” as higher challenges than the total average.
It must be noted that given the relatively small difference in responses and low
numbers of respondents, these differences cannot be considered significant.

181
 Adopt AI Study

The table below provides a summary of the results. Red indicates those challenges
that are higher than the total average.

182
 Adopt AI Study

Table 8 Comparison of procurement process challenges between sectors.

Procurement process challenges Mobility/

Total Health E-gov Education
Total transport
average sector sector Sector
rating sector
response rating rating rating
(N=77) rating
(N=77) (N=19) (N=22) (N=29)
(N=16)

Burdensome administrative requirements 3.92 1 1 1 1 1

Lack of clarity from the public sector regarding their needs/demands 3.83 2 2 3 3 2

An emphasis on price/cost-savings over service quality and non-financial benefits 3.83 3 3 2 2 3

Unclear regulatory requirements 3.78 4 4 4 4 4

The complexity of the writing of technical specifications 3.49 5 5 5 5 5

Unclear procurement processes 3.46 6 6 7 7 6

Lack of a common language across organisations/sectors 3.37 7 7 8 8 7

The need to demonstrate in-year cost savings 3.31 8 8 6 6 8

Lack of ethical frameworks/ ethical oversight specific to AI 3.29 9 9 9 9 9

Source: Online survey, N=77.

183
 Adopt AI Study

Data challenges
The most significant data challenges for all sectors are:
 Unsatisfactory sharing of data across organisational boundaries
 Insufficient access to large volumes of high-quality data
 Lack of data to understand where AI is needed/ best suited
 Underdeveloped data governance
 Lack of clear “data ownership”/ “data sovereignty”
 Inadequate data management
 Lack of trust/ public acceptability with public sector data handling
 Absence of data standards
When comparing the significance of the challenges between the health, e-government,
transport/mobility and education sectors, there were slight sectoral differences in
rating. E-government and mobility/transport sector respondents rated “Insufficient
access to large volumes of high-quality data” as a higher challenge than the total
average. Also, the “Lack of clear “data ownership”/ “data sovereignty” was rated
higher by the e-government sector.
It must be noted that given the relatively small difference in responses and low
numbers of respondents, these differences cannot be considered significant.
The table below provides a summary of the results. Red indicates those challenges
that are higher than the total average.

184
 Adopt AI Study

Table 9 Comparison of data challenges between sectors.

Data challenges Mobility/

Total Health E-gov
Total transport Education
average sector sector
rating sector sector
response rating rating
(N=77) rating (N=29)
(N=77) (N=19) (N=22)
(N=16)

Unsatisfactory sharing of data across organisational boundaries 4.05 1 1 2 2 1

Insufficient access to large volumes of high-quality data 3.95 2 2 1 1 2

Lack of data to understand where AI is needed/ best suited 3.93 3 3 3 3 3

Underdeveloped data governance 3.78 4 4 5 4 4

Lack of clear “data ownership”/ “data sovereignty” 3.77 5 5 4 5 5

Inadequate data management 3.71 6 6 6 6 6

Lack of trust/ public acceptability with public sector data handling 3.68 7 7 7 7 7

Absence of data standards 3.53 8 8 8 8 8

Source: Online survey, N=77.

185
 Adopt AI Study

AI technology challenges
The most significant AI technology challenges for all sectors are:
 Lack of transparency in AI systems decision support/making processes
 Difficulty to inspect and assess an AI solution before their actual deployment
 Potential for biases/discrimination within the systems
 Difficulty in establishing liability and responsibility for the AI system
 Lack of limited regulatory spaces ("sandboxes") for experimenting with AI
solutions and monitoring their impacts
 Requirements for the explainability of AI solutions are excessively high
When comparing the significance of the challenges between the health, e-government,
transport/mobility and education sectors, there were slight sectoral differences in
rating. E-government and mobility/transport sector respondents rated “Difficulty to
inspect and assess an AI solution before their actual deployment” as higher than the
total average.
In addition, the “Lack of limited regulatory spaces ("sandboxes") for experimenting
with AI solutions and monitoring their impacts” was rated higher than the total
average by respondents in the e-government, transport/mobility and education
sectors.
It must be noted that given the relatively small difference in responses and low
numbers of respondents, these differences cannot be considered significant.
The table below provides a summary of the results. Red indicates those challenges
that are higher than the total average.

186
 Adopt AI Study

Table 10 Comparison of AI technology challenges between sectors.

AI technology challenges Total Total Health E-gov Mobility/ Education

average rating sector sector transport sector
response (N=77) rating rating sector (N=29)
(N=77) (N=19) (N=22) rating
(N=16)

Lack of transparency in AI systems decision support/making processes 3.83 1 1 2 2 1

Difficulty to inspect and assess an AI solution before their actual 3.80 2 2 1 1 2

deployment

Potential for biases/discrimination within the systems 3.73 3 3 3 3 3

Difficulty in establishing liability and responsibility for the AI system 3.70 4 4 5 5 5

Lack of limited regulatory spaces ("sandboxes") for experimenting with AI 3.68 5 5 4 4 4

solutions and monitoring their impacts

Requirements for the explainability of AI solutions are excessively high 3.48 6 6 6 6 6

Source: Online survey, N=77.

187
 Adopt AI Study

Organisational capacity challenges

The most significant organisational capacity challenges for all sectors are:
 Lack of human resources for managing the system
 Lack of political support
 Lack of understanding of the capabilities/benefits of AI solutions
 Lack of human resources for procuring the system
 System complexity and lack of single-entry point
 Lack of system interoperability
 Lack of trust within organisations
 Lack of managerial support
 Lack of digital skills
 Lack of political support
When comparing the significance of the challenges between the health, e-government,
transport/mobility and education sectors, there were slight sectoral differences in
rating. E-government and mobility/transport sector respondents rated “Lack of
human resources for procuring the system”, “System complexity and lack of single-
entry point” and “Lack of digital skills” higher than the total average.
Respondents from the education sector rated “System complexity and lack of single-
entry point”, “Lack of digital skills” and “Lack of political support” higher than the total
average.
It must be noted that given the relatively small difference in responses and low
numbers of respondents, these differences cannot be considered significant.
The table below provides a summary of the results. Red indicates those challenges
that are higher than the total average.

188
 Adopt AI Study

Table 11 Comparison of organisational capacity challenges.

Organisational capacity challenges Total Total Health E-gov Mobility/ Education

average rating sector sector transport sector
response (N=77) rating rating sector (N=29)
(N=77) (N=19) (N=22) rating
(N=16)

Lack of human resources for managing the system 3.84 1 1 1 1 1

Lack of political support 3.64 2 2 2 2 2

Lack of understanding of the capabilities/benefits of AI solutions 3.64 3 3 5 5 3

Lack of human resources for procuring the system 3.63 4 4 3 3 5

System complexity and lack of single-entry point 3.63 5 5 4 4 4

Lack of system interoperability 3.58 6 6 6 6 6

Lack of trust within organisations 3.49 7 7 7 7 7

Lack of managerial support 3.43 8 8 9 9 10

Lack of digital skills 3.42 9 9 8 8 8

Lack of political support 3.42 10 10 10 10 9

Source: Online survey, N=77.

189
 Adopt AI Study

3.7.2 Comparison of policy recommendations in the online survey

To see if there are significant differences in recommendations between the sectors, we
took the recommendations identified in the recent JRC Science for Policy Report AI
Watch: Road to the Adoption of Artificial Intelligence by the Public Sector 475 and asked
stakeholders to rate how beneficial they are in the online survey.
First, the qualitative scale in the online survey was converted into a numerical scale,
ranging from 1 to 5 (from not beneficial to strongly beneficial). Second, the average
response was calculated, which allowed us to rank the recommendations according to
how beneficial they are perceived to be. Third, the same was done for each of the
sectors, which allows us to provide a comparison of recommendations between the
sectors.
The most beneficial recommendations to increase the uptake and public procurement
of AI are:
 Optimise funding in support of AI in the public sector to promote the spreading
and scaling of reusable solutions.
 Support multidisciplinary research and knowledge creation amongst European
universities and Research and Development (R&D) institutions around AI for the
public sector.
 Reinforce and advance existing initiatives on open data and interoperability
 Build a common European Data Space for public sector bodies and their
operators, drawing from the compilation of relevant AI datasets and related
Registries throughout Europe.
 Design national and European capacity-building programmes for public
sector innovators aiming to develop and/or adopt AI in support of the digital
transformation of public services.
 Build upon and promote the use of regulatory sandboxes for public
administrations, allowing experimentation of AI-enabled solutions in controlled
environments.
 Promote the adoption of ethical principles, the development of guidelines,
and the identification of mitigating measures to minimize the risks of
deployment of AI by the public sector.
 Harmonise and complement EU regulations to promote human-centric and
trustworthy AI-enabled public services for all citizens.
 Share reusable and interoperable AI components at all operational levels of
European public administrations.
 Promote AI in the public sector in support of sustainability while developing
sustainable AI, in compliance with environmental principles, and leveraging
civic engagement and participation.
When comparing the recommendations between the health, e-government,
transport/mobility and education sectors, there were slight sectoral differences in
rating. E-government and mobility/transport sectors rated the recommendations on
supporting multidisciplinary research and knowledge creation amongst European
universities and Research and Development (R&D) institutions around AI for the public
sector, building a common European Data Space for public sector bodies and their
operators, harmonising and complementing EU regulations to promote human-centric

475
https://ai-watch.ec.europa.eu/publications/ai-watch-road-adoption-artificial-intelligence-public-
sector_en

190
 Adopt AI Study

and trustworthy AI-enabled public services, and sharing reusable and interoperable AI
components at all operational levels of European public administrations higher than
the total average.
Respondents in the education sector rated recommendations on building a common
European Data Space for public sector bodies and their operators, and building and
promoting the use of regulatory sandboxes for public administrations higher than the
total average.
It must be noted that given the relatively small difference in responses and low
numbers of respondents, these differences cannot be considered significant.
The table below provides a summary of the results. Orange indicates those
recommendations that are rated higher than the total average.

191
 Adopt AI Study

Table 12 Comparison of recommendations

Recommendations on the procurement/uptake of AI Total Total Health E-gov Mobility/ Education

average rating sector sector transport sector
response (N=77) rating rating sector (N=29)
(N=77) (N=19) (N=22) rating
(N=16)

Optimise funding in support of AI in the public sector to promote the spreading and 4.12 1 1 2 2 1
scaling of reusable solutions.

Support multidisciplinary research and knowledge creation amongst European 4.07 2 2 1 1 2

universities and Research and Development (R&D) institutions around AI for the
public sector.

Reinforce and advance existing initiatives on open data and interoperability 3.96 3 3 4 4 4

Build a common European Data Space for public sector bodies and their operators, 3.94 4 4 3 3 3
drawing from the compilation of relevant AI datasets and related Registries
throughout Europe.

Design national and European capacity-building programmes for public sector 3.86 5 5 5 5 6
innovators aiming to develop and/or adopt AI in support of the digital
transformation of public services.

Build upon and promote the use of regulatory sandboxes for public administrations, 3.84 6 6 8 8 5
allowing experimentation of AI-enabled solutions in controlled environments.

Promote the adoption of ethical principles, the development of guidelines, and the 3.83 7 7 9 9 7
identification of mitigating measures to minimize the risks of deployment of AI by
the public sector.

Harmonise and complement EU regulations to promote human-centric and 3.82 8 8 6 6 8

trustworthy AI-enabled public services for all citizens.

Share reusable and interoperable AI components at all operational levels of 3.82 9 9 7 7 9

European public administrations.

192
 Adopt AI Study

Promote AI in the public sector in support of sustainability while developing 3.78 10 10 10 10 10

sustainable AI, in compliance with environmental principles, and leveraging civic
engagement and participation.

Promote the development of multilingual guidelines, criteria and tools for public 3.74 11 11 11 11 11
procurement of AI solutions in the public sector throughout Europe.

Create a European marketplace for GovTech solutions in support of public sector 3.68 12 12 12 12 12
digital transformation.

Set up an EU observatory on AI, built on a pan-European network of AI national 3.68 13 13 13 13 13

observatories, to gather, share, and collectively manage best practices and
experiences learned from different stakeholders in the public sector throughout
Europe.

Create an EU-wide network of governance bodies for streamlined management of AI 3.59 14 14 14 14 14

in the public sector.

Develop and apply umbrella impact assessment frameworks based on key 3.54 15 15 15 15 15
influencing factors to measure the use and impact of AI in the public sector.

Develop and promote dedicated AI-enabled solutions based on co-creation 3.47 16 16 16 16 16

approaches (e.g., through citizen participation in the development and deployment
of AI) to increase citizens' and businesses' relevance trust and confidence in the use
of AI by the public sector.

Source: Online survey, N=77.

193
 Adopt AI Study

3.8 Conclusion
The stakeholder consultations undertaken as part of this study show that the main
challenges in the uptake of AI centres on the procurement process, data, AI
technology and organisation:
 The most significant procurement process challenges are related to
burdensome administrative requirements, lack of clarity from the public sector
regarding their needs/demands, an emphasis on price/cost-savings over service
quality and non-financial benefits, unclear regulatory requirements, and the
complexity of writing technical specifications.
 The most significant data challenges are related to unsatisfactory sharing of
data across organisational boundaries, insufficient access to large volumes of
high-quality data, lack of data to understand where AI is needed/ best suited,
underdeveloped data governance, and lack of clear “data ownership”/ “data
sovereignty”.
 The most significant AI technology challenges are lack of transparency in AI
systems' decision support/making processes, difficulty to inspect and assess an
AI solution before their actual deployment, the potential for
biases/discrimination within the systems, difficulty in establishing liability and
responsibility for the AI system, lack of limited regulatory spaces ("sandboxes")
for experimenting with AI solutions and monitoring their impacts, and high
requirements for the explainability of AI solutions.
 The most significant organisational challenges are a lack of human resources
for managing the system, lack of political support, lack of understanding of the
capabilities/benefits of AI solutions, lack of human resources for procuring the
system, system complexity and lack of single-entry point, and lack of system
interoperability.
The comparative analysis showed that these challenges are not significantly
different across the sectors. The challenges were raised and discussed by
participants in all the workshops, irrespective of the sector. In addition, the differences
between the sectors in the online survey were minor.
Similarly, the comparison of the policy recommendations between the health, e-
government, transport/mobility and education sectors in the online survey showed no
significant sectoral differences in policy recommendations. The policy
recommendations are transversal and important for all the sectors analysed.

194
 Adopt AI Study

4 Chapter 4: Policy recommendations on the uptake of AI in

the public sector
4.1 Introduction
This chapter explores the recommendations to increase the uptake and public
procurement of AI. It includes the results from:
 The policy workshop dedicated to identifying and exploring policy solutions and
recommendations for the uptake of AI in the public sector.
 An online survey, dedicated to assessing recommendations for the uptake of AI
in the public sector. The recommendations are based on the recent JRC Science
for Policy Report AI Watch: Road to the Adoption of Artificial Intelligence by the
Public Sector476.
The more detailed methodologies for these consultations can be found in the annex.
The chapter proceeds by providing the analysis of the policy workshop, the
comparative assessment of recommendations from the policy workshop and the online
survey. The chapter concludes by triangulating the results from the two sets of
recommendations.
4.2 Comparative assessment of policy recommendations
This section provides a comparative assessment of the policy recommendations
identified in the sectoral and the dedicated policy workshop presented above
and in the online survey conducted as part of this study. The recommendations
selected for comparison are the ones that were identified as the most important
challenges by the participants in the policy workshop and the potential policy actions
are the most significant and direct solutions offered by the participants. The
recommendations in the online survey are taken from the recent JRC Science for
Policy Report AI Watch: Road to the Adoption of Artificial Intelligence by the Public
Sector477.
After presenting the recommendations from the policy workshop and the online
survey, section 5 concludes by triangulating the results between the two sets of
recommendations and presenting the recommendations in order of significance.
We provide an assessment of the most important recommendations identified in the
policy workshop on a scale of low/medium/high based on the following criteria:
 Effectiveness. The extent to which recommendations achieve their goals and
the level of benefits they provide.
 Feasibility. The extent to which recommendations can be implemented and
whether they have significant limits.
 Efficiency. The extent to which recommendations provide value for money and
if they require significant resources.
 EU added value. The extent to which recommendations provide benefits at a
European level and the Single Market and which level of implementation,
European or Member State, is the most appropriate, keeping in mind the
principle of subsidiarity.

476
https://ai-watch.ec.europa.eu/publications/ai-watch-road-adoption-artificial-intelligence-public-
sector_en
477
https://ai-watch.ec.europa.eu/publications/ai-watch-road-adoption-artificial-intelligence-public-
sector_en

195
 Adopt AI Study

The following recommendations are analysed in the rest of the section:

 Ensure a clear regulatory framework for AI
 Foster coordinated procurement strategies
 Increase expertise among public buyers
 Improve alignment between industry and public sector expectations
 Increase clarity and harmonisation around cross-border data flows
 Promote interoperability
 Facilitate “data ownership”
 Address the challenges surrounding the sharing of data between and across
sectors
 Address bias within AI and data sources
 Improve technological accountability and transparency
 Increase trust in AI solutions
 Facilitate the integration of new AI technologies/services into existing systems
(education/health/transport)
 Focus on long-term implementation
 Improve access to funding and resources
 Foster the development of skills and competencies

Ensure a clear regulatory framework for AI

Potential policy actions:
 Adopt a regulatory framework for AI.
 Provide a clear definition of what is AI and clarity on the scope of what AI
regulations cover.
 Provide clarity as to how EU legislation and regulatory frameworks interact
together.

Criteria Rating Description

Effectiveness/ High This challenge is identified both by the public administrations
(seen in reports from the Danish, Austrian and French
benefits
governments) and the private sector (the lack of laws or
regulations poses an obstacle for 69% of the surveyed
private sector enterprises by the European Commission478).
Consequently, given its significance, addressing the lack of
clarity on AI regulation can be highly effective and provide
the following benefits:
1. Improved accountability: Defined rules and regulations
can help ensure that AI systems are developed and
used responsibly and that those who do not follow
these rules are held accountable.

478
European Commission, Directorate-General for Communications Networks, Content and
Technology. 2020. European enterprise survey on the use of technologies based on artificial intelligence :
final report. URL: https://data.europa.eu/doi/10.2759/759368

196
 Adopt AI Study

2. Greater transparency: Well-defined regulations can

also help increase transparency around the
development and use of AI, enabling stakeholders to
better understand how these systems work and how
they might be affecting society.
3. Enhanced public trust: By establishing clear guidelines
for the development and use of AI, regulators can help
build public trust in these technologies and address
concerns about their potential negative impacts.
4. Increased innovation: Establishing clear regulations
can also create a more stable and predictable
regulatory environment, which can encourage
innovation and investment in AI.
5. Better protection of human rights: Defined regulations
can help ensure that AI systems do not infringe on
individuals' human rights, such as their right to privacy
or their right to be free from discrimination.
Feasibility/ Medium There are several potential limitations to addressing the lack
of clarity on AI regulation:
limits
1. Lack of agreement: There may be disagreement
among governments, industries, and experts on how
to regulate AI, making it difficult to develop clear and
effective regulations.
2. Technological complexity: AI is a rapidly evolving field,
and it can be difficult to keep up with the latest
developments and their potential impacts. This can
make it difficult to craft regulations that are both
effective and flexible enough to adapt to changing
technology.
3. Legal and ethical considerations: There are many legal
and ethical considerations to take into account when
regulating AI, including issues related to privacy, bias,
and accountability. Balancing these competing
interests can be challenging.
4. Limited understanding of the potential consequences:
It can be difficult to predict the long-term
consequences of AI and how it will be used, which
makes it hard to anticipate the potential impacts of
regulatory measures.
Efficiency/ High significant resources required, and at the same time, the
work might build on the existing regulation. Thereby,
resources
efficiency might be maximised. Some examples of EU
regulations related to AI include:
1. The General Data Protection Regulation (GDPR): This
legislation, which came into effect in 2018, sets out
rules for the protection of personal data, including data
processed by AI systems.
2. The Ethical Guidelines for Trustworthy AI: These
guidelines, developed by the European Commission's
High-Level Expert Group on Artificial Intelligence (AI
HLEG), provide a framework for the development and
deployment of trustworthy AI.
3. The AI Act: This legislation, which was proposed by
the European Commission in 2020, aims to establish a
framework for the regulation of AI in the EU. It
includes provisions on the development, deployment,

197
 Adopt AI Study

and use of AI, as well as measures to ensure the

transparency and accountability of AI systems.
4. Public procurement of AI: The EU has also established
guidelines for the procurement of AI systems by public
authorities479, which aim to ensure that such systems
are reliable, robust, and respectful of fundamental
rights. The AI Act will be fully consistent with these
guidelines as noted in the impact assessment
accompanying the proposal480.
Overall, the EU has taken a proactive approach to regulate
AI, with a focus on ensuring that the technology is developed
and used in a way that is transparent, accountable, and
respectful of fundamental rights.
EU added High There are several potential benefits to the European Union
value/ (EU) of clarifying the use and procurement of artificial
intelligence (AI):
subsidiarity
1. Promoting the development of trustworthy AI: By
establishing clear guidelines and regulations for the
development and deployment of AI, the EU can help to
ensure that the technology is developed and used in a
way that is transparent, accountable, and respects
fundamental rights. This can help to build trust in AI
and encourage its adoption by businesses and other
organizations.
2. Protecting fundamental rights: Clarifying the use and
procurement of AI can help to protect the fundamental
rights of EU citizens, including their right to privacy,
non-discrimination, and data protection.
3. Levelling the playing field: By establishing clear rules
and standards for the procurement of AI systems by
public authorities, the EU can help to ensure that all
companies have an equal opportunity to compete for
these contracts.
4. Supporting innovation: By establishing a clear and
predictable regulatory environment for AI, the EU can
help to foster innovation and encourage the
development of new technologies.
Overall, clarifying the use and procurement of AI can help to
ensure that the technology is developed and used in a way
that is beneficial to society and supports the EU's goals of
innovation, competitiveness, and respect for fundamental
rights.

Foster coordinated procurement strategies

Potential policy actions:
 Identify 'owners' for procurement strategy and their needs.
 Reinforce ideas through strategy documents and activities.
 Consider the procurement power of larger-scale organisations and assess what
their processes and needs are.

479
https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai
480
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence

198
 Adopt AI Study

 Provide an inventory of best practices from similar entities.

Criteria Rating Description
Effectiveness/ High There are several potential benefits that the European Union
(EU) could realize by addressing the lack of a coordinated
benefits
procurement strategy:
1. Improved efficiency and cost savings: A coordinated
procurement strategy could help streamline the
procurement process, reducing duplication and
bureaucracy, and allowing the EU to leverage its
purchasing power to negotiate better deals with
suppliers. This could lead to significant cost savings for
the EU.
2. Increased transparency and fairness: A coordinated
procurement strategy could help increase transparency
in the procurement process, ensuring that all suppliers
have a fair and equal opportunity to participate in EU
tenders. This could help reduce the risk of corruption
and ensure that public funds are used effectively and
efficiently.
3. Greater alignment with EU priorities: A coordinated
procurement strategy could help ensure that the EU's
procurement decisions are aligned with its broader
policy priorities, such as sustainability, innovation, and
regional development.
4. Enhanced competitiveness: By coordinating
procurement, the EU could become more competitive by
adopting a more agile and flexible approach to
procurement, and by making it easier for small and
medium-sized enterprises to participate in the EU's
procurement market.
5. Stronger bargaining power: A coordinated procurement
strategy could give the EU greater bargaining power
when negotiating with suppliers, allowing it to secure
better terms and conditions, and to drive innovation and
value for money.

Feasibility/ High There are several potential limits to addressing the lack of a
coordinated procurement strategy in the European Union (EU).
limits
Some of these limits may include:
1. Resistance to change: Implementing a coordinated
procurement strategy may require significant changes to
existing procurement processes and systems, which
may be met with resistance from procurement officials
and other stakeholders who are used to working in a
decentralized way.
2. Complexity and bureaucracy: A coordinated
procurement strategy may introduce additional
bureaucracy and complexity into the procurement
process, which could lead to delays and increased costs.
3. Loss of local control: A coordinated procurement
strategy may involve the transfer of decision-making
power from local authorities to a centralised agency,
which could lead to concerns about the loss of local
control and accountability.

199
 Adopt AI Study

4. Legal and regulatory challenges: A coordinated

procurement strategy may require changes to EU
procurement law and regulations, which could be a
complex and time-consuming process. It would be
important to explain why changes would be needed and
what would be the provision in existing procurement
laws that currently prevent the adoption of such
strategies.
5. Political challenges: A coordinated procurement strategy
may be controversial and may face political opposition,
particularly if it is perceived as a threat to the
sovereignty of member states (MS). Thus, for MS it
should be most likely voluntary and presented as a
support and good practice tool.
Despite these potential limits, it is important to note that a
coordinated procurement strategy could also bring significant
benefits, including improved efficiency and cost savings,
increased transparency and fairness, and greater alignment
with EU policy priorities.
Efficiency/ High Several resources would be required in addressing the lack of a
coordinated procurement strategy in the European Union (EU).
resources
Some of these resources may include:
1. Funding: Establishing a procurement agency or support
unit, and implementing a coordinated procurement
strategy, may require additional funding to cover the
costs of setting up and operating them, as well as any
necessary changes to existing procurement systems and
processes481.
2. Expertise: A coordinated procurement strategy may
require the development of specialized expertise in
areas such as procurement law, AI procurement, and
data analytics, to support the design and
implementation of the strategy. A procurement agency
or unit might be responsible for this.
3. Infrastructure: Implementing a coordinated
procurement strategy may require the development of
new infrastructure, such as a procurement portal or
platform, to support the centralised procurement
process. An example is the European Commission’s
procurement portal.
4. Stakeholder engagement: Engaging with a range of
stakeholders, including procurement officials, suppliers,
and other interested parties, will be important to ensure
that the coordinated procurement strategy is well-
understood, supported, and implemented effectively.
5. Legal and regulatory frameworks: Establishing a
coordinated procurement strategy may require the
development of new legal and regulatory frameworks,
the modification of existing ones, or the alignment with
existing or planned legislation, such as the AI Act, to
support the coordinated procurement process.
It will be important for the EU to carefully consider these and
other resource requirements, and to plan and allocate

481
A coordinated strategy might necessarily need to be enforced by a single agency or unit, compliance may
be decentralised. If the strategy is only for EU institutions, it could be internal and decentralised. MS might
have a similar system also.

200
 Adopt AI Study

resources appropriately, to ensure the success of a coordinated

procurement strategy.
Efficiency can be maximized and resources minimized by
adopting a mix of coordination and de-centralisation. In the
United States, AI is both coordinated under the federal
government and decentralized among state governments482.
Coordination is enabled through the National Technical
Information Service (NTIS) under the U.S. Commerce
Department and the Federal Risk and Authorization
Management Program (FedRAMP). The former is responsible
for helping federal agencies rapidly analyze, manage, and
implement scalable data solutions by leveraging an extensive
NTIS network of technical talent from private industry, which is
often difficult to locate in today’s competitive information
technology landscape. FedRAMP’s mission is to promote the
adoption of secure cloud services across the federal
government by providing a standardized approach to security
and risk assessment.
EU added High Coordination of procurement at the EU level can have
value/ significant benefits, as discussed in the effectiveness section
above.
subsidiarity
There may be several benefits to coordinated AI procurement
at the member state level in comparison to the EU level:
1. Closer alignment with local needs and priorities:
Coordinating AI procurement at the member state level
may allow for a better alignment of procurement
decisions with the specific needs and priorities of the
member state, as well as with the local context and
conditions in which the AI solutions will be used.
2. Greater flexibility and agility: Coordinating AI
procurement at the member state level may allow for a
more agile and flexible approach to procurement,
allowing member states to respond more quickly to
changing needs and opportunities.
3. Improved accountability and transparency: Coordinating
AI procurement at the member state level may help
improve accountability and transparency, as
procurement decisions will be made by officials who are
directly accountable to the member state's citizens and
stakeholders.
4. Enhanced competitiveness: Coordinating AI
procurement at the member state level may allow
member states to better leverage their purchasing
power to negotiate better deals with suppliers, and to
support the development of local AI capabilities and
industries.
5. Greater autonomy: Coordinating AI procurement at the
member state level may allow member states to retain a
greater degree of autonomy in their procurement
decisions, which may be important for some states that
are particularly sensitive about the sovereignty of their
procurement processes.
Overall, the benefits of coordinating AI procurement at the
member state level in comparison to the EU level will depend
on the specific needs, priorities, and context of the member

482
The World Bank. 2020. Artificial Intelligence in the Public Sector: Maximizing Opportunities, Managing
Risks. Available at: https://openknowledge.worldbank.org/handle/10986/35317

201
 Adopt AI Study

state, and will need to be carefully balanced against the

potential benefits of a coordinated procurement strategy at the
EU level.

Increase expertise among public buyers

Potential policy actions:
 Introduce national and European points of contact for practical advice.
 Use Digital Innovation Hubs (DIHs) and GovTech incubators for consulting, or
competence centres for aiding public bodies to adopt/set up fitting solutions.
 Educate and raise awareness of the potential of AI with real-world success
stories.
 Development of a specific Community of Practice on the development and
testing of EU-wide AI procurement Clauses, currently being developed483.

Criteria Rating Description

Effectiveness/ High The lack of expertise is a well-documented barrier and
challenge in the adoption of AI solutions in public
benefits
institutions484. Addressing the lack of expertise among public
buyers in AI procurement in the EU can have several
benefits. Some of these benefits include:
1. Improved decision-making: By increasing the expertise
of public buyers, they will be better equipped to make
informed decisions about which AI products and
services to purchase. This can help to ensure that the
AI procurement process is more efficient and effective.
2. Increased value for money: With more expertise,
public buyers will be better able to evaluate the quality
and value of different AI products and services. This
can help to ensure that the government is getting the
best value for its money when it comes to AI
procurement.
3. Greater transparency: By increasing the expertise of
public buyers, there will be greater transparency in the
AI procurement process. This can help to build trust
between the government and the public, and ensure
that AI procurement is carried out fairly and
transparently.
4. Improved innovation: By increasing the expertise of
public buyers, they will be better able to identify and
support innovative AI solutions. This can help to drive
innovation in the AI industry, and lead to the
development of new and improved AI products and
services.
Overall, addressing the lack of expertise among public buyers
in AI procurement can help to improve the efficiency,
effectiveness, and transparency of the procurement process,
and drive innovation in the AI industry.

483
https://ec.europa.eu/newsroom/growth/items/746858/en
484
Assaf Ariel et al, 2021, Barriers and challenges of e-government services: A systematic literature review
and meta-analyses, https://iopscience.iop.org/article/10.1088/1757-899X/1125/1/012027/pdf

202
 Adopt AI Study

Feasibility/ High There are several potential limits to addressing the lack of
expertise among public buyers in AI procurement in the EU.
limits
Some of these limits include:
1. Time and resources: Building expertise in AI
procurement can take time and resources, which may
be in short supply for public buyers.
2. Limited budgets: Public buyers may have limited
budgets for training and development, which can limit
the extent to which they can build expertise in AI
procurement.
3. Difficulty finding skilled personnel: There may be a
shortage of skilled personnel with expertise in AI
procurement, which can make it difficult for public
buyers to find the resources they need to build their
expertise.
4. Resistance to change: Some public buyers may be
resistant to changing their existing procurement
processes or adopting new technologies, which can
limit the extent to which they can build expertise in AI
procurement.
5. Legal and regulatory challenges: There may be legal
and regulatory challenges associated with AI
procurement, which can make it difficult for public
buyers to fully understand and navigate the process.
Overall, while addressing the lack of expertise among public
buyers in AI procurement in the EU can have many benefits,
it may also be subject to several potential limits and
challenges.
Nevertheless, the feasibility is high given that there are
already existing examples to build on. Models on national
contact points and Digital Innovation Hubs already exist and
can be tailored to promote expertise in AI. There are several
other examples of efforts to address the lack of expertise
among public buyers in AI procurement in the EU.
One example is the European Artificial Intelligence
Alliance485, which was established in 2018 to promote the
development and deployment of AI in the EU. The Alliance
brings together stakeholders from across the EU to share
knowledge and expertise on AI, and to develop
recommendations for the development and deployment of AI
in the region.
Another example is the European AI Procurement
Guidelines486, which were developed by the European
Commission in collaboration with the European Artificial
Intelligence Alliance. The guidelines guide public buyers on
how to effectively procures AI products and services, and
how to ensure that these products and services meet the
necessary quality and ethical standards. The AI Act will be
fully consistent with these guidelines as noted in the impact
assessment accompanying the proposal487.

485
https://futurium.ec.europa.eu/en/european-ai-alliance
486
https://digital-strategy.ec.europa.eu/en/policies/european-ai-alliance
487
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence

203
 Adopt AI Study

Another example is DG GROW’s Big Buyers platform488. It

establishes working groups, which focus on a specific unmet
public procurement need. The groups meet regularly, both
online and in person, and consist of a small number of public
purchasing organisations, such as local and regional
governments, central purchasing bodies, and other public
agencies. One working group – “Digital solutions in the
healthcare sector” – focuses on more efficient procurement of
innovative medical devices, digital services and IT systems
through the collaboration of public hospitals and purchasing
bodies that can lower the pressure on health budgets while
facilitating better value for money489.
There are also many individual initiatives and programs being
undertaken by governments and other organizations across
the EU to help build expertise in AI procurement. These
initiatives may include training programs, workshops, and
other resources designed to help public buyers understand
the AI procurement process and how to effectively evaluate
and select AI products and services.
Efficiency/ Medium Several resources may be required in addressing the lack of
expertise among public buyers in AI procurement in the EU.
resources
Some of these resources may include:
1. Training programs: Public buyers may need to invest
in training programs to help them build their expertise
in AI procurement. These programs could include
courses, workshops, or other educational materials
designed to help public buyers understand the AI
procurement process and how to effectively evaluate
and select AI products and services.
2. Expert support: Public buyers may need to seek out
external expertise and support from organizations or
individuals with experience in AI procurement. This
could include consulting firms, industry experts, or
other organizations that can provide guidance and
support to public buyers as they navigate the AI
procurement process.
3. Educational materials: Public buyers may need access
to a range of educational materials, including guides,
tutorials, and other resources that can help them
understand the AI procurement process and how to
effectively evaluate and select AI products and
services.
4. Legal and regulatory guidance: Public buyers may
need access to legal and regulatory guidance to help
them understand and navigate the complex legal and
regulatory requirements associated with AI
procurement in the EU.
Overall, addressing the lack of expertise among public buyers
in AI procurement in the EU may require a range of
resources, including training programs, expert support,
educational materials, and legal and regulatory guidance.
EU added Medium It may be beneficial to address the lack of expertise among
value/ public buyers in AI procurement at both the EU and Member
State levels.
subsidiarity

488
https://bigbuyers.eu/about/bigbuyers
489
https://bigbuyers.eu/working-groups/digital-solutions-in-the-healthcare-sector

204
 Adopt AI Study

At the EU level, efforts to build expertise in AI procurement

could include the development of common standards and
regulations, the sharing of best practices and expertise
among Member States, and the provision of guidance and
support to Member States as they seek to build their
expertise in AI procurement.
At the Member State level, efforts to build expertise in AI
procurement could include the development of training
programs, the provision of educational materials, and the
establishment of dedicated AI procurement units or teams
within the government.
Overall, a combination of EU-level and Member State-level
efforts could be the most effective way to address the lack of
expertise among public buyers in AI procurement in the EU.
This approach could help to ensure that there is a consistent
approach to AI procurement across the EU, while also taking
into account the specific needs and circumstances of
individual Member States.

Improve alignment between industry and public sector expectations

Potential policy actions:
 Introduce “best practice” dialogues between industry, the public sector and
academia, and “roadshow” type events for building networks and shared goals
and understanding.
 Conduct internal research within public organisations to understand what
expectations and ideas on AI public employees have.

Criteria Rating Description

Effectiveness/ High Addressing the lack of alignment between industry and public
sector expectations in AI procurement in the EU could have
benefits
several potential benefits. Some of these benefits include:
1. Improved efficiency: By ensuring that there is greater
alignment between industry and public sector
expectations in AI procurement, the procurement
process may be more efficient, as it will be clearer
what is expected from the industry and what the public
sector is looking for in terms of AI products and
services.
2. Increased value for money: By ensuring that there is
greater alignment between industry and public sector
expectations, the public sector may be able to get
better value for its money when it comes to AI
procurement. This could be because the industry will
be able to offer products and services that more
closely meet the needs and expectations of the public
sector.
3. Enhanced innovation: By fostering greater alignment
between industry and public sector expectations, the
public sector may be able to encourage the
development of innovative AI solutions that meet the
needs of both sides. This could lead to the
development of new and improved AI products and
services.
4. Improved relationships: By addressing the lack of
alignment between industry and public sector

205
 Adopt AI Study

expectations, the public sector may be able to improve

its relationships with industry. This could lead to a
more collaborative and productive relationship
between the two sides, which could be beneficial for
both parties.
Overall, addressing the lack of alignment between industry
and public sector expectations in AI procurement in the EU
could have a range of benefits, including improved efficiency,
increased value for money, enhanced innovation, and
improved relationships.
There are several examples of efforts to address the lack of
alignment between industry and public sector expectations in
AI procurement in the EU, as discussed in the previous
recommendation.
Feasibility/ High There are several potential limits to efforts to address the
lack of alignment between industry and public sector
limits
expectations in AI procurement in the EU. Some of these
limits include:
1. Misalignment of goals: There may be fundamental
differences in the goals and priorities of industry and
the public sector, which can make it difficult to achieve
alignment in AI procurement.
2. Limited understanding: The industry and the public
sector may have a limited understanding of each
other's needs and expectations when it comes to AI
procurement, which can make it difficult to achieve
alignment.
3. Communication barriers: There may be communication
barriers between the industry and the public sector,
which can make it difficult to effectively share
information and build understanding.
4. Misperceptions: There may be misperceptions or
misunderstandings on both sides, which can further
complicate efforts to achieve alignment in AI
procurement.
5. Limited resources: Industry and the public sector may
have limited resources to devote to efforts to build
alignment in AI procurement, which can make it more
difficult to achieve.
Overall, while there are many potential benefits to addressing
the lack of alignment between industry and public sector
expectations in AI procurement in the EU, there are also
several potential limits and challenges that may need to be
overcome.
In light of the several examples of efforts to address the lack
of alignment between industry and public sector expectations
in AI procurement in the EU, including the European Artificial
Intelligence Alliance and the European AI Procurement
Guidelines. These initiatives suggest that it is possible to
achieve some level of alignment between industry and the
public sector in AI procurement in the EU.
Efficiency/ High Several resources may be required to address the lack of
alignment between industry and public sector expectations in
resources
AI procurement in the EU. Some of these resources could
include:
1. Staff and personnel: Addressing the lack of alignment
between industry and public sector expectations in AI

206
 Adopt AI Study

procurement may require the allocation of staff and

personnel to work on this issue. This could include
individuals with expertise in AI procurement, as well as
individuals with strong communication and
relationship-building skills to facilitate collaboration
between industry and the public sector.
2. Funding: Addressing the lack of alignment between
industry and public sector expectations in AI
procurement may require the allocation of funding to
support initiatives and programs aimed at building
alignment in this area. This could include funding for
training programs, workshops, and other resources
designed to help build understanding and foster
collaboration between industry and the public sector.
3. Time and effort: Achieving alignment between industry
and public sector expectations in AI procurement may
require a significant investment of time and effort. This
could include time spent building relationships and
understanding between industry and the public sector,
as well as time spent developing and implementing
strategies to foster greater alignment in AI
procurement.
Overall, addressing the lack of alignment between industry
and public sector expectations in AI procurement in the EU
may require the allocation of a range of resources, including
staff and personnel, funding, and time and effort.
There are several examples of efforts to address the lack of
alignment between industry and public sector expectations in
AI procurement in the EU, including the European Artificial
Intelligence Alliance and the European AI Procurement
Guidelines. These initiatives suggest that it is possible to
build on them to aim for higher efficiency.
EU added Medium Addressing the lack of alignment between industry and public
value/ sector expectations in AI procurement in the EU could be
done at both the EU and Member State levels.
subsidiarity
At the EU level, efforts to build alignment between industry
and public sector expectations in AI procurement could
include the development of common standards and
regulations, the sharing of best practices and expertise
among Member States, and the provision of guidance and
support to Member States as they seek to build alignment in
AI procurement.
At the Member State level, efforts to build alignment between
industry and public sector expectations in AI procurement
could include the development of training programs, the
provision of educational materials, and the establishment of
dedicated AI procurement units or teams within the
government.
Overall, a combination of EU-level and Member State-level
efforts could be the most effective way to address the lack of
alignment between industry and public sector expectations in
AI procurement in the EU. This approach could help to ensure
that there is a consistent approach to AI procurement across
the EU, while also taking into account the specific needs and
circumstances of individual Member States.

207
 Adopt AI Study

Increase clarity and harmonisation around cross-border data flows

Potential policy actions:
 Formulate clear incentives and use cases for data sharing.
 Promote open application programming interfaces (APIs).
 Develop more federated data networks and data spaces.
Criteria Rating Description
Effectiveness/ High Formulating clear incentives and use cases for data sharing
specifically focused on supporting the implementation of AI
benefits
technologies in the public sector would complement the
objectives of the European Data Strategy and support their
achievement. Promoting AI use cases through publications or
repositories such as the ones published by the JRC490
contribute to the existing body of knowledge on the topic and
moves from a more theoretical view to a solutions-based
analysis, built around many concrete examples.
Federated data networks and data spaces already being
implemented such as the Mobility Data Space described in
Section 2.2 provide a safe and trusted ecosystem for data
sharing, as stakeholders have access to the data they need
for new use cases and AI applications. The Commission Staff
Working Document on Common European Data Spaces
published in 2022491 highlighted numerous benefits such as
security, data control and governance, interoperability, and
openness.
Promoting open application programming interfaces (APIs)
has been explored by the European Commission492 and is
envisaged in the Open Data EU Directive 2019/1024, which
requires the use of APIs for ‘high-value’ and dynamic
datasets. The APIs present many benefits for the public
sector, including fostering data sharing and innovation in
governments and related public services through AI
technologies.
Addressing the lack of clarity and harmonisation around
cross-border data flows in AI procurement in the EU could
have several potential benefits. Some of these benefits
include:
1. Improved efficiency: By clarifying and harmonising
cross-border data flows in AI procurement, it may be
possible to make the procurement process more
efficient, as it will be clearer what is expected in terms
of data management and data protection.
2. Increased confidence: Clarifying and harmonising
cross-border data flows in AI procurement may
increase confidence among industry and the public
sector, as it will be clearer what is expected in terms
of data management and data protection. This could
encourage more companies to engage in AI
procurement with the public sector, and may

490
See Selected AI cases in the public sector available at https://data.jrc.ec.europa.eu/dataset/7342ea15-
fd4f-4184-9603-98bd87d8239a
491
Available at https://digital-strategy.ec.europa.eu/en/library/staff-working-document-data-spaces
492
See JRC (2020) Application Programming Interfaces in government - Why, what and how, or JRC (2020)
An Application Programming Interfaces (APIs) framework for digital government

208
 Adopt AI Study

encourage the public sector to be more active in

procuring AI products and services.
3. Enhanced trust: Clarifying and harmonising cross-
border data flows in AI procurement may enhance
trust between industry and the public sector, as both
sides will have a better understanding of their
respective roles and responsibilities concerning data
management and data protection.
4. Greater legal certainty: Clarifying and harmonising
cross-border data flows in AI procurement may
provide greater legal certainty for both industry and
the public sector, as it will be clearer what is expected
in terms of data management and data protection.
This could help to reduce the risk of legal disputes
arising concerning AI procurement.
Overall, addressing the lack of clarity and harmonisation
around cross-border data flows in AI procurement in the EU
could have a range of benefits, including improved efficiency,
increased confidence, enhanced trust, and greater legal
certainty.
Feasibility/ High Given that all three actions foreseen under this
recommendation are already being implemented at the EU
limits
and Member State levels under various forms, their feasibility
has been proven to be high.
There are several potential limits to efforts to address the
lack of clarity and harmonisation around cross-border data
flows in AI procurement in the EU. Some of these limits
include:
1. Complex legal landscape: The legal landscape
surrounding cross-border data flows in the EU is
complex, and it may be difficult to achieve clarity and
harmonisation in this area.
2. Different national approaches: There may be different
approaches to cross-border data flows in different
Member States, which can make it challenging to
achieve clarity and harmonisation at the EU level.
3. Conflicting priorities: Industry and the public sector
may have conflicting priorities when it comes to cross-
border data flows, which can make it difficult to
achieve clarity and harmonisation.
4. Limited resources: There may be limited resources
available to devote to efforts to clarify and harmonise
cross-border data flows in AI procurement, which can
make it more difficult to achieve.
5. Technological challenges: There may be technological
challenges associated with clarifying and harmonising
cross-border data flows in AI procurement, which
could limit progress in this area.
For example, in adopting APIs, governments can
encounter certain risks and challenges related to
cybersecurity issues, missing API governance
structures, or the lack of an API culture.493 However,
the adoption of such actions at the EU level could be
supported by the development of standards, guidelines
or awareness campaigns.

493
Ibid.

209
 Adopt AI Study

Overall, while there are many potential benefits to addressing

the lack of clarity and harmonisation around cross-border
data flows in AI procurement in the EU, there are also several
potential limits and challenges that may need to be
overcome. Nevertheless, as regards potential limits, these
are quite low when compared to the overall benefits provided
by the implementation of the actions.
Efficiency/ High The implementation of APIs could carry certain technical and
organisational costs related to their set-up, but these are
resources
expected to be minimal. Similarly, to implement federated
data networks and data spaces, further digitalisation is
needed at a national level. At the same time, it is necessary
to set up interoperable EU-wide infrastructures to enable the
cross-border use of data in the EU. However, these will be
supported through different EU funds and instruments.
Nevertheless, the benefits of all actions are expected to
exceed the potential costs and resources. For example,
overall, the European Health Data Space is expected to save
the EU around €11 billion over ten years.494
EU added Medium The creation of federated data networks and data spaces at
value/ the EU level has clear benefits linked to its cross-border
characteristic. For example, the COVID-19 pandemic has
subsidiarity
highlighted the critical role of data, and in this particular
context, a well-functioning common European data space for
health, which has been marked as one of the priority
initiatives by the European Council, is expected to make a
crucial difference in tackling potential health pandemics in the
future through EU cooperation.495
Formulating clear incentives and use cases for data sharing
can be implemented both at the EU level and MS level,
however, when done at the EU level, such as in the case of
the JRC databases, the benefits could be expected to be
higher, as such measures would have a higher level of
exposure and outreach.
Promoting open application programming interfaces (APIs),
again is one measure that could be implemented both at the
EU level and MS level, although less likely at the EU level.
However, certain measures such as interoperability standards
and guidelines developed at the EU level which would support
the widespread adoption of APIs by national and local
governments could have an increased added value.

Promote interoperability
Potential policy actions:
 Develop standards for data exchange.
 Develop national strategies for data complemented by organisation/sector-
specific strategies for data.
 Introduce incentives for creating interoperable systems.

494
Questions and answers - EU Health: European Health Data Space (EHDS) available at
https://ec.europa.eu/commission/presscorner/detail/en/QANDA_22_2712
495
European Commission (2022) Commission Staff Working Document on Common European Data Spaces

210
 Adopt AI Study

 Introduce documentation for semantic interoperability (data dictionaries,

ontologies, etc.).

Criteria Rating Description

Effectiveness/ High Lack of interoperability has been highlighted as an important
challenge for the adoption of AI by the public sector (see
benefits
Section 1.3.1.4 on Organisational capacity challenges and
Section 3.7.2 on cross-sectoral challenges reflected by the
stakeholder consultations). This was an issue that was
especially pointed out concerning the Mobility sector (see
section 2.3) and also in the Education sector (see section
2.5).
The research showed that the fragmentation of the European
data market is a barrier to business competitiveness and the
deployment of AI solutions and greater interoperability is a
key prerequisite for the development of new AI applications
in the public sector. For example, the study found that one
uniquely European obstacle to the wider uptake of AI
solutions in the mobility sector is the fragmentation of its rail
market causing a lack of interoperability (see section 2.3 on
challenges in the Mobility sector). Initiatives like the
Shift2Rail project were therefore launched to develop a
common interoperability framework at the European level.496
Developing standards, coordinated strategies and
interoperability requirements for data exchanges would bring
numerous benefits to the development of new AI solutions for
the public sector:

 Standards can help to ensure that data is shared

consistently and reliably between different systems,
organisations, and agencies. This can make it easier to
develop new AI solutions that can make use of that
data.
 Coordinated strategies can help to ensure that different
organisations and agencies are working together in a
coordinated way to achieve common goals related to
the development and use of AI in the public sector.
This can help to increase the efficiency and
effectiveness of AI development efforts.
 Interoperability requirements can ensure that new AI
solutions can be integrated with existing systems and
work seamlessly together. This can help to minimize
disruptions and maximize the benefits of new AI
solutions to the public sector. Interoperability can
foster the democratisation of data and AI across
different domains, allowing smaller organisations and
agencies to benefit from the same data and AI tools as
larger ones.

Feasibility/ High These types of actions focused on increasing interoperability

are already being put in place at the national and EU level,
limits
therefore their level of feasibility is expected to be high.
The Commission has recently adopted the Interoperable
Europe Act497 to strengthen cross-border interoperability and
cooperation in the public sector across the EU. It aims to
implement interoperability by design approach and calls for
mandatory interoperability assessments to evaluate the

496
See https://rail-research.europa.eu/about-shift2rail/mission-and-objectives/
497
COM(2022) 720 final

211
 Adopt AI Study

impact of changes in IT systems related to cross-border

interoperability in the EU.
The SEMIC Support Centre498 is an EU initiative aimed at
building an interoperable European administration. In line
with the Interoperable Europe Act, the SEMIC community
facilitates the co-creation of technical solutions and sharing of
best practices. The Support Center is also home to a
community that shares tooling, standards and good practices
to promote semantic interoperability, including for the
common European data spaces.499 The recent 2022 SEMIC
Conference, for example, offered support in overcoming
obstacles in implementing data spaces and aimed to clarify
how interoperability within and across sectors can be
achieved.500
For example, ‘Core Vocabularies’ is one EU initiative designed
to provide simplified, reusable, and extensible data models
that capture the fundamental characteristics of a public
organisation, in a context-neutral manner.501 Open e-TrustEx
is an open-source solution which provides a secure file
exchange platform for European public administrations502.
Many other such technical solutions are available on the
Joinup platform, which is the European Commission's one-
stop shop for interoperable, open and free digital government
ICT solutions.503
Efficiency/ Medium Interoperable digital public services are essential for building
a digital single market. Apart from the economic benefits and
resources
efficiency gains, case studies (see section 2.3 on the Mobility
Data Space in Germany) show that interoperability positively
affects public values, such as improving trust from citizens in
their governments.
In the past years, digital government and data experts have
developed wide-ranging common interoperability cooperation
practices based on the European Interoperability Framework
(EIF). However, recent evaluations have exposed the serious
limitations of this entirely voluntary cooperation approach.504
For this purpose, the Commission adopted in November 2022
the Interoperable Europe Act proposal to strengthen cross-
border interoperability and cooperation in the public sector
across the EU and it is expected to save costs, and cross-
border interoperability can lead to cost-savings between €5.5
and €6.3 million for citizens and between €5.7 and €19.2
billion for businesses dealing with public administrations.505
Any measures designed to increase interoperability would
therefore have a high positive impact on the uptake of AI
solutions in the public sector as well. However, measures
improving interoperability could be expected to incur

498
See SEMIC Support Centre | Joinup (europa.eu)
499
See https://joinup.ec.europa.eu/collection/semic-support-centre/data-spaces
500
See https://semic2022.eu/
501
See https://ec.europa.eu/isa2/solutions/core-vocabularies_en/
502
See https://ec.europa.eu/isa2/solutions/open-e-trustex_en/
503
See https://joinup.ec.europa.eu/collection/joinup/interoperability-solutions
504
Press Release, November 2022, New Interoperable Europe Act to deliver more efficient public services
through improved cooperation between national administrations on data exchanges and IT solutions
available at https://ec.europa.eu/commission/presscorner/detail/%20en/ip_22_6907
505
European Commission Press Release, 2022, New Interoperable Europe Act to deliver more efficient
public services through improved cooperation between national administrations on data exchanges and IT
solutions available at https://ec.europa.eu/commission/presscorner/detail/en/ip_22_6907

212
 Adopt AI Study

significant costs in the short term for public administrations

even though they are offset over the longer term by the
benefits generated. Such measures would be built on already
existing frameworks developed at the EU level and could be
funded by EU funding instruments such as the Digital Europe
Programme.
EU added Medium In the past years, digital government and data experts have
value/ developed wide-ranging common interoperability cooperation
practices based on the current European Interoperability
subsidiarity
Framework (EIF) and will continue to do so under the
Interoperable Europe Act, which may see the revision of the
EIF.
The actions could be implemented both at the EU level and
MS level. However, certain measures such as interoperability
standards and guidelines developed at the EU level could
have an increased added value over the national solutions as
they would ensure higher levels of technical interoperability
across the whole EU and would have a wider outreach.
The objectives to ensure a coherent EU approach and EU-
wide interoperability for data exchanges and a dynamic data
ecosystem cannot be achieved by Member States' action
alone but need EU action.

Facilitate “data ownership”

Potential policy actions:
 Define data control, access and other processes in relevant legislation (GDPR,
Data Governance Act and the Data Act).
 Engage in public discussion, education and seminars around “ownership” and
usage of individuals' data to promote greater understanding by individuals of
how their data may be used.

Criteria Rating Description

Effectiveness/ Medium The study found that often in the public sector, there is a
reluctance towards data sharing due to challenges related to
benefits
“data ownership” and “data sovereignty” or because of
insufficient tools and appropriate digital infrastructures to
permit data sharing more easily (see Section 1.3.1.2. on data
challenges).
Rules on the B2B and B2G side of data exchanges and
interoperability have already been pushed forward through
legislation such as the Data Act and the Data Governance Act
as part of the European Strategy for Data. They are expected
to have a high impact on creating a single market for data
that will ensure Europe’s global competitiveness and “data
sovereignty”.
Decentralised data architectures, for example, can provide
several benefits for the adoption of AI solutions:

 Data privacy and security: Decentralised data

architectures distribute data across multiple nodes,
rather than storing it in a central location. This can
make it more difficult for unauthorized parties to
access or tamper with the data, which can be
especially important in applications that involve
sensitive or personal information.

213
 Adopt AI Study

 Scalability: They can help to ensure that the system

can handle large amounts of data, even as the number
of users or the amount of data grows.
 Interoperability: Decentralised systems typically use
open standards and protocols, which makes it easier
for different systems to communicate and share data.
This can be useful in situations where multiple AI
solutions need to work together or share data.
 More accessible AI: Decentralised architecture allows
data to be shared with a wider population, which could
enable more groups or individuals to have access to
the data, allowing more inclusive AI development.
Nevertheless, most of the actions envisaged under this
recommendation are soft law measures intended to increase
clarity and awareness and encourage more data sharing,
hence their effectiveness can be expected to be lower.
Benefits might also be indirect and difficult to capture in
practice.
Feasibility/ High As the actions envisaged under this recommendation are soft
law measures intended to increase clarity over “data
limits
ownership” and awareness and to encourage more data
sharing, their level of feasibility should be high with very few
possible limits endangering their implementation. Moreover,
the actions could build on already existing initiatives based
on the European Data Strategy.
For example, Gaia-X is a project whereby representatives
from business, politics, and science from Europe and around
the globe are working together, hand in hand, to create a
federated and secure data infrastructure.506 The architecture
of Gaia-X is based on the principle of decentralization.
Similarly, the study also explored the case study of the
Mobility Data Space from Germany with its possible
implications for the development of AI solutions for the public
sector (see Section 2.3.4).
Efficiency/ Medium The necessary resources required for the actions envisaged
would be minimal as these would build on already existing
resources
frameworks and solutions. Moreover, the actions are mostly
designed to clarify certain concepts, encourage data sharing,
and create awareness over existing solutions, therefore they
can be expected to not require high implementation costs or
technical investments.
Nevertheless, for measures such as decentralized data
architectures the trade-offs also need to be evaluated
carefully, as these could also bring challenges such as higher
complexity, increased maintenance costs and lower
performance in some cases.
EU added Medium All actions could be developed both at the EU and national
value/ levels, however, if implemented and coordinated at the EU
level, they could support the creation of a data ecosystem
subsidiarity
based on European values and can provide input to
discussions on innovation at the European level.

506
Further information is available at: https://gaia-x.eu/what-is-gaia-x/

214
 Adopt AI Study

Address the challenges surrounding the sharing of data between and across
sectors
Potential policy actions:
 Introduce guidelines as to what is and isn't legally possible and data sensitivity.
 Encourage sharing of data between different data spaces.
 Ensure that confidential data (business secrets) are end-to-end encrypted and
promote privacy-preserving techniques.
 Promote common sharing or format and metadata standards.
Criteria Rating Description
Effectiveness/ Medium On top of challenges related to a lack of interoperability and
uniform data quality and unharmonised rules for cross-border
benefits
data flows, a differentiated GDPR implementation has also
been signalled as an important issue influencing data sharing
across sectors. In the case of AI/big data applications, the
uncertainties are aggravated by the novelty of the
technologies, their complexity and the broad scope of their
individual and social effects.
Further clarifying GDPR and other data privacy rules as well
as introducing guidelines which are specifically aimed at
specifying the obligations of actors implementing AI solutions
would create a better understanding of existing legislation
and have indirect positive effects on the level of data sharing
occurring in practice, but these are expected to be limited on
the uptake of AI by the public sector.
Encouraging the sharing of data between different data
spaces can provide several benefits for the adoption of AI
solutions by the public sector:

 Increased efficiency: Sharing data between different

data spaces can allow for more efficient use of
resources since multiple organisations can access and
use the same data instead of each having to collect
and maintain their separate datasets.
 Improved decision-making: Sharing data can allow for
a more accurate and comprehensive analysis of the
data, which can lead to better decisions. By pooling
data from multiple sources, organisations can gain a
more complete understanding of the issues they are
addressing.
 Better public services: By sharing data, public sector
organisations can more easily identify patterns and
correlations that can help to improve the delivery of
public services.
 Cost savings: Sharing data can also help organisations
save money since they won't have to spend as much
on collecting and maintaining their datasets.
Additionally, by sharing data, organisations can reduce
duplication of effort and avoid having to perform the
same analysis multiple times.
 Innovation: Sharing data can also lead to new insights
and innovations that would not have been possible with
data siloed in one organisation. Exposing data to the
wider population can enable new uses and discoveries,
which can drive innovations.

215
 Adopt AI Study

Feasibility/ High Rules on the B2B and B2G side of data exchanges and
interoperability have already been pushed forward through
limits
legislation such as the Data Act and the Data Governance Act
as part of the European Strategy for Data. They are expected
to have a high impact on creating a single market for data
that will ensure Europe’s global competitiveness and “data
sovereignty”.
Given that the actions envisaged under this recommendation
are soft law measures intended to increase clarity over data
privacy regulation and awareness of existing solutions and to
encourage more data sharing, their level of feasibility should
be high with very few possible limits endangering their
implementation. Moreover, the actions could build on already
existing initiatives based on the European Data Strategy.
Efficiency/ Medium The necessary resources required for the actions envisaged
would be minimal as these would build on already existing
resources
frameworks and solutions. Moreover, the actions are mostly
designed to clarify certain concepts, encourage data sharing,
and create awareness over existing solutions, therefore they
can be expected to not require high implementation costs or
technical investments.
It's also important to note that increased data sharing also
comes with a set of challenges such as privacy concerns,
security, legal and ethical issues, and technical challenges
around data integration, all of which need to be addressed
through appropriate governance, regulations, and technical
solutions.
For example, for the implementation of end-to-end
encryption and privacy-preserving techniques for data
sharing, there are several potential investments necessary
including:

 Hardware and infrastructure costs are associated with

acquiring and maintaining the necessary hardware and
infrastructure to support encryption and privacy-
preserving techniques.
 Development and implementation costs, including costs
for software development, testing, and deployment.
 Training and support costs: Organizations may need to
provide training and support to their employees and
users to ensure they understand how to properly use
and maintain the systems.
 Maintenance and management costs.
 Legal and regulatory compliance costs depend on the
specific regulations and laws that apply to the data
being shared.

EU added Medium All actions could be developed both at the EU and national
value/ levels, however, if implemented and coordinated at the EU
level, they could support the creation of a data ecosystem
subsidiarity
based on European values and can provide input to
discussions on innovation at the European level.
Providing guidance requires a multilevel approach, which
involves data protection authorities, civil society,
representative bodies, and all other stakeholders. The
guidance provided by political authorities, such as the
European Parliament, the European Commission and the
European Data Protection Supervisor could have a higher
level of effectiveness and wider outreach. Nevertheless,
National Data Protection Authorities should also provide

216
 Adopt AI Study

guidance, in particular when contacted for advice by

controllers, or in response to data subjects' queries.507

Address bias within AI and data sources

Potential policy actions:
 Promote ethics by design (interdisciplinary awareness/competence in
developing/procuring teams).
 Develop tools and promote education to make understanding bias a standard
part of AI workflow.
 Have consistent monitoring and evaluation of bias.
Criteria Rating Description
Effectiveness High The study found that in the public sector especially when
/ procuring a technology, the potential for bias and discrimination
in AI is an important factor to address during the
benefits
procurement/development process (see section 1.3 on AI
Technology challenges). The question of understanding who
bears responsibility and liability for AI technology is also
extremely important in ensuring confidence in AI technologies
and therefore increases its likelihood to be adopted by public
administration.
The study also showed that several best practices can already
be highlighted for higher uptake and avoiding social rejection of
AI solutions: greater involvement of stakeholders, experts,
workers and citizens that will be facing/using the AI, and
enhancement of digital infrastructure, or the creation of an “AI
Data Officer” (see eGovernment section on savings in
operational efficiency).508
The actions envisaged under this recommendation have the
potential to tackle some of the issues surrounding bias within
data sources and data science teams as the concept of ‘ethics by
design’ is developed at the EU level through various initiatives.
For example, guidance has been drafted by a panel of experts509
and builds on the work of the Independent High-Level Expert
Group on AI and their ‘Ethics Guidelines for Trustworthy AI’510 as
well as on the results of the EU-funded SHERPA511 and SIENNA
projects.512 All these initiatives have shown that embedding
principles of ethics early on in the implementation of AI can
bring numerous benefits, such as more fairness, transparency,
accountability and oversight, which in turn enhance trust in AI
systems.
For high-risk AI systems, this will be addressed by the upcoming
AI Act (Articles 9 and 10) and the conformity and monitoring
procedures. For non-high risk, there will be the option of Article
69 with codes of conduct513.

507
EPRS (2020) The impact of the General Data Protection Regulation (GDPR) on Artificial intelligence
508
AI Multiple, April 2022, AI in government: examples, challenges & best practices in 2022,
https://research.aimultiple.com/ai-government/
509
European Commission (2021) Ethics By Design and Ethics of Use Approaches for Artificial Intelligence
510
See more at https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai
511
https://www.project-sherpa.eu/
512
https://www.sienna-project.eu/
513
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence

217
 Adopt AI Study

Feasibility/ High The actions envisaged within this recommendation would have a
high level of feasibility given that they would build on existing
limits
initiatives and frameworks.
Most importantly, the EU AI Act would provide the necessary
legal framework to implement monitoring and enforcement
structures as, under the Regulation, public authorities would be
required to have appropriate human resources and technical
tools when implementing AI systems, especially when these
technologies would fall under the high-risk categories
established by the act. This will apply to risk management, data
quality and bias testing and auditing. There will also be an
expectation that the private actors who procure the systems
comply with these requirements (they will be providers under
the AI Act and will directly be obliged under the AI Act).
Efficiency/ High The actions envisaged within this recommendation would have a
high level of efficiency given that they would build on existing
resources
initiatives and frameworks. Nevertheless, the field of data
science can be expensive, both in terms of education, training
and of technology and software, which can be a barrier for
organisations in obtaining representation in their data science
teams.
EU added High All actions could be developed both at the EU and national
value/ levels, however, if implemented and coordinated at the EU level,
they could support the creation of a data ecosystem based on
subsidiarity
European values and can provide input to discussions on
innovation at the European level.
The guidance provided by EU authorities, such as the European
Parliament, or the European Commission could have a higher
level of effectiveness and wider outreach.
The AI Act requires post-market monitoring plans (Article 61)
that providers must establish to document the performance of
high-risk AI systems throughout their lifetimes. This can form
the basis for an EU-wide ecosystem for conducting AI auditing.

Improve technological accountability and transparency

Potential policy actions:
 Use transparency tools along EU guidelines.
 Promote certification of AI on all assurance levels and conformity assessment of
transparency requirements, ideally via independent third parties for high-risk
areas.

Criteria Rating Description

Effectiveness/ High The study found that the transparency of the decisions made
by algorithms is essential to make AI in procurement
benefits
democratic and acceptable to European citizens (see section
1.3 on AI technology challenges).514 The EU’s High-Level
Group on AI has also called for further work to define
pathways to achieving explainability.515

514
World Economic Forum, June 2020, AI Procurement in a Box: Procurement guidelines,
https://www3.weforum.org/docs/WEF_AI_Procurement_in_a_Box_AI_Government_Procurement_Guideli
nes_2020.pdf
515
EU High Level Group on AI (2019) Ethics guidelines for trustworthy AI. Available at:
https://ec.europa.eu/digital-singlemarket/en/news/ethics-guidelines-trustworthy-ai

218
 Adopt AI Study

Transparency tools and ensuring the transparency of AI

systems would be beneficial for several reasons: they would
give users confidence in the system, they would safeguard
against bias, would help ensure regulatory standards or
policy requirements are met and would improve system
design.516
Feasibility/ High The actions envisaged within this recommendation would
have a high level of feasibility given that they would build on
limits
existing initiatives and frameworks.
A range of policy instruments already exists that seek to
promote or enforce some form of transparency in the use of
data and AI. For example, the JRC has stressed the
possibility to develop methodologies to evaluate the impacts
of AI on society, built on the model of the Data Protection
Impact Assessments (DPIA) introduced in the General Data
Protection Regulation (GDPR). Another possible solution was
also the promotion of transparency systems in sensitive
systems, through the implementation of transparency-by-
design approaches in AI components that would provide a
guarantee of the respect of fundamental rights.517
The path for transparency mechanisms is also set by the
proposed EU AI Act. Specifically, the AI Act sets some new
minimum requirements of transparency and interpretability)
for AI systems labelled as “high-risk”. Transparency tools and
metrics would be built at the EU level to check the technical
documentation required by the AI Act and, in turn, verify the
extent to which the AI system is aligned with the
requirements of the regulation.
Efficiency/ Medium The actions envisaged within this recommendation would
have a medium level of efficiency given that they would build
resources
on existing initiatives and frameworks. However, designing
and implementing transparency tools which go beyond
guidelines as well as developing industry standards could
require certain significant resources and investment.
One example of an ongoing project is NL4XAI, the first
European Training Network on Explainable Artificial
Intelligence funded by Horizon 2020, which aims to train 11
researchers to leverage the usage of AI models and
techniques and integrate their findings into a common open-
source software framework for Explainable AI that will be
accessible to all the European citizens.518

516
The Royal Society (2019) Explainable AI: the basics – policy briefing available at
https://ec.europa.eu/futurium/en/system/files/ged/ai-and-interpretability-policy-
briefing_creative_commons.pdf
517
JRC (2020) Robustness and Explainability of Artificial Intelligence
518
See https://nl4xai.eu/about/

219
 Adopt AI Study

EU added Medium Different users require different forms of explanation in

value/ different contexts, therefore transparency tools could be
difficult to harmonise across sectors or MS.
subsidiarity
However, in terms of the development of standards, their
value stems from their high level of harmonization at the EU
level. For this purpose, according to the proposal for the EU
AI Act, the European Committee for Standardisation (CEN)
and the European Committee for Electrotechnical
Standardisation (CENELEC) will be in charge of developing
the technical standards for the AI Act. Technical standards
will play a key role in the implementation of the AI Act, as
the companies that apply them will be considered by default
in conformity with the EU rules. Nevertheless, this will only
be finalised once the AI Act has been finalised and by the EU
co-legislators, which is unlikely before 2023.

Increase trust in AI solutions

Potential policy actions:
 Adopt and prepare for the implementation of the AI regulatory framework,
including the AI Act and the AI Liability Directive.
 Use storytelling to communicate the value and the worth of AI solutions.
 Promote testing, inspection, and certification.
 Promote greater education on AI at all levels of society.
 Promote the adoption of ethical frameworks for AI.
Criteria Rating Description
Effectiveness/ Medium The study found that some of the main barriers to the
adoption of AI technologies are related to social aspects. In
benefits
this context, it is essential to lessen the possibility of
decreasing citizens’ trust towards public authorities, by
properly addressing and minimising the dehumanisation of
public services and attracting experts and finding synergies
with the ICT industry (see Section on main drivers and
barriers for the E-Government sector).
Rules proposed by the AI Liability Directive, which will come
after the AI Act, will promote trust in AI (and other digital
technologies) by ensuring that victims are effectively
compensated if damage occurs despite the preventive
requirements of the AI Act and other safety rules. The
proposal for the AI Act seeks to establish mechanisms to
preserve fundamental rights and values throughout the whole
life cycle of AI-based systems, thus ensuring legal certainty
that encourages innovation and investments in AI systems.
On top of these frameworks, any measures increasing
societal trust would benefit all actors involved in the AI-value
chain, because strengthening citizens’ confidence will
contribute to faster uptake of AI.
Feasibility/ High The actions envisaged within this recommendation would
have a high level of feasibility given that they would build on
limits
existing initiatives and frameworks.

220
 Adopt AI Study

Efficiency/ High While relying as much as possible on existing frameworks

and solutions to minimize costs, the mandatory nature of the
resources
AI Act and Liability Directive would still require some
resources to follow all procedures and ensure compliance
which public authorities will need to prepare for.
EU added Medium The role of both the Liability Directive and the AI Act is to
value/ contribute to the rollout of AI. The conditions for the roll-out
and development of AI technologies in the internal market
subsidiarity
can be significantly improved by preventing fragmentation
and increasing legal certainty through harmonised measures
at the EU level, compared to possible adaptations of liability
rules at a national level.
The actions could be implemented both at the EU level and
national level, however, guidance or awareness campaigns
provided by EU Commission and the future EU AI Board could
have a higher level of effectiveness and wider outreach.

Facilitate the integration of new AI technologies/services into existing

systems (education/health/transport)
Potential policy actions:
 Create regulatory sandboxes
 Aim for a longer cycle of integration because this happens with any new IT.
 Address security considerations by undertaking risk-adequate controlled
experiments via new solutions.
 Use "data ambassador" campaigns to better involve stakeholders.
 Work alongside existing systems as a support rather than immediately replacing
existing systems.

Criteria Rating Description

Effectiveness/ High Difficulties in integrating new AI technologies/services into
existing systems have been reported in several sectors
benefits
analysed by the study. For example, in the Health sector, this
comes from a lack of complementary solutions to enable the
use/uptake of AI MedTechs (e.g., digital medical records
require new integration software and platforms) (see Section
2.4.2 on the Health sector’s value chain analysis).
Improving the integration of new AI technologies/services
into existing systems rather than immediately replacing
existing systems has several benefits:

 Cost-effectiveness: Integrating AI into existing systems

rather than replacing them altogether can save a
significant amount of money.
 Familiarity: Employees and users are already familiar
with existing systems, so integrating AI into those
systems will cause less disruption and make the
transition smoother.
 Data compatibility: Integrating AI into existing systems
allows the new technology to work with the same data
sets and interfaces that are already in use.
 Scalability: Integrating AI into existing systems allows
for incremental implementation, making it easier to
scale up as needed.

221
 Adopt AI Study

 Flexibility: Integrating AI into existing systems allows

organisations to pick and choose which processes and
tasks they want to automate, rather than having to
replace entire systems.
 Reduced risk: Integrating AI into existing systems
rather than replacing them altogether can help
organisations to reduce the risk of data loss, system
failure or other unexpected problems.

Feasibility/ High The actions envisaged within this recommendation would

have a high level of feasibility given that they would build on
limits
existing initiatives and frameworks. Moreover, they are soft
law measures intended to increase awareness, promote good
practices and enhance the testing of AI solutions and their
integrations within existing systems, which suggests that
there would not be significant limits to their implementation.
As regards the measure aimed at promoting the undertaking
of risk-adequate controlled experiments via new solutions, it
is worth noting that the proposal for the EU AI Act envisages
setting up coordinated AI 'regulatory sandboxes' to foster
innovation across the EU. A regulatory sandbox is a tool
allowing businesses to explore and experiment with new and
innovative products, services or businesses under a
regulator's supervision. It provides innovators with incentives
to test their innovations in a controlled environment, allows
regulators to better understand the technology, and fosters
consumer choice in the long run.519
Efficiency/ Medium The necessary resources required for the actions envisaged
would be minimal as these would build on already existing
resources
frameworks and solutions. Moreover, the actions are mostly
designed to clarify certain concepts, encourage testing,
inspecting and certification, and create awareness over
existing solutions, therefore they can be expected to not
require high implementation costs or technical investments.
Nevertheless, improving the integration of new AI
technologies/services into existing systems might require
several investments related to auditing and testing the
existing systems, preparing them for integration with AI
systems as well as the actual integration costs, which relate
to measures needed to make sure that existing and new
systems work together seamlessly.
undertaking risk-adequate controlled experiments via new
solutions and working alongside existing systems as a
support rather than immediately replacing existing systems
are also both measures that can reduce overall resources
needed for the implementation of AI technologies. They allow
for faster deployment by allowing organisations to test new
solutions in a controlled environment before fully
implementing them.
EU added High The actions could be implemented both at the EU level and
value/ national level, however, guidance or awareness campaigns
provided by EU authorities could have a higher level of
subsidiarity
effectiveness and wider outreach.
For example, developing testing and regulatory sandboxes at
the EU level rather than at the national level can facilitate
cross-border deployment of new AI technologies across the

519
European Parliament (2022) Briefing on Artificial intelligence act and regulatory sandboxes available at
https://www.europarl.europa.eu/RegData/etudes/BRIE/2022/733544/EPRS_BRI(2022)733544_EN.pdf

222
 Adopt AI Study

EU, therefore promoting the Single Market and the EU's

digital agenda. They can also provide consistency in the way
that new AI technologies are tested and regulated, which can
reduce uncertainty for companies operating in multiple
member states and can allow EU Member States and can lead
to greater efficiencies by avoiding duplication of efforts.

Focus on long-term implementation

Potential policy actions:
 Fund multi-disciplinary projects with a focus on sustainability over time.
 Avoid “project silos” and “reinventing the wheel” each time and promote a more
modular longer-term approach. The overarching purpose should be a long-term
solution for citizens involving real-world solutions with a service mentality.
 Promote, invest and facilitate (e.g., via legislation on data sharing) open-source
solutions and libraries.
 Make policy with input from representative groups of users.
Criteria Rating Description
Effectiveness/ High Involving multi-disciplinary teams, encouraging collaboration,
and focusing on implementation over the longer term has
benefits
several benefits:

 Diverse perspectives: Multi-disciplinary teams bring

together individuals with different backgrounds, skills,
and expertise, which can lead to more creative and
effective solutions.
 Increased efficiency: Collaboration allows team
members to share their knowledge and work together
to achieve a common goal, resulting in faster and more
efficient implementation of AI technologies.
 Improved decision-making: By involving diverse
experts, decision-making can be based on a more
comprehensive understanding of the problem at hand,
leading to better and more informed decisions.
 Better alignment with business goals: By focusing on
implementation over the longer term, teams can
ensure that AI technologies are being used in a way
that aligns with the organisation's overall goals and
objectives.
 Better generalisation: By involving experts from
different fields, the model can be trained on more
diverse data, which can improve its generalisation
capabilities.
 Better ethical consideration: Multi-disciplinary teams
can better consider the ethical implications of AI
technology, and help to mitigate the negative
consequences

Feasibility/ Medium The actions envisaged within this recommendation would

have a high level of feasibility given that they would build on
limits
existing initiatives and frameworks. Moreover, they are soft
law measures intended to increase awareness, promote good
practices and support the implementation of AI solutions by
the public sector with a focus on the long term, which
suggests that there would not be significant limits to their
implementation.

223
 Adopt AI Study

The Digital Europe Programme, as well as the EU Recovery

and Resilience Facility (RRF) programme, include funding
opportunities dedicated to AI for the public sector.
Efficiency/ High The necessary resources required for the actions envisaged
would be reasonable as these would build on already existing
resources
frameworks and solutions. However, the actions described
under this set of recommendations are related to funding,
providing financial incentives and establishing programmes
that would ensure long-term involvement in AI
implementation projects. As a consequence, the actions can
be expected to require significant implementation costs or
technical investments.
It is possible to improve efficiency by, for example, including
scalability as a formal precondition to benefit from EU funding
programmes foreseeing AI-related activities520. There is also
a need to ensure resources are appropriately allocated as a
part of this action, here, to ensure funding schemes and
initiatives are optimised an initial focus could be placed on
those specific innovations with a clear focus on replicability
and scalability over the long term.
EU added High EU-level funding adds significant value added in
value/ complementing national funding through:
subsidiarity  Scalability - enhancing the development and
implementation of AI technologies on a larger scale,
across multiple countries,
 Harmonisation: helping to harmonise the development
and implementation of AI technologies across different
countries, leading to greater consistency and
interoperability,
 Access to expertise: providing access to a wider pool of
expertise and resources, which can support the
development and implementation of AI technologies,
 International cooperation: encouraging international
cooperation and the sharing of best practices among
different countries, leading to a more effective and
efficient implementation of AI technologies.

Improve access to funding and resources

Potential policy actions:
 Strengthen current initiatives that provide funding and resources, such as EIC
Accelerator and Horizon 2020.
 Establish new cascading funding and grants for research and development of AI
solutions for the public sector.

Criteria Rating Description

Effectiveness/ High A lack of funding is a central barrier to AI-enabled public
sector innovation521. Many national and EU-level projects
benefits
require significant funding levels to ensure the appropriate
resourcing for the designing and developing of AI solutions.

520
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1
521
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1

224
 Adopt AI Study

This funding is particularly needed for replicating and scaling

up best practices.
Whilst a range of measures here focuses on the availability of
financial resources, the reputation and position of the
Commission are also important in facilitating engagement
and partnerships between the public sector and industry. The
study found that there are several clear examples and cases
(Case Studies 2.2.4, 2.3.4) where multi-disciplinary teams
within sectors are effective in breaking down silos of both
knowledge and data and bridging the gap between the public
and private sectors. Similarly, such ventures demonstrated
the value of user-led, rather than technology-led design.
Feasibility/ High The Digital Europe Programme, as well as the EU Recovery
and Resilience Facility (RRF) programme, include funding
limits
opportunities dedicated to AI for the public sector.
Build on these programmes and promote the use of
international funding together with national, regional and
local funding in a complementary manner522. In this instance,
a range of existing projects already exists that could be
chosen for further funding to help scale up good AI practices
at the national and EU level. Such action would help spread
contextual experiences across settings (local-regional,
country-country) as well as help scale the system up from
local to central units of administration. Being led by the
Commission in this instance would help ensure alignment
with European regulations and values.
Efficiency/ High It is possible to improve efficiency by, for example, including
scalability as a formal precondition to benefit from EU funding
resources
programmes foreseeing AI-related activities523.
There is also a need to ensure resources are appropriately
allocated as a part of this action, here, to ensure funding
schemes and initiatives are optimised an initial focus could be
placed on those specific innovations with a clear focus on
replicability and scalability.
EU added High EU-level funding adds significant value added in
value/ complementing national funding through:
subsidiarity  Scalability - enhancing the development and
implementation of AI technologies on a larger scale,
across multiple countries,
 Harmonisation: helping to harmonise the development
and implementation of AI technologies across different
countries, leading to greater consistency and
interoperability,
 Access to expertise: providing access to a wider pool of
expertise and resources, which can support the
development and implementation of AI technologies,
 International cooperation: encouraging international
cooperation and the sharing of best practices among
different countries, leading to a more effective and
efficient implementation of AI technologies.

522
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1
523
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1

225
 Adopt AI Study

Foster the development of skills and competencies

Potential policy actions:
 Introduce digital Academy-type models for the public sector where AI is part of
a curriculum.
 Develop and maintain digital skills within the workforce through a centralised
training portal.
 Improve digital literacy through the establishment of communities of practice,
or peer learning groups across Member States.

Criteria Rating Description

The study found that the lack of skills and expertise is one of
Effectiveness/ High
the most significant barriers to AI adoption (Section 1.3.1.4 on
benefits organisational capacity challenges). The successful adoption
and deployment of AI solutions require a mixture of technical
and socio-technical skills to ensure an understanding of the
technology itself, and its potential524.

Employees often do not possess the necessary skills, and it is

often challenging for public administrations to identify, recruit,
and retain the required talent. The use of trusted training
packages is one way of building skills and knowledge within
the workforce, and are effective as the case study of AI MOOC
in Section 2.5.4 demonstrated.

Addressing the lack of digital skills and competencies can bring

several benefits:

 Improved implementation: individuals and

organisations will have the necessary skills and
knowledge to effectively use and manage these
technologies,
 Increased efficiency: can lead to more efficient use of
AI technologies, as individuals and organisations will be
able to better understand and utilize these technologies
to achieve their goals,
 Greater innovation: individuals and organisations will
be better equipped to develop new and creative uses
for AI technologies,
 Better decision-making: better ability to analyse and
interpret the data generated by AI technologies,
 More effective use of resources: better utilisation of AI
technologies to achieve their goals.
 Better security: individuals and organisations will be
better equipped to identify and mitigate potential
security risks associated with AI technologies.
 Greater transparency: individuals and organisations
will be better able to understand and explain the
decision-making processes and outcomes generated by
AI technologies.

Feasibility/ High Given that the three actions under this recommendation are all
currently being implemented at the EU and Member State level
limits
in various ways, their feasibility has already proven to be high.

524
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1

226
 Adopt AI Study

For addressing existing workforce skills there is a range of

existing and effective databases, initiatives, and platforms
across the EU such as Digital Skill Up525, the Digital Skills &
Jobs platform526, and the ALL-DIGITAL academy527. Moreover,
Master’s programmes are also funded through the Connecting
Europe Facility (CEF) Telecom programme. Under this
initiative, top-level universities, SMEs and research centres
joined forces to design Master programmes on AI, focusing on
human-centric AI, AI ethics, AI for the public sector and AI in
healthcare. The programmes follow a highly practical approach
that will offer hands-on experiences, and opportunities to work
closely with business partners and to learn from real case
studies.528
The potential limit here is down to the training provision itself.
Certain industries may require more tailored, or highly curated
packages. These more technical training offers may fall outside
the scope of this recommendation, but more work can be done
on the provision and dissemination of general skills
development pathways to help increase the basic digital skills
of the population.
AI professionals are relatively scarce in the global market, as
the demand outweighs the supply529. In addition, such
professionals tend to prefer engagement in the private sector,
given its higher benefits. The higher salaries expected by AI
experts are also one of the main variables behind the high cost
of adopting AI solutions in the public sector.
Efficiency/ High As mentioned, several solutions already exist that could be
leveraged at low cost and to increase efficiency. Platforms
resources
such as the ‘elements of AI’ MOOC can continue to be used to
assist in educating citizens on the basics of AI. If the
commission wished to produce or procure further, sectoral-
specific training packages through these platforms then this
would increase the cost. Similarly, any expansion of existing
systems would incur additional costs, but this would be less
than the full production of such systems Currently the
resource costs are needed. to secure the appropriate AI
professionals is high, and the expansion of trusted training
platforms here would help over the medium term.
EU added Low Efforts to address the lack of skills and competencies should
value/ be undertaken at both the EU and national levels. However,
given that Education is a national competence, it is expected
subsidiarity
that the EU added value is rather low and that Member States
would wish to tailor measures and initiatives to the specifics of
their national educational systems, especially in what concerns
school and higher education.
The EU added value in this instance would focus on the use of
a centralized and trusted training platform that could ensure
shared training across the Member States which is important
for establishing a common understanding of the topics, as well
as ensuring alignment with the European Digital Decade policy
programme.

525
See https://www.digitalskillup.eu/catalog/
526
See https://digital-skills-jobs.europa.eu/en/opportunities/training
527
See ALL DIGITAL Academy – Erasmus+ Project
528
See https://digital-skills-jobs.europa.eu/en/artificial-intelligence-masters-supported-cef
529
JRC (2022) AI Watch: Road to the Adoption of Artificial Intelligence by the Public Sector. Available at:
https://op.europa.eu/en/publication-detail/-/publication/34251428-dc12-11ec-a534-01aa75ed71a1

227
 Adopt AI Study

Ranking of policy recommendations

The table below provides a ranking of the most important recommendations based on
the analysis above. It must be noted that the ranking is based on mostly qualitative
evidence. To have a more systematic quantitative assessment and ranking, a full
impact assessment would be needed, with a dedicated cost-benefit analysis. This was
beyond the scope of the current study.

228
 Adopt AI Study

Table 13 Ranking of policy recommendations

Effectiveness/ Feasibility/ Efficiency/ EU added value/

Recommendations
Benefits limits resources subsidiarity Ranking

Improve access to funding and resources High High High High 1

Address bias within AI and data sources High High High High 2

Foster coordinated procurement strategies High High High High 3

Increase clarity and harmonisation around cross-border data flows High High High Medium 4

Improve alignment between industry and public sector expectations High High High Medium 5

Ensure a clear regulatory framework for AI High Medium High High 6

Facilitate the integration of new AI technologies/services into existing systems

High High Medium High 7
(education/health/transport)

Focus on long-term implementation High Medium High High 8

Foster the development of skills and competencies High High High Low 9

Promote interoperability High High Medium Medium 10

Increase expertise among public buyers High High Medium Medium 11

Improve technological accountability and transparency High High Medium Medium 12

Increase trust in AI solutions Medium High High Medium 13

Facilitate “data ownership” Medium High Medium Medium 14

Address the challenges surrounding the sharing of data between and across sectors Medium High Medium Medium 15

Source: Authors’ elaboration.

229
 Adopt AI Study

5 Conclusion
The comparison and triangulation of recommendations from the policy workshop and
the online survey provide the following set of recommendations in order of
importance. For each recommendation, we note the recent policy actions that have
been taken and the potential policy actions that could be taken by the European
Commission to increase the uptake of AI by the public sector in Europe.
It's worth mentioning that these policy recommendations and potential actions are not
exhaustive and might vary depending on the context and the specific needs of each
country or sector. Also, it's important to note that some of these policy actions were
proposed in recent publications of the European Commission and might not have been
adopted yet.
1. Increase funding and resources for AI in the public sector:
Potential policy actions the Commission could take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Establishing a dedicated AI fund to support AI projects in the public sector and
provide grants for public sector organizations to develop and implement AI
solutions.
• Co-funding AI projects in the public sector together with private sector
companies, to encourage collaboration and knowledge sharing.
On-going activities:
The European Commission has recently provided funding for research and
development of AI in the public sector through programs such as Horizon Europe, and
the EU Framework Program for Research and Innovation (2021-2027)530. They also
established a dedicated funding stream for AI in the public sector under the European
AI Alliance, to support the scaling and spread of reusable solutions (2018)531. The
Commission set up a European AI Alliance, a network of national AI alliances to
exchange best practices and support the development and uptake of AI in the public
sector (2018)532.
2. Reduce bias within AI and data sources:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Establishing a dedicated AI bias detection and mitigation centre to support the
public sector in identifying and addressing bias in data and AI models.
• Providing training on bias detection and mitigation techniques for data
scientists and other professionals in the public sector.
• Developing a set of common ethical guidelines for data collection,
management, and sharing in the public sector to ensure data quality and
reduce bias533.
• Encouraging the use of interpretable AI methods in the public sector to make
AI models more transparent and accountable for their decisions 534.
On-going activities:

530
https://research-and-innovation.ec.europa.eu/funding/funding-opportunities/funding-programmes-and-
open-calls/horizon-europe_en
531
https://ec.europa.eu/digital-single-market/en/european-ai-alliance
532
Ibid
533
This will be covered in the upcoming AI Act Articles 9 and 10.
534
This will be covered in the upcoming AI Act Article 13.

230
 Adopt AI Study

The European Commission has recently established guidelines and best practices for
data collection, management, and sharing in the public sector to ensure data quality
and reduce bias (2021)535. They funded research on bias detection and mitigation
techniques and invest in training programs to educate data science teams on the
importance and methods of mitigating bias in data and AI models (2021) 536. The
upcoming AI Act537 proposal complements existing Union law on non-discrimination
with specific requirements that aim to minimise the risk of algorithmic discrimination,
in particular concerning the design and the quality of data sets and mandating bias
examination, testing and risk mitigation measures.
3. Encourage coordination of AI procurement strategies:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Establishing a procurement office for AI solutions in the public sector, to
streamline the procurement process and ensure compliance with guidelines and
best practices.
• Developing a set of common criteria and guidelines for procurement of AI
solutions by public sector organizations, to ensure transparency and
accountability in the procurement process.
• Encouraging the use of AI solutions developed by European companies and
SMEs, to support the development of the European AI ecosystem.
• Offering dedicated training and certification programs for public procurement
officers, to ensure they are equipped with the knowledge and skills needed to
evaluate and select AI solutions.
On-going activities:
The European Commission has recently developed guidelines and best practices for
the procurement of AI solutions by public sector organizations (2021). They also aim
to establish a European marketplace for GovTech solutions, to facilitate the
procurement of AI solutions by public sector organizations (2021). The upcoming AI
Act538 proposes a governance system at the Member State level, plus a cooperation
mechanism in European Artificial Intelligence Board. It also envisages mandatory
requirements and procedures to be followed by providers and users of high-risk AI
systems, including public authorities.
4. Increase clarity and harmonization around cross-border data flows:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a common European data space for public sector bodies and their
operators, to facilitate the sharing and reuse of relevant AI datasets and
related registries across Europe.
• Establishing a European data governance framework to ensure that data can be
used and shared across borders in a secure and trusted way.
• Encouraging the development of cross-border data-sharing agreements
between public sector organizations, to support the development and
deployment of AI-enabled services.

535
https://digital-strategy.ec.europa.eu/en/policies/data-governance-act-explained
536
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
537
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence
538
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence

231
 Adopt AI Study

• Encouraging the use of open data standards and common data models in the
public sector, to support the interoperability of AI-enabled services.
On-going activities:
The European Commission has recently proposed several policy actions to ensure the
free flow of data in the EU, including the establishment of a European Data
Governance Act (2021)539 and the adoption of the Data Act (2022) 540. These initiatives
aim to facilitate data sharing across sectors and Member States. The Commission also
proposed regulations to protect personal data and ensure “data sovereignty”, including
the General Data Protection Regulation (GDPR) (2018) 541. The upcoming AI Act542
proposes harmonised rules for the development, placement on the market and use of
AI systems. It also proposes a single definition of AI.
5. Promote alignment between industry and public sector expectations:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Encouraging the development of public-private partnerships to support the
development and deployment of AI-enabled services in the public sector.
• Establishing a dedicated AI observatory to gather, share, and collectively
manage best practices and experiences learned from different stakeholders in
the public sector throughout Europe.
• Developing a set of common guidelines and standards for the development and
deployment of AI-enabled services in the public sector, to ensure alignment
between industry and public sector expectations.
• Encouraging the participation of public sector organizations in international AI
initiatives and standards development.
On-going activities:
The European Commission has recently established the European AI Alliance, a
network of national AI alliances to exchange best practices and support the
development and uptake of AI in the public sector (2018) 543. They also established
dedicated funding streams for AI in the public sector under the European AI Alliance,
to support the scaling and spread of reusable solutions (2018) 544. The Commission
also plays an active role in the OECD AI Observatory, which was launched in 2019,
which acts as a global hub for information and knowledge-sharing on AI policies and
practices545. Finally, the upcoming AI Act will provide a harmonised approach for the
development, placement on the market and use of AI systems in the EU.
6. Establish a clear AI regulatory framework:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Adopting AI regulatory framework, including the AI Act and the AI Liability
Directive.

539
https://digital-strategy.ec.europa.eu/en/policies/data-governance-act
540
https://digital-strategy.ec.europa.eu/en/policies/data-act
541
https://ec.europa.eu/info/law/law-topic/data-protection_en
542
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence
543
https://ec.europa.eu/digital-single-market/en/european-ai-alliance
544
Ibid
545
https://oecd.ai/en/about

232
 Adopt AI Study

• Establishing a regulatory sandbox for public sector organizations to experiment

with AI-enabled solutions in controlled environments.
• Developing a set of common guidelines and standards for the development and
deployment of AI-enabled services in the public sector.
• Encouraging the participation of public sector organizations in international AI
initiatives and standards development.
• Encouraging the development of multilingual guidelines, criteria and tools for
public procurement of AI solutions in the public sector throughout Europe.
On-going activities:
The European Commission has recently proposed several policy actions to ensure the
responsible development and deployment of AI in the EU, including the adoption of a
European AI Act (2021)546. The High-level expert group on AI also established a set of
ethical guidelines for the development and deployment of AI in the EU - the Ethics
Guidelines for Trustworthy AI (2019)547. The upcoming AI Act548 proposes harmonised
rules for the development, placement on the market and use of AI systems. It also
proposes a single definition of AI.
7. Promote the integration of new AI technologies and services into
existing systems in education, health and transport:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a set of guidelines and best practices for the integration of AI in
existing systems, to support the development and deployment of AI-enabled
services in the public sector.
• Encouraging the development of multidisciplinary research and knowledge
creation amongst European universities and R&D institutions around AI for the
public sector.
• Offering dedicated capacity-building programs for public sector innovators
aiming to develop and/or adopt AI in support of the digital transformation of
public services.
• Supporting the development of reusable and interoperable AI components at all
operational levels of European public administration.
On-going activities:
The European Commission has recently proposed several policy actions to support the
integration of AI in existing systems, including the establishment of a European AI
Fund supporting the integration of AI in existing systems (2021) 549. They also
established several dedicated funding streams for AI in the public sector, including
funding for the development and deployment of AI in the fields of health, education,
and transportation550
8. Promote interoperability:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:

546
https://eufordigital.eu/discover-eu/eu-digital-single-market/
547
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
548
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence
549
https://eufordigital.eu/discover-eu/eu-digital-single-market/
550
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

233
 Adopt AI Study

• Developing a set of guidelines and best practices for interoperability and data
sharing in the public sector.
• Encouraging the development of open data and data sharing initiatives
between public sector organizations.
• Supporting the development of reusable and interoperable AI components at all
operational levels of European public administration.
• Promoting the use of a common European Data Space for public sector bodies
and their operators, drawing from the compilation of relevant AI datasets and
related Registries throughout Europe.
On-going activities:
The European Commission has recently proposed several policy actions to support
interoperability, open data and data sharing, including the European Data Strategy
(2020)551 as well as to strengthen public sector interoperability in general, through the
Interoperable Europe Act552. They also established several dedicated funding streams
for AI in the public sector to support the interoperability and data-sharing initiatives553
9. Build trust in AI solutions through transparency and accountability:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a set of guidelines and best practices for transparency and
accountability in the public sector.
• Encouraging the development of transparency and accountability initiatives
between public sector organizations.
• Promoting the adoption of ethical principles, the development of guidelines,
and the identification of mitigating measures to minimize the risks of
deployment of AI by the public sector.
• Harmonizing and complementing EU regulations to promote human-centric and
trustworthy AI-enabled public services for all citizens.
On-going activities:
The European Commission has recently proposed several policy actions to support
building trust in AI solutions through transparency and accountability, including the
Coordinated Plan on Artificial Intelligence (2021)554. They also established several
dedicated funding streams for AI in the public sector to support transparency and
accountability initiatives555. The upcoming AI Act will place predictable, proportionate
and clear obligations on providers and users of “high-risk” AI systems to ensure the
safety and respect of existing legislation protecting fundamental rights throughout the
whole AI systems’ lifecycle. For some specific AI systems, only minimum transparency
obligations are proposed, in particular when chatbots or ‘deep fakes’ are used.
10.Harmonize EU regulations to promote human-centric and trustworthy
AI-enabled public services:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:

551
https://eufordigital.eu/discover-eu/eu-digital-single-market/
552
COM(2022) 720 final
553
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
554
https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review
555
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

234
 Adopt AI Study

• Developing a set of guidelines and best practices for harmonization of EU

regulations in the public sector556.
• Encouraging the development of harmonization of EU regulations initiatives
between public sector organizations.
• Promoting the adoption of ethical principles, the development of guidelines,
and the identification of mitigating measures to minimize the risks of
deployment of AI by the public sector 557.
• Harmonizing and complementing EU regulations to promote human-centric and
trustworthy AI-enabled public services for all citizens558.
• Creating a European marketplace for GovTech solutions in support of public
sector digital transformation.
On-going activities:
The European Commission has recently proposed several policy actions to support
harmonizing EU regulations to promote human-centric and trustworthy AI-enabled
public services, including the AI Act and the Coordinated Plan on Artificial Intelligence
(2021)559. They also established several dedicated funding streams for AI in the public
sector to support the harmonization of EU regulations initiatives 560. The upcoming AI
Act561 will address most of the policy actions proposed. It aims for consistency with
existing Union legislation applicable to sectors where high-risk AI systems are already
used or likely to be used. The Commission is supporting the creation of a GovTech
incubator as well as a knowledge platform through the Digital Europe Programme.
11.Focus on long-term implementation in the use of AI in the public
sector:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a set of guidelines and best practices for long-term implementation
in the public sector.
• Encouraging the development of long-term implementation initiatives between
public sector organizations.
• Creating an EU-wide network of governance bodies for streamlined
management of AI in the public sector.
On-going activities:
The European Commission has recently proposed several policy actions to support
focusing on long-term implementation in the use of AI in the public sector, including
the Coordinated Plan on Artificial Intelligence (2021) 562. The Commission also
established several dedicated funding streams for AI in the public sector to support
long-term implementation initiatives563.

556
This aspect will be covered by the upcoming AI Act.
557
This aspect will be covered by the upcoming AI Act.
558
This aspect will be covered by the upcoming AI Act.
559
https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review
560
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
561
https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-
artificial-intelligence
562
https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review
563
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

235
 Adopt AI Study

12.Develop dedicated AI-enabled solutions based on co-creation

approaches:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a set of guidelines and best practices for co-creation approaches in
the public sector.
• Encouraging the development of co-creation initiatives between public sector
organizations and citizens.
• Leveraging civic engagement and participation in the development and
deployment of AI.
• Creating an EU-wide network of governance bodies for streamlined
management of AI in the public sector.
• Developing and applying umbrella impact assessment frameworks based on
key influencing factors to measure the use and impact of AI in the public
sector.
On-going activities:
The European Commission has recently proposed several policy actions to support the
development of dedicated AI-enabled solutions based on co-creation approaches,
including the Coordinated Plan on Artificial Intelligence (2021) 564. The Commission
also established several dedicated funding streams for AI in the public sector to
support co-creation initiatives565.
13.Create a European marketplace for GovTech solutions in support of
public sector digital transformation:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Developing a set of guidelines and best practices for GovTech solutions in the
public sector.
• Encouraging the development of GovTech initiatives between public sector
organizations.
• Promoting the development of multilingual guidelines, criteria, and tools for
public procurement of AI solutions in the public sector throughout Europe.
• Creating an EU-wide network of governance bodies for streamlined
management of AI in the public sector.
• Developing and applying umbrella impact assessment frameworks based on
key influencing factors to measure the use and impact of AI in the public
sector.
On-going activities:
The European Commission has recently proposed several policy actions to support
creating a European marketplace for GovTech solutions in support of public sector
digital transformation, including the Communication on the European Data Strategy
(2020)566. They also established several dedicated funding streams for AI in the public
sector to support the GovTech initiatives 567. The JRC has also recently published two

564
https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review
565
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
566
https://ec.europa.eu/info/strategy/data-strategy_en
567
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

236
 Adopt AI Study

reports related to GovTech; on scoping GovTech dynamics in the EU and on GovTech

practices in the EU.568
14.Strengthen the role of the EU Artificial Intelligence Observatory:
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Fostering a pan-European network of AI national observatories.
• Encouraging the sharing of best practices and experiences between different AI
observatories.
• Developing a set of guidelines and best practices for the governance of AI.
• Developing an EU-wide network of governance bodies for streamlined
management of AI in the public sector.
• Developing and applying umbrella impact assessment frameworks based on
key influencing factors to measure the use and impact of AI in the public
sector.
On-going activities:
The European Commission has created AI Watch569, an initiative of the European
Commission (EC) jointly developed by the EC Joint Research Centre (JRC) and the
Directorate General for Communications Networks, Content and Technology (DG
CONNECT). AI Watch monitors industrial, technological and research capacity, policy
initiatives in the Member States, uptake and technical developments of Artificial
Intelligence and its impact on the economy, society and public services. It provides
analyses necessary to monitor and facilitate the implementation of the European
Strategy for AI.
15.Promote the development of sustainable AI
Potential policy actions the Commission can take to increase the uptake of AI in the
public sector in Europe that are specific to this recommendation include:
• Establishing environmental standards and guidelines for AI development and
use.
• Encouraging the use of renewable energy sources.
• Promoting the development of energy-efficient AI technologies.
• Establishing a certification program for sustainable AI.
• Providing education and training on sustainable AI.
• Collaboration with international organizations.
On-going activities in this regard:
European Green Deal570, launched in December 2019, is a comprehensive plan to
make the European Union climate-neutral by 2050. The Green Deal includes measures
to promote the use of sustainable AI, such as funding for research and development of
energy-efficient AI technologies and support for the use of renewable energy sources.
AI4EU571 is a European Union-funded project aimed at promoting the development and
use of AI in Europe. The project includes a focus on sustainable AI and has established

568
See: https://www.researchgate.net/publication/359001983_Scoping_GovTech_dynamics_in_the_EU
and https://op.europa.eu/en/publication-detail/-/publication/b0ce2f17-a021-11ec-83e1-
01aa75ed71a1/language-en
569
https://ai-watch.ec.europa.eu/about_en
570
https://commission.europa.eu/strategy-and-policy/priorities-2019-2024/european-green-deal_en
571
https://www.ai4europe.eu/

237
 Adopt AI Study

a working group on "AI for Sustainable Development." The European AI Alliance572 is a

forum for stakeholders to discuss and provide input on AI policies and initiatives in
Europe. The Alliance has established a working group on "AI and Climate Change,"
which is focused on promoting the use of AI to address climate change and other
environmental challenges.

572
https://digital-strategy.ec.europa.eu/en/policies/european-ai-alliance

238
 Adopt AI Study

6 Annexes

6.1 Chapter 1 Annex

6.1.1 Methodology
The main subject for the analysis of the scoping report is public procurement, and
more specifically the following research questions:
 What are the main gaps and obstacles that hinder public procurement and the
uptake of AI?
 What are the benefits and added value that can be generated by a wider uptake
and procurement of AI systems in the EU?
 What are the main sectors/areas and AI technologies with high economic
potential (e.g. technological readiness, high growth performance) where
accelerated public procurement could bring the highest EU-added value?
Based on these research questions, the selection of the sectors will be made.
The following approach was taken to tackle the research questions above and get to
the selection of the public sector priority policy areas ready for large-scale deployment
of AI technologies:
 State of play focused on public procurement
Qualitative and quantitative mapping and analysis of national initiatives on AI
(e.g. strategies, action plans) and AI projects.
 Challenges and benefits analysis in public procurement
Analysis and mapping of identified challenges and benefits for the public
procurement of AI, complemented by findings of the private sector.
 Economic analysis
Quantitative mapping of current investments in AI projects and future national
projections. Moving beyond public procurement, an analysis of the indicators of
innovation in selected economic sectors (based on NACE classification) is
complemented by an analysis of market indicators of AI typologies.
 The selection of the public sector policy areas
Selection of the four priority areas, based on the above steps and triangulation
of findings based on the specific criteria of identified trends in governmental
initiatives, projects, and the market; maturity of AI technologies; and the
economic potential of the sectors that will create European added value for the
public procurement of AI.
The figure below visually describes how the analysis performed funnels into the final
selection of the (four) public sector policy areas to be the focus of the work for this
study.

239
 Adopt AI Study

Figure 71 Task 1 Methodology

Source: Author’s elaboration

6.1.2 Public sector areas

As generally recognised, the public sector is defined as consisting “of governments
and all publicly controlled or publicly funded agencies, enterprises, and other entities
that deliver public programs, goods, or services”573. Within this, there are multiple
functions that it undertakes by further breaking down into policy areas. These policy
areas may be covered by any of the three levels of the public sector within one
country: national, regional, and local.
According to the COFOG (Classification Of the Functions Of Government)574, which is
also used by the AI Watch, there are ten categories of governmental expenditure,
which are depicted in the figure below. For the goals of this research, this report will
rely on the COFOG classification and report on further public sector policy areas that
were identified throughout the research that falls under the depicted main categories
of COFOG below.

573
Dube, S. and Danescu, D., Supplemental Guidance, 2011. Public Sector Definition. URL: www.globaliia.org/standards-guidance
574
The COFOG (Classification Of the Functions Of Government) is an international classification which breaks down data on general
government expenditure from the System of National Accounts according to the different purposes or functions in which the funds are
used. More information: https://ec.europa.eu/eurostat/statistics-
explained/index.php?title=Glossary:Classification_of_the_functions_of_government_(COFOG)

240
 Adopt AI Study

Figure 72 Public sector policy areas

Source: Authors’ elaboration based on COFOG.

To further break down the COFOG categorisation, according to the official definitions
and research in the national political initiatives on AI and the AI projects, further sub-
categories to the COFOG classification were identified, as shown in Figure 73 below.
Figure 73 Full overview of public sector policy areas

Source: Authors’ elaboration. Note: *Sub-category added by authors based on research findings.

241
 Adopt AI Study

6.1.3 Mapping of public sector areas versus economic sectors

The overview of national government initiatives and AI projects relies on the COFOG
(and its expanded) classification of public sector policy areas. As the economic
analysis will turn towards market-related developments and statistics, that section of
the research will base itself on economic sectors, as prescribed by the European
statistical classification of economic activities, NACE codes 575.
Looking at the public sector policy areas as classified by COFOG, mapping was carried
out against the NACE classification to determine the most relevant correspondences.
Based on the mapping, the following selection of economic areas will be studied, as
illustrated in the figure below.
Figure 74 Mapping of public sector areas vis-à-vis NACE codes

Source: Authors’ elaboration.

Note: * Sub-category added by authors based on research findings.

The NACE economic activities that are not represented and thus will not be studied for
this research are Electricity, gas, steam and air conditioning supply (NACE D);

575
https://nacev2.com/en

242
 Adopt AI Study

Wholesale and retail trade, repair of motor vehicles and motorcycles (NACE G);
Accommodation and food services activities (NACE I); and Real estate activities (NACE
L). Furthermore, the defence does not form part of the Coordinated Plan on Artificial
Intelligence, for which reason it will not be considered for this study.
6.1.4 AI technologies
Throughout the literature, there are multiple ways of categorising AI as a technology
into different types or typologies.
The system of classification that is more generally used in tech parlance is the
classification of the technology into Artificial Narrow Intelligence (ANI), Artificial
General Intelligence (AGI), and Artificial Superintelligence (ASI).576 Another alternative
is to classify AI according to its ability to ‘think’ by itself: reactive machines, limited
memory machines, theory of mind, and self-aware AI.577 Furthermore, AI can also be
classified according to the main purpose behind the technology: analytic, functional,
interactive, text, and visual.578 These classifications are combined and summarised in
Figure 75.
Figure 75 AI general classification

Source: Author’s elaboration based on Medium, Forbes, and Science Soft.

Finally, the researchers at AI Watch 579 have created a detailed overview of AI

typologies and their purpose in public services, depicted and defined in the figure
below. The AI technologies (typologies) that will be used for this scoping report are
based on the AI Watch classification.

576
https://medium.com/mapping-out-2050/distinguishing-between-narrow-ai-general-ai-and-super-ai-
a4bc44172e22
577
https://www.forbes.com/sites/cognitiveworld/2019/06/19/7-types-of-artificial-
intelligence/?sh=2305ba1233ee
578
https://www.scnsoft.com/blog/artificial-intelligence-types
579
European Commission. 2020. AI Watch Artificial Intelligence in public services – Overview of the use and
impact of AI in public services in the EU. URL:
https://publications.jrc.ec.europa.eu/repository/handle/JRC120399

243
 Adopt AI Study

Figure 76 AI Watch AI typologies

Source: European Commission. 2020. AI Watch Artificial Intelligence in public services – Overview of the use and impact
of AI in public services in the EU.

The studied AI projects are mapped according to these typologies, and further detailed
or added on when relevant, to be explored in the section on the Overview of AI
projects in the public sector. These AI typologies will serve as a basis to carry out the
market analysis of these existing AI solutions’ maturity levels. Throughout the
economic analysis, indicators analysed will not always adopt the AI watch
classification, hence, the analysis will make conclusions based on the closest fit with
the AI watch classification.

6.1.5 List of initiatives

Country Title (Hyperlinked) Year

Austria 32nd Amendment to the Austria Motor Vehicles Act 2016

Austria Automated Driving Regulation 2019

Austria Code of Practice: Automated 2018

244
 Adopt AI Study

Country Title (Hyperlinked) Year

Austria Digital Roadmap Austria 2017

Austria Artificial Intelligence Mission Austria 2030 2019

Belgium Autonomous vehicles - code of practice for testing in 2016

Belgium

Belgium AI 4 Belgium 2019

Belgium Plan next tech Brussels 2017-2020

Belgium Regional Innovation Plan 2016-2020

Belgium Action Plan AI (Flanders) 2019-2024

Belgium DIGITALWALLONIA4AI 2019

Bulgaria Concept for the development of AI in Bulgaria until 2020

2030

Croatia National Plan for the Development of Artificial N/A

Intelligence (not yet published)

Cyprus National Artificial Intelligence (AI) Strategy: Key action 2019

for promoting the integration and development of AI in
Cyprus

Czechia Digital Czech Republic 2018

Czechia Digital Education Strategy to 2020 2014-2020

Czechia Innovation Strategy 2019-2030

Czechia Artificial Intelligence Strategy 2019-2035

Czechia National Strategy of Open Access to Scientific 2017-2020

Information and Data

Denmark Law on the disclosure of Data Ethics Policy 2020

Denmark National Strategy for Artificial Intelligence 2019-2022

Denmark, Declaration on AI in the Nordic-Baltic Region 2018

Estonia,
Finland, the
Faroe Islands,
Iceland, Latvia,
Lithuania,
Norway,
Sweden, and
the Ãland
Islands

Estonia Bürokratt 2020

Estonia National Artificial Intelligence Strategy 2019-2021

245
 Adopt AI Study

Country Title (Hyperlinked) Year

Finland Leading the way into the age of Artificial Intelligence 2019-2025

Finland National Regulation on automated decision-making 2020

Finland Artificial Intelligence 4.0 2020

France National Strategy on Artificial Intelligence 2018-2022

France Addressing human rights concerns arising from facial 2019

recognition technology

France Digital Republic Bill 2016

France Law on mobility orientation 2019

Germany Artificial Intelligence Strategy 2018

Germany Action Plan for Digitalisation and AI in Mobility 2018

Germany Automated Vehicles Bill in the Road Traffic Act 2017

Germany Ethical Guidelines for Self-driving Cars 2016

Greece National Strategy on AI 2020

Hungary Artificial Intelligence Strategy 2020-2030

Ireland Data Sharing and Governance Act 2019

Ireland National Artificial Intelligence Strategy 2021

Italy National Artificial Intelligence Strategy 2019

Italy Proposals and Report for an Italian Strategy on AI 2018

Latvia National AI Strategy 2020

Lithuania Law on Road Traffic Safety (self-driving cars) 2017

Lithuania Artificial Intelligence Strategy: A vision of the future 2019

Lithuania LT.AI - Fostering AI and the creation of Lithuanian 2020-2021

language technological resources for AI

Luxembourg Artificial Intelligence: A strategic vision for 2019

Luxembourg

Luxembourg Digital Luxembourg 2014

Malta National AI Strategy 2019

Malta Towards Trustworthy AI: Malta's ethical AI framework 2019

Netherlands Strategic action plan on Artificial Intelligence 2019

Netherlands Dutch Digitalisation Strategy (DDS) 2021

246
 Adopt AI Study

Country Title (Hyperlinked) Year

Netherlands Experimental law on self-driving vehicles 2019

Netherlands Non-discrimination by design 2021

Poland Artificial Intelligence Development Policy in Poland 2020-2030

Poland National AI Strategy 2020

Portugal Advanced Computing Portugal 2030 2019-2030

Portugal National Strategy for Artificial Intelligence - AI 2019-2030

Portugal 2030

Romania National AI Strategy (not yet published) N/A

Slovakia Action Plan for Digital Transformation 2019-2022

Slovenia National Programme on AI 2019-2025

Spain National AI Strategy 2020

Spain Spanish RDI Strategy in Artificial Intelligence 2019

Spain Digital Rights Charter 2021

Spain National Plan for the Advancement of Language 2016-2020

Technologies

Spain Notice 15/V - 113 (Authorisation to test automated 2015

driving systems)

Spain Notice 16 TV/89 (Automatic parking systems) 2016

Spain Plan for the Digitalisation of Public Administration 2021-2025

Spain Recovery, Transformation and Resilience Plan 2021

Spain Spanish Strategy for Science, Technology, and 2021-2027

Innovation

Spain Strategy for Connected Industry 4.0 2015

Sweden National Approach to Artificial Intelligence 2018

Sweden AI Agenda for Sweden 2019

6.1.6 List of projects

Country Title (hyperlinked) Year

Austria Mona - Public chatbot on relevant questions for 2020

companies on the subject of the corona crisis and the
economy

247
 Adopt AI Study

Country Title (hyperlinked) Year

Austria ELAK - Electronic file In development

(AI aspect)

Austria; IMAGINE 2017-2021

Germany;
France; Spain;
Turkey

Austria; Automated damage detection for different road 2018-2021

Switzerland surfaces using deep learning techniques (DACH
cooperation) - ASFALT

Belgium Camera System - Mobile phone usage on vehicles 2021

Belgium ILVO - AI4Agriculture 2020

Belgium Jobnet - Reskilling, upskilling and retraining people 2018

Belgium Verontrustingen - Enabling accurate predictions to 2014

detect day-care services which require further
inspection

Belgium Walloon - Agricultural subsidy monitoring with the use 2020

of GeoAI

Belgium AcPaas - Technical procurement documents 2017

comparison

Belgium CitizenLab - Citizen participation platform 2019

Belgium Flemish Infoline - Automatic classification of incoming 2021

phone calls

Belgium VDAB - Chatbot for job seekers 2018

Bulgaria Plovdiv - City Concierge Chatbot 2018

Croatia Andrija - Virtual assistant supporting and advising 2020-2021

suspected COVID-19 infections (over)

Croatia Hope - Chatbot informing work of the Civil Protection 2020

Staff

Cyprus WaterAnalytics - Water quality monitoring 2018-2020

Denmark Corti's AI - Emergency medical services real-time 2018

speech analysis for cardiac predictions

Denmark CHAIN - Smart Water optimising energy consumption 2020

248
 Adopt AI Study

Country Title (hyperlinked) Year

Denmark Erhvervsstyrelsen - Detection of errors in business 2019

statements

Denmark Kiri - a chatbot for citizens' enquiries 2019

Denmark Sprogteknologi (Language technology) 2020

Estonia Datel - Speed cameras for detecting speeding 2010

violations

Estonia Traffic load - Machine vision for collecting information 2018

on traffic load

Estonia OTT - Predicting employment pathways 2019

Estonia SATIKAS - Detecting the agricultural grasslands 2019

changes

Estonia X-Road AI - Detect anomalies and incidents in Data 2020

Exchange Layer

Estonia Monitoring and species identification using images and 2021

artificial intelligence

Estonia Remote monitoring of forest resources 2020

Estonia Ice map based on satellite data 2021

Estonia e-Residency - Chatbot for customer assistance 2018

Estonia Hans - A system for preparing verbatim reports 2020

Estonia Iti - Chatbot for Statistics Estonia 2020

Estonia Riigikogu - Language speech synthesiser for the 2020

Estonian Parliament

Estonia SUVE - Chatbot for questions about emergency Unknown

situations

Estonia Texta Toolkit - Identifying documents published 2019-2022

without authorisation

European Union Digitranscope 2018-2020

(JRC)

249
 Adopt AI Study

Country Title (hyperlinked) Year

Finland Case Fiva (RPA) - Support the Financial Supervisory 2018

Authority

Finland Helsinki HUH - Intensive care for a premature baby 2017

Finland AuroraAI - Programme with many AI examples 2019-2023

Finland Kamu - Chatbot about immigration processes 2017

Finland OuloBot - Virtual assistant for business 2020

Finland PatRek - Chatbot for business 2019

Finland VeroBot - Chatbot for citizens about tax management 2019

Finland Parking chatbot Unknown

Finland Oodi's book recommendation service Unknown

Finland Intelligent material management system 2019

Finland Health centre chatbot Unknown

Finland Maternity clinic chatbot NeRo Unknown

Finland; FABULOS 2018-2021

Estonia;
Greece;
Portugal;
Netherlands;
Norway

Finland; AI4Cities 2020-2022

France;
Denmark;
Estonia;
Norway;
Germany

France Datakalab - Covid-19 Mask wearing detection In some 2020-2021

French cities (over)

France French Fraud - Fraud detection value declarations 2017

France Datajust - AI in judicial decisions 2020

France SignauxFaibles - Predictive analysis of business 2020

difficulties

France Aria - Chatbot answering retired people 2018

250
 Adopt AI Study

Country Title (hyperlinked) Year

France ClaudIA - Chatbot for the invoicing portal for 2017

purchases by public authorities

France IAlim - Targeting restaurant health inspections 2019

France RenoiRH - Chatbot in HR management rules to 2020

facilitate access

France LaBonneBoite - Identification of companies with a high 2016

probability of hiring

Germany MARS 2016

Germany Germany's Federal Office for Migrants and Refugees - 2017

Refugee Language Detection

Germany Against pornography - Analysis of suspicious material 2019

Germany ARC-D - Automatic Road Condition Detection 2019-2020

Germany C-19 - Interdepartmental chatbot on the topic of 2020

Corona

Germany Child benefit - Voice assistant for parents' support 2020-2021

Germany Notification procedure - Automatic control of accidents 2019

occurring at the workplace

Germany Study certificates - Automatic recognition for applying 2018

for child benefit

Germany Govbot - Administrative search engine for citizens' 2012

enquiries

Germany AI - supported data analysis and simulation of the 2021

rescue service - AI-Rescue

Germany Weather-independent and highly automated 2021-2023

ridesharing service in Kelheim - KelRide

Germany Adaptive maps and AI-based infrastructure monitoring 2021

- AK-KII

Germany AI-based cross-type mobility optimization in non-urban 2020-2022

regions - KIMoNo

Germany Artificial intelligence in Ingolstadt's traffic system - 2020-2023

KIVI

Germany Improved quality of rail traffic through intelligent, 2018-2021

data-based damage pattern detection on rail vehicles -
QUISS

251
 Adopt AI Study

Country Title (hyperlinked) Year

Germany Mobility digital High Franconia - MobiDig 2017-2020

Germany Fleet Weather Map - FloWKar 2017-2021

Germany Holistic network approach to identify systemic 2017-2020

obstacles and coordination potential in mobility
planning - GaNEsHA

Germany Applications of V2X technology in rail traffic - Rail2X 2017-2020

Smart Services

Germany Dynamic transport optimization through anticipation 2017-2020

and real-time data analysis - TransData

Germany ally Mobility Analytics Platform - allyMAP 2017-2019

Germany Data analytics and artificial intelligence for safe and 2017-2020
reliable mobility - DAYSTREAM

Germany Intelligent pump station and lock control in the port - 2018-2021
Tide2Use

Germany Proactive video-based use of telecommunications 2016-2019

technologies in innovative highway scenarios -
Providentia

Germany Prediction of image sequences from FAS video 2017-2020

sequences - FAS-VidGen

Germany Satellite-based system for displaying, forecasting and 2018-2021

simulating air pollutants for sustainable urban and
regional development - SAUBER

Germany Intelligent information technologies for process 2018-2021

optimization and automation in inland ports -
Binntelligent

Germany Port operation optimization through predictive 2018-2021

embedded condition monitoring of the track
infrastructure - HavenZuG

Germany Interactive robotic system for emptying sea containers 2017-2020

- Iris

Germany Innovative fields of application for Augmented Reality 2018-2021

(AR) in inland and seaports - InnoPortAR

Germany Hybrid 3D as-built data acquisition and model-based 2018-2021

inspection of waterway structures for sustainable
infrastructure lifecycle management - 3D HydroMapper

Germany Autonomous machines in road construction 4.0 - 2017-2020

ROBOT road construction 4.0

252
 Adopt AI Study

Country Title (hyperlinked) Year

Germany Optimization of utilization in long-distance truck traffic 2018

through the use of artificial neural networks -
Cargonexx

Germany Techniques for Interactive Nautical AIS Data Analysis - 2018-2019

TINA

Germany The digital construction assistant of the future - 2018-2021

DeepSpaceBIM

Germany Acquisition system for high-precision and up-to-date 2018-2020

maps in buildings, data processing for accessibility and
BIM integration - indoorRobot

Germany Open Data Crowd Sensing Service for the easy fusion 2018-2021
of annotated and swarm-based mass data - OCROSS

Germany; Solve-RD 2018-2022

Netherlands;
United
Kingdom;
France; Spain;
Czechia; United
States;
Belgium; Italy;
Portugal

Greece, i-Prognosis 2016-2020

Belgium,
Germany,
Portugal,
Sweden and
United Kingdom

Greece; AI4Media 2020-2024

Belgium; Italy;
United
Kingdom;
Romania;
Netherlands;
Malta;
Switzerland;
France;
Germany;
Austria;
Ireland;
Portugal;
Denmark;
Bulgaria

Greece; iASIS 2017-2020

Germany;
Spain; United
Kingdom;
United States

Greece; BigO 2016-2021

Sweden;

253
 Adopt AI Study

Country Title (hyperlinked) Year

Netherlands;
Ireland; Spain

Hungary TÉBA - Decision on the Maternity Benefit following the 2010-2013

family support scheme of the Hungarian State
Treasury

Hungary Home Quarantine App - Supporting the use of 2020

electronic control for home quarantine

Hungary Véda - System Operated by the Police for Traffic 2018

Safety Automated Processing

Ireland VDA - Voicebot on calls from Irish taxpayers on tax 2018

clearance

Ireland Area Monitoring System (AMS) 2021

Ireland The Dublin Beat - Citizen opinion analysis 2020

Ireland SynchroniCity initiative (Dublin) 2019

Ireland; Spain; Cybele 2019-2022

France; Italy;
Germany;
Poland; Greece;
Finland;
Cyprus;
Luxembourg;
Netherlands;
Serbia; Austria;
United
Kingdom;
Belgium;
Denmark

Italy R1 - humanoid robot 2016

Italy Toscana Open Research - Allows users to access 2017

information through "queries" without having to know
the technical terms

Italy Noovle - Management of health documentation 2018

Italy Borbot - Virtual assistant for museums 2017

Italy Pierino - PIatform for the extraction and retrieval of 2016

online information

254
 Adopt AI Study

Country Title (hyperlinked) Year

Italy TALIA SF - Extract and organize knowledge coming 2017

from the analysis of a huge quantity of textual
documents

Italy, Belgium, FANDANGO 2018-2021

Greece, Ireland,
Spain

Italy; Belgium; Pantheon - Precision Farming of Hazelnut Orchards 2017-2021

Germany;
Luxembourg

Italy; Belgium; Exscalate4Cov 2020-2021

Poland;
Germany;
Spain;
Switzerland;
Sweden

Italy; Belgium; Cyberlegs++ 2017-2021

Slovenia; Italy;
Iceland

Italy; Denmark; CERTH-MUKA - Public Organizations Multi-factor 2020-2023

Norway; Misinformation Handling (ETAPAS)
Greece;
Belgium;
Sweden;
France; Austria

Italy; Spain; SIMPATICO 2016-2019

United Kingdom

Italy; Spain; Nevermind 2016-2020

United
Kingdom;
Portugal;
Sweden;
Germany

Latvia FITS ITEMS - Road safety 2018

Latvia Automatic decision-making of the Electronic 2019

Declaration System of the State Revenue Service

Latvia Hugo.lv - Virtual Assistant for governments 2018

Latvia Tom - Chatbot facilitating customer communication 2020

with the State Revenue Service

Latvia UNA - Chatbot regarding the process of enterprise 2018

registration

Latvia VARIS - Reviews all the applicants to Rural Support 2018

Service

255
 Adopt AI Study

Country Title (hyperlinked) Year

Lithuania viLTe' - Chatbot answering relevant questions 2020

regarding the COVID-19

Netherlands Amsterdam - Holiday rental home fraud 2021-2022

Netherlands ANPR - Car plate recognition 2021

Netherlands Cameras - Detection holding a phone in a car 2020

Netherlands Rotterdam – Improving the usage of rooftops 2017

Netherlands CAS - Crime anticipation system 2013 (over)

Netherlands RWS - Machine learning for road accident predictions 2017

Netherlands SyRi - Detect welfare fraud 2014-2019

(over)

Netherlands Xomnia - The Burglary Predictor Unknown

Netherlands Xomnia - Law enforcement optimization system Unknown

Netherlands Amberscript - Automated speech translation for public 2019

organisations

Netherlands SIA - Notifications of public space 2019

Netherlands Automated parking control Unknown

Netherlands; MyLeg 2018-2022

Italy; Iceland;
Australia

Netherlands; TECHNEQUALITY 2019-2021

United
Kingdom;
Sweden;
Germany; Italy;
Estonia

Norway; United SelfBACK 2016-2021

Kingdom;
France;
Denmark;
Netherlands

Poland Construction of a central system of automatic road 2011-2015

traffic supervision

Poland Home Quarantine - app on COVID-19 risk 2020

256
 Adopt AI Study

Country Title (hyperlinked) Year

Poland Poland - Fraud Detection COVID-19 support 2020

Portugal IPOscore - Predicting complications of surgery and 2018

prognosis of cancer patients

Portugal CCM-SNS - Verification of medical prescriptions 2016

Portugal Derm.AI - Tele Dermatological Screening 2019-2021

Portugal Failstopper - Study the compressed air system of the 2018-2021

vehicle

Portugal EPISA - Entity and property inference for semantic 2019

archives

Portugal Sigma - Chatbot on changing citizens' address 2019

Portugal Data2Help - Optimizing the allocation of resources for 2019

a better and faster response to medical emergencies

Romania Bucharest airport - Face recognition 2019

Romania Antonia - Automatic processing of requests for public 2018

services

Slovakia PCA - AI algorithms in audit methodology detecting 2019

potential fraud when prescribing medicines

Slovakia OverSi - Chatbot to faster bureaucracy 2018

Slovakia Taxana - Chatbot for financial administration 2019

Slovenia Semantic analyser - Document analyser for the public 2019

administration

Slovenia, RECONCYCLE 2020-2024

Germany, Italy

Spain Fuengirola Town Hall - measuring beach attendance 2020

Spain Mercé - citizen science for better urban life 2020

Spain VeriPol - Detect false police reports 2018

Spain VioGen - Forecast gender violence 2007

Spain Chatbot - Boost relations with citizens 2018

Spain Misty II - Robot for elderly people 2020

257
 Adopt AI Study

Country Title (hyperlinked) Year

Spain, France, VI-DAS 2016-2019

Ireland,
Germany,
Netherlands,
United Kingdom

Spain, Greece, FCIS - Machine learning to identify suspicious patterns 2019-2021

Poland, France, (trade, financial flows)
Romania,
Germany,
Lithuania

Spain; France; PRIMAGE 2018-2022

Germany; Italy;
United
Kingdom;
Austria;
Belgium;
Poland

Spain; Italy; BETTER 2017-2021

Germany;
United
Kingdom;
Switzerland;
Portugal

Spain; Deep Health 2019-2022

Netherlands;
Romania;
Greece; France;
Germany; Italy;
Sweden;
Switzerland

Spain; PILOTING 2020-2023

Switzerland;
France;
Czechia;
Netherlands;
Greece; Norway

Sweden Tengai - Robot in recruitment processes 2019

Sweden Scapis - Prevention of heart- and lung diseases 2021

Sweden SKOSA - Interpreting medical information 2019

Sweden Stockholm3 - Diagnosing prostate cancer 2019

Sweden AIDA - Interpreting detailed plan provisions 2021

Sweden Skatti - Answering system about population 2022

registration and income tax return

Sweden Swedish Land Registry (SLR) - Fostering efficiency 2019

when dealing with land registry requests

258
 Adopt AI Study

Country Title (hyperlinked) Year

Sweden; SmokeBOT 2015-2018

Germany;
Austria; United
Kingdom

259
 Adopt AI Study

6.2 Chapter 2 Annex

6.2.1 Definitions
A clear definition of the core concepts is paramount for this chapter to ensure
coherence and clear demarcation between the sectors. For example, the first analysis
of the “e-government” sector needs to be well-defined to avoid overlap with the other
three sectors of mobility, health and education.
Figure 77 The 4 sectors studied in the chapter

Source: COFOG

Public sector
The chapter uses Dube and Danescu´s definition of the public sector as “governments
and all publicly controlled or publicly funded agencies, enterprises and other entities
that deliver public programmes, good or services” 580. This definition caters for the
significant diversity of public bodies involved with the introduction and development of
AI in a particular sector. In terms of the sectors themselves, the COFOG (both levels I
and II) serve as a useful starting point for deciding the limits of the sectors. For
example, the first section on “e-government” uses the narrower COFOG classification
of “general public services” to define its scope more. The sectors on “health” and
“education” also have their COFOG level I classification while “transport” has a level II
classification falling within the scope of the “economic activity” level I classification.
Nevertheless, there are inevitably areas of sectoral overlap which need to be
addressed. For example, the category of “research”, is an activity carried out in each
of the 4 sectors yet this is categorized in the “education” COFOG level I. A similar
problem arises with data infrastructure and data spaces, an essential base for AI
development which could fall into multiple categories. As a general rule, the chapter´s
approach places initiatives into the sector that is explicitly stated (eg. a healthcare-
specific chatbot or a mobility data space). If the policy is more general, it falls within
the remit of “e-government” and general public services in the first section.
Procurement
Increasing public procurement of AI and accelerating uptake is important for the EU to
reach its AI policy goals. This process, whereby public authorities purchase work,
goods or services from companies 581, has long been seen as a key lever for enacting
systemic change in specific sectors and the wider economy. Accounting for €2 trillion

580
Dube, S. and Danescu, D., Supplemental Guidance, 2011. Public Sector Definition. URL:
www.globaliia.org/standards-guidance
581 https://ec.europa.eu/growth/single-market/public-procurement_en

260
 Adopt AI Study

of the EU´s annual spending (14% of total GDP) 582, public procurement is rightly
considered one of the public sector´s principal tools in achieving its policy ends.
Moreover, procurement from central, regional and local governments can have myriad
social, political and environmental benefits beyond the initial payment for goods or
services.
More specifically to this chapter, AI is both something that is procured by a
government, usually a technological solution and a tool for improving the procurement
process itself through deploying it in areas such as strategic sourcing, contract
management and predictive analytics583. This chapter focuses on both of these faces
of AI and procurement. However, the paucity of publicly available information and the
complexity of the technology means that cases of simple traditional procurement as
defined above are uncommon. Subsequently, the emphasis of the report will be on the
use of the technology, its value chain and the policy architecture behind it.

A typology of Public Sector AI Policy

A useful starting point when thinking about the public sector and AI is to ascertain
the principal forms of policy instruments used throughout the EU. Past research and
databases on the subject have categorized public policy in distinct yet
complementary ways (as seen in figure 4). These typologies, when combined into a
single list, provide a helpful lens through which to measure the progress of different
member states with AI and identify the main policy trends. These categories are
also used in the economic analyses in each section.
As of late 2022, almost every EU member state has published a National AI Strategy
and, in the process, signalled significant intent when it comes to AI adoption.
Indeed, amongst the literature, there is a relative consensus that, with EU-level
backing, Europe as a whole is moving into a new stage of macro-AI policy584 where
the public sector is an increasingly central actor.
Sources such as the OECD´s AI Observatory and European Commission AI
Watch´s National Strategy Reports highlight different policy types in use such as
regulation, governance financial support, infrastructure etc. Alternatively, others
such as AI Watch´s report on the public sector by Misuraca and Van Noordt585
and its 3rd Peer learning workshop586 describe the public sector´s relationship
with AI as more of a progression or trajectory. The latter starts with regulating and
facilitating AI Innovation followed by using the technology and finally developing it
themselves. Of course, good regulation is a constant process that needs to be
updated to remain relevant, especially with such a fast-moving technology. Leading
countries in AI will have a repertoire containing each type of policy however using
and developing the technology should be seen as the normative policy goal for the
public sector.

582
https://ec.europa.eu/growth/single-market/public-procurement_en
583
https://sievo.com/resources/ai-in-procurement#:~:text=the%20full%20experience.-
,Use%20of%20AI%20in%20procurement,contract%20management%20and%20strategic%20sourcing.
584
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services, EUR 30255 EN,
Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-19540-5 (online),
doi:10.2760/039619 (online), JRC120399
585
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services, EUR 30255 EN,
Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-19540-5 (online),
doi:10.2760/039619 (online), JRC120399
586 van Noordt, C., Alishani, A., Tangi, L., Gattwinkel, D. and Pignatelli, F., AI Watch. Artificial Intelligence
for the public sector. Report of the “3rd Peer Learning Workshop on the use and impact of AI in public
services”, 24 June 2021, Publications Office of the European Union, Luxembourg, 2021, ISBN 978-92-76-
41530-5, doi:10.2760/162795, JRC126501.

261
 Adopt AI Study

In addition to a member state in its entirety, this typology can be also used to
assess the role of the public sector as it relates to individual policies. This report A)
identifies the role of the public entity within each AI policy during the initial policy
mapping and B) Gives countries a category based on the sum of these policies
which is used in the economic analysis. The report relies on the following categories
which are a blend of the typologies in figure 4. Below is the list of five modes of
government action as they relate to AI policy.
Planning
The likely first stage in a government´s AI journey will be as a planner for
subsequent other actions. In terms of policies, these could be national strategies,
agendas, plans587 or white papers. Finally, in the economic models, this category is
used for countries which have not yet passed any concrete AI policy in that sector.
Regulating
If the policy is a form of guidance and regulation then the government´s role is
deemed to be a regulator. These actions could include regulatory oversight and
ethical advice bodies or standards and certification 588.
Facilitating
The facilitator category denotes a policy in the area of infrastructure or
networking. Also, as a form of AI Enabler (OECD), actions such as building and
maintaining dataspaces and conducting public awareness campaigns on AI fall into
this category.
Funding
A distinct and more direct form of public investment is the category of a funder.
Any policy in the OECD´s “financial support” or AI Watch´s “from the lab to market”
sections fall into this section. In terms of specific instruments, this includes “grants
for public research”, “business R&D and innovation” and “procurement programmes
for R&D and innovation” mentioned above.
Using
The final category, User, is used in cases where public bodies are using AI
technology themselves. These may have been developed themselves or procured
from the private sector.

587
https://oecd.ai/en/dashboards
588
https://oecd.ai/en/dashboards

262
 Adopt AI Study

Table 14 Typologies of AI public sector activity from the literature

Source AI Watch: AI AI Watch´s 3rd OECD´s AI Policy AI-Watch

in public Peer learning Observatory National
services workshop Strategies
(Misuraca Report
and Van
Noordt)

Categories Regulator Data-focused Guidance and Regulation

of Public approach Regulation
Facilitator Infrastructure
Sector
Private sector Governance
Policy User Networking
leadership
Financial support
Developer Human Capital
AI4GOV front-
AI Enablers and
runner From the lab to
other support
the market

Source: Authors’ elaboration.

6.2.2 Methodology
This chapter deploys a mixed-methods research design to fully investigate the four
sectors in question. With the synergistic value of using multiple methods, each
subsection will address a different aspect of the research question(s) to produce a
holistic and triangulated analysis of AI adoption and uptake.
The approach draws an AI Watch report by Misuraca and Van Noordt and their
proposal (see below) for studying public sector innovation. With the qualitative
components of the chapter, the study identifies the context, internal and external
drivers and barriers related to the implementation of AI into public administrations in
each sector. Through the policy mapping and challenges and solutions sections,
inferences regarding the propensity of the public sector to adopt AI technology are
possible. Furthermore, the value chain analysis in conjunction with the SWOT and
PESTEL analyses approaches the implementation of AI itself. Likewise, the
quantitative aspect, where possible, performs the role of measuring outcomes
against the expected theory. The conclusion will summarize the main lessons from the
research and forward a series of suggestions for other administrations looking to enact
policy in AI. In its totality, the chapter aims to illustrate the dynamics underpinning
the introduction of AI into public administrations.
Figure 78 Summary of the structural factors affecting the adoption of AI in public services taken from an AI Watch study

Source: Misuraca and Van Noordt

263
 Adopt AI Study

Policy Mapping
To establish the state of play for each sector and provide a solid base for further
investigation, each section will start with a mapping of the policy landscape and build
on the existing inventory carried out in previous research. Using desk research, the
policy landscape indicates the broader trends in each sector such as areas of
geographic strength or the popularity of a certain type of policy or AI being used (e.g.
a chatbot in “eGovernment”). Furthermore, this catalogue of policies will dictate the
category (e.g. regulator, facilitator, user) into which member states are placed for the
economic analysis (explained in more detail below).
High-performance computing has not been included in this chapter which would
expand an already wide scope even further. The full list of policies can be found as an
annex and is broken down into each sector. These tables also include other categories
of interest including the implementing body, the type of AI technology used (where
relevant), the level of government (national/subnational etc), the source and crucially
government´s role in the policy. As displayed in figure 3, this draws on earlier
typologies but, rather than the country writ large, the governmental role is assessed
for each policy. This includes the categories of the planner, regulator, facilitator,
funder and user as outlined above.
Value Chain Analysis
The value chain analysis was carried out by leveraging publicly available information.
The methodology started by looking for existing representations in the literature.
When sufficient, these models were recovered and are used in the present study (see
e-government sector). However, for both the mobility and the health sector, different
models were brought together and merged to allow a joint analysis of the R&D value
chain, the physical value chain and the digital value chain. For the education sector,
existing models were considered unsatisfactory for this study and a new model was
developed from scratch.
These models were developed to identify the key areas where public procurement and
interventions from public authorities could support the deployment and uptake of
Artificial Intelligence. Each link in the value chain was analysed separately. To make
the study concise and focused on the key elements, these links are not always
analysed separately, but jointly. This is especially interesting to identify synergies.
Furthermore, when relevant, the impact of the COVID-19 outbreak on the redefinition
of the value chain was also considered and analysed.
SWOT and PESTEL analyses
For the identification of the main drivers and barriers in the identified four sectors, two
main tools are used. The PESTEL analysis enabled the classification of the divers and
drawbacks identified at six different levels: Political, Economic, Social, Technological,
Environmental and Legal. The SWOT (Strengths, Weaknesses, Opportunities and
Threats) gave an accurate diagnosis of the current situation and the surrounding
environment
To construct these two analyses, the main challenges and solutions identified and
presented in the previous section were gathered and reorganized. Based on this
information, the different barriers and drivers were identified and categorised in the
corresponding level of the PESTEL. This categorisation enabled, at a later stage, the
synthesis action of similar ideas that have connections among them. This integration
work led to the definition of the strengths, weaknesses, opportunities and threats of
the uptake of AI technologies in public procurement in the identified sectors.
Furthermore, the weaknesses and threats identified in the SWOT analysis allow a
deeper understanding of the barriers detected during the study.

264
 Adopt AI Study

Case studies
In terms of the selection strategy, the case studies were selected on several criteria.
The first, central to the idea of best practices and key lessons to be followed by other
government actors, was innovation and the verifiable success of the case. Secondly, to
promote the adoption of AI by public authorities in mind, the idea of replicability was
also key. Cases which required levels of resources likely unavailable to most member
states such as a very large budget or a pre-existing level of development in a certain
sector were not considered. The areas examined in the case study were the basic
timeline of the policy/project/initiative, the type and nature of the
procurement/development such as the funding dynamics and a series of key lessons
that can serve as a guide for other public entities wanting to involve themselves in AI.
Economic analysis
The final part of each section is a brief economic analysis that complements the
preceding qualitative components. The two main aims of this subsection are to test
and quantify some of the theories and findings of the prior analysis and wider
literature as well as compare the relationships between different types of government
intervention with factors of interest. While the models do not make any sweeping
causal claims, the aim is generally to support the claim with quantitative evidence that
investment and development of AI by governments is beneficial.
The quantitative component of each sector (except mobility) will use linear
regression to compare the performance of different modes of government policy on a
selection of sector-specific indicators drawn from the literature and the prior research
in each section.
Each of the models uses the same categorical independent variable based on the
types of government action highlighted in the typology of the public policy above.
Each member state will be given a category according to the relationship of its public
sector to AI based on the policy mapping. This draws on previous literature (see figure
24 above), including research conducted by AI Watch, in clustering countries
according to their approach to AI. This section uses the five modes of policy action
explained above: user, regulator, facilitator, funder and user. if a member state is
already using AI technology at the national level, it is categorized as a “user” in the
model much like a country with no AI policies is categorized as a “planner”. Each
member state´s designation in each section is illustrated in a multi-coloured map at
the start of each subsection. As highlighted above, the journey of a country is not a
linear path from the planner to the regulator to the user. Nevertheless, the evidence
from the mapping in these sectors indicates that countries which have reached a
certain threshold (ie. AI use) will only continue on that path.
The dependent variables differ in each sector. These variables are taken from the
literature and other findings in the report and are drawn from “relevant social and
economic indicators being representative of the context in which AI is embedded and
integrating into the analysis of other direct and indirect factors that are also
influencing impact”589. Generally, these indicators quantify some of the theoretical
benefits of AI in the public sector such as improving internal efficiency, policy delivery,
public services and citizen-government interaction590. While the lack of indicators on
the public sector and AI are frustrating, to the extent that the mobility section does

589
Misuraca, G. and Van Noordt, C., AI Watch - Artificial Intelligence in public services, EUR 30255
EN, Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-19540-5
(online), doi:10.2760/039619 (online), JRC120399.
590
Manzoni, M., Medaglia, R., Tangi, L., Van Noordt, C., Vaccari, L. and Gattwinkel, D., AI
WatchRoad to the adoption of Artificial Intelligence by the Public Sector: A Handbook for
Policymakers, Public Administrations and Relevant Stakeholders, EUR 31054 EN, Publications
Office of the European Union, Luxembourg, 2022, ISBN 978-92-76-52131-0, doi:10.2760/693531,
JRC129100.

265
 Adopt AI Study

not contain any linear regression models, the economic analyses still test and quantify
some of the expected outcomes of the theory behind AI public sector implementation.
Some examples of this hypothesis testing in the report include: whether countries with
adult education on AI have a higher concentration of AI talent or whether is there any
evidence that member states using AI will have a more accessible government in
terms of the number of digital users. Although the lack of indicators and recency of AI
investment mean that proving causality between public sector activity and these
trends is challenging, the correlations are still insightful for the reasons laid out above.
6.2.3 Data sources and collection
In terms of data collection of policies, several sources were used to carry out the
mapping including the OECD AI observatory´s policy inventory, the national AI
strategy reports from AI watch, and the AI-X initiative inventory. Another source that
was used heavily was the JRC´s database on use cases of AI in the public sector 591.
This used the COFOG headings according to the sector in question and only included
policies that had been “implemented”. Nevertheless, given the reliability of these data
sources and the relative paucity of AI policies, this mapping should encompass most of
the policies on AI that have publicly available information. All of the policies used in
the mapping, and tangentially in the economic analysis, can be found in the annex.
The drivers and barriers to the uptake of AI technologies in public procurement in the
identified four sectors build on the findings from the value chain analysis.
Interviews were also carried out mostly concerning the case studies. Interviewees
were selected from representatives of organizations that had been part of the chosen
initiatives. This included both public and private sector actors such as representatives
from the Mobility Data Space and Reaktor, a Finnish technology consultancy which
was involved in the Elements of AI MOOC, the case study in the education section.
While the principal focus was on the intricacies of the cases, interviewers were also
asked about the wider sector with their responses contributing to other subsections in
the chapter.

591
https://ipsoeu.github.io/ips-explorer/case/

266
 Adopt AI Study

6.2.4 Policy overview

Member State Policy Name Implementing Government Policy Type Source Level of
body Role Government
Austria MONA UnternehmensseService User Chatbots, AI-X National
Portal Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Austrian Council for Federal Ministry for Regulator Regulation OECD National
Robotics and AI Transport, Innovation
and Technology
Belgium Innovative public Flemish Department of Funder From the lab to National Regional
procurement programme Economy, Science and market strategy report
Innovation
Datastore.Brussels Brussels Regional Facilitator Infrastructure National Regional
Informatics Centre strategy report
Opendata.Brussels City of Brussels Facilitator Infrastructure National Regional
strategy report
ACPaas City of Antwerp User Platform AI-X Regional
CitizenLab Traffic Institute Vias User Platform AI-X National
Flemish Infoline Information Flanders User Audio processing AI-X Regional
VDAB – Chatbot for job Vlaamse Dienst voor User Chatbots, AI-X Regional
seekers Arbeidsbemiddeling en Intelligent Digital
Beroepsopleiding Assistants, Virtual
Agents and
Recommendation
Systems
Veronstrustingen Kind en Gezin User Predictive analytics AI-X Regional
Reducing night noise Leuven User Audio Processing, AI Watch Regional
through nudging Machine Learning Landscaping
Report
eTranslation: Machine European Parliament ** User Natural language JRC National
translation for public processing
administrations
Bulgaria Plovdiv – City Concierge UMNI User Chatbots, AI-X Regional
Chatbot Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Cyprus National Open Data Portal Data.gov.cy Facilitator Infrastructure National National
strategy report

267
 Adopt AI Study

Denmark National Centre for Public Danish Government Funder From the lab to National National
Sector Innovation Market strategy report
Investment Fund Danish Government Funder From the lab to National National
Market strategy report
Performance contracts with Danish Government Funder From the lab to National National
7 GTS Institutes Market strategy report
Kiri Frederiksberg User Chatbots, AI-X Regional
Municipality Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Intelligent Control Platform Danish Business User Machine Learning, AI Watch National
Authority Automated Landscaping
Reasoning Report
Taxation and the Danish Tax Authority User Automated JRC National
automatization of property reasoning
evaluation
Estonia Open data portal Ministry of Economic Facilitator Infrastructure National National
Affairs and strategy report
Communications
Estonian catalogue of Information System Facilitator Infrastructure National National
public sector information Authority strategy report
systems
e-Residency – Chatbot for e-Residency organization User Chatbots, AI-X National
customer assistance Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
EE Parliament – a system Estonian Parliament User Audio processing AI-X National
for preparing verbatim
reports
Iti - Chatbot for Statistics Estonian Statistical Office User Chatbots, AI-X National
Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Riigikogu – language Estonian Parliament User Chatbots, AI-X National
speech synthesizer for the Intelligent Digital
Estonian Parliament Assistants, Virtual
Agents and

268
 Adopt AI Study

Recommendation
Systems
Texta Toolkit Ministry of Education and User Natural language AI-X National
Research processing
X-Road AI State Information User Predictive Analytics, AI-X National
Systems Board RIA Simulation and Data
Visualisation
Automatic translation in Commercial Register User Natural language JRC National
the information system of processing
the Commercial Register
KRATID - detecting traffic Information system User Computer vision JRC National
anomalies and incidents of authority
the Estonian data
exchange layer (X-Road)
Finland Aurora AI Ministry of Finance Regulator Regulation OECD National
AI Registers City of Helsinki Regulator Regulation National Regional/International
Strategy Report
Case Fiva – Support the The Financial User Natural language AI-X National
financial supervisory Supervisory Authority processing
authority
Kamu Chatbot about Finnish Immigration User Chatbots, AI-X National
immigration processes Authority Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Oulobot – virtual assistant City of Oulo User Chatbots, AI-X Regional
for business Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Automation of subtitling Finnish Tax User Audio Processing JRC National
videos and audio - Administration
Improving accessibility of
public websites
Explore state spending Hansel Oy User Automated JRC National
reasoning
France AI sandbox programme of National Data Protection Facilitator AI Enablers and OECD National
the National Data Authority other support
Protection Authority
Labor AI – Centre of Foreign Office, Ministry Facilitator Governance OECD International
expertise of the global of Economy and Finance,
partnership on AI INRIA, Ministry of Labour

269
 Adopt AI Study

CASD secure Data Hub National Institute of Facilitator Infrastructure National National
Statistics strategy report
Aria – Chatbot answering National Old Age User Chatbots, AI-X National
retired people Insurance Fund (Cnav) Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
ClaudIA – a chatbot for the State Financial IT User Chatbots, AI-X National
invoicing portal for Agency, France Intelligent Digital
purchases by public Assistants, Virtual
authorities Agents and
Recommendation
Systems
RenoiRH – Chabot in HR Interministerial Centre User Chatbots, AI-X National
Management rules to for IT services relating to Intelligent Digital
facilitate access Human Resources Assistants, Virtual
Agents and
Recommendation
Systems
Noa chatbot Prefecture de l'Ile de User Chatbots, JRC Regional
France Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Germany Reality lab for AI in Civil German Federal Agency Funder From the lab to National National
Protection for Technical Relief Market strategy report
Universal AI platform for Federal Ministry of Facilitator Governance OECD National
the BFV Finance (BMF)
Plattform Lernende System Federal Ministry of Facilitator Networking National National
Education and Research strategy report
Observatory for AI in Work Federal Ministry of Regulator Regulation OECD National
and Society Labour and Social Affairs
Govbot – administrative Ministry of Home Affairs User Chatbots, AI-X Regional
search engine for citizens' and Local NRW Intelligent Digital
enquiries Assistants, Virtual
Agents and
Recommendation
Systems
Project ML-SAST Federal Office for User Machine Learning JRC National
Information Security
(BSI)

270
 Adopt AI Study

Enhancing metadata for Federal Office for User Natural language JRC National
textual information - Security in Information processing
improving the quality of Technology (Bundesamt
search and automatic für Sicherheit in der
reasoning Informationstechnik -
BSI)
ChatBot Botty Bonn City of Bonn User Natural language JRC Regional
processing
Bobbi - Berlin City of Berlin User Natural language JRC Regional
administration processing
Frag-den-Michel! Free and Hanseatic City User Natural language JRC Regional
Innovativer Online- of Hamburg processing
Bürgerservice in Betrieb
Crisis Prevention: German Federal Foreign User Planning and JRC National
PREVIEW Office scheduling
Greece CERTH-MUKA Municipality of Katerini User Natural language AI-X Regional
processing
Hungary TEBA – Decision on Hungarian State User Expert and Rule- AI-X National
maternity benefit Treasury based Systems,
Algorithmic
Decision Making
Automated administrative Ministry for Innovation Facilitator AI enablers and OECD National
procedures in Hungarian and Technology infrastructure
Ireland The Dublin Beat – Citizen Municipality of Dublin User Natural language AI-X Regional
opinion analysis processing
VDA – voicebot on calls Irish Revenue User Chatbots, AI-X National
from Irish taxpayers on tax Commissioners Intelligent Digital
clearance Assistants, Virtual
Agents and
Recommendation
Systems
Virtual Digital Assistant Office of the Revenue User Chatbots, JRC National
with the Revenue Commissioners Intelligent Digital
Commissioners of Ireland Assistants, Virtual
Agents and
Recommendation
Systems
Italy Toscana Open Research Regional Conference for User Chatbots, AI-X Regional
Research and Innovation Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems

271
 Adopt AI Study

Evasometro Anonimizzato Italian Tax Authority User Planning and JRC National
- Big data analysis of scheduling
taxpayer behaviour
SegnalaCI - Transparent Genova Municipality User Planning and JRC Regional
management of citizen scheduling
reports about City status
AI Autonomation - CSI Piemonte User Robotics and JRC Regional
Automatic classification automation
tools and integration with
Robot Process Automation
(RPA)
Citizen Conversational AI CSI Piemonte User Natural language JRC Regional
Platform processing
ConsolONE: proactive Smartnet Srl* User Automated JRC National
monitoring and reasoning
management system for
real-time control of all
company resources,
IBM Watson Content Hub AGID (Agenzia per Italia User Automated JRC National
Digitale) reasoning
Amazon Lex: service for AGID (Agenzia per Italia User Natural language JRC National
creating communication Digitale) processing
interfaces via voice and
text for any type of
application
Amazon Polly: Text-to- AGID (Agenzia per Italia User Automated JRC National
Speech service Digitale) reasoning
LombardiaInforma: Regione Lombardia User Automated JRC Regional
proactive Citizen reasoning
Communication
YUCCA-Smart Data Regione Piemonte User Machine learning JRC Regional
Platform
ANAC, Machine learning ANAC User Machine learning JRC National
tool
Caterina, the first (virtual) Siena Municipality User Natural language JRC Regional
municipal employee for processing
identity cards and changes
of residence
Chatbot Comune di Collegno Municipality User Natural language JRC Regional
Collegno (TO) processing
Chatbot Covibot Provincia Autonomous Province of User Natural language JRC Regional
Trento (TN) Trento - Health Agency processing
Latvia Automatic decision-making State Revenue Service User Chatbots, AI-X National
of the electronic Intelligent Digital

272
 Adopt AI Study

declaration system of the Assistants, Virtual

state revenue service Agents and
Recommendation
Systems
Hugo - Virtual Assistant for Culture Information User Chatbots, AI-X National
governments Systems Centre Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Toms – chatbot facilitating State Revenue Service User Chatbots, AI-X National
customer communication Intelligent Digital
with the State Revenue Assistants, Virtual
Service Agents and
Recommendation
Systems
Luxembourg AI: a strategic vision for Ministry of State Facilitator Networking OECD National
Luxembourg
Unlocking digitized Bibliothèque nationale de User Computer vision JRC National
documents - using OCR, Luxembourg
extracting entities and
organising image content
Netherlands AI Register City of Amsterdam Regulator Regulation National (Inter)National
strategy report
AI Coalition of the Ministry of Economic Funder From the lab to National National
Netherlands Affairs and Climate Policy market strategy report

Non-discrimination by Ministry of the Interior Regulator Regulation OECD National

design and Kingdom Relations
Amberscript Multiple local User Audio processing AI-X Multiple Regional
governments
SIA – Notifications of public Municipality of Dordrecht User Chatbots, AI-X National
space Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Calculemus FLINT - User Natural language JRC National
Ensuring understandability processing
of normative law texts
Digital assistant for Ministry of Social Affairs User Natural language JRC National
answering Parliamentary and Employment processing
questions

273
 Adopt AI Study

Amsterdam City Archive - Ancient B.V., Picturae, User* Natural language JRC Regional
Handwritten Text Sioux Technologies, processing
Recognition for enabling Islands of Meaning,
searches on old Amsterdam City Archive.
handwritten texts
Reporting issues in public Research, Information & User Planning and JRC National
space Statistics (OIS) scheduling
Poland GovTech Government of Poland Facilitator Networking National National
strategy report
Fraud Detection COVID-19 National Tax User Natural language AI-X National
support Administration processing
State Clearance Chambre – State Clearance User Predictive Analytics, AI-X National
Financial Risk Indicator Chambre Simulation and Data
Visualisation
Virtual Clerk - Assistance Chorzow User Natural language JRC Regional
with the search for processing
information
Intelligent Management Plonsk User Machine Learning JRC Regional
Systems
Resident's Virtual Advisor Wroclaw User Natural language JRC Regional
processing
Kąty Wrocławskie Virtual Kąty Wrocławskie User Natural language JRC Regional
clerk processing
iVoting - a cyberdemocracy Jawor User Natural language JRC Regional
tool processing
Portugal Forum AI AI Portugal 2030 Facilitator Networking National National
strategy report
iSIMPLEX Administrative Planner National Strategies, OECD National
Modernization Agency agendas and plans
LabX Centre for Innovation Funder From the lab to National National
market strategy report
InnoLabs Centre for Innovation Funder From the lab to National International
market strategy report
Sigma – Chatbot on Administrative User Chatbots, AI-X National
changing citizens' address Modernization Agency Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
ASAE Intelligent AMA (Agência para a User Planning and JRC National
Management and Modernização scheduling
Operational Control Center Administrativa)
- CIGESCOP

274
 Adopt AI Study

Romania Antonia – Automatic Municipality of Cluj- User Chatbots, AI-X Regional

processing of requests for Napoca Intelligent Digital
public services Assistants, Virtual
Agents and
Recommendation
Systems
Slovakia OverSi – Chatbot to faster Office of the Deputy User Chatbots, AI-X National
bureaucracy Prime Minister of the Intelligent Digital
Slovak Republic Assistants, Virtual
Agents and
Recommendation
Systems
Taxana – Chatbot for Financial Administration User Chatbots, AI-X National
financial administration Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Slovenia Semantic analyser – Ministry of Public User Natural language AI-X National
document analyser for the Administration processing
public administration
Spain GovTechLab Comunidad de Madrid Funder From the lab to National National
market strategy report
060 Service – Answer FAQ General Secretariat of User Chatbots, AI-X National
from citizens and Digital Administration Intelligent Digital
bureaucratic terms in Assistants, Virtual
informal language Agents and
Recommendation
Systems
Chatbot – Boost relations Municipality of Sant Just User Chatbots, AI-X Regional
with citizens Desvern Intelligent Digital
Assistants, Virtual
Agents and
Recommendation
Systems
Mercé – citizen science for City of Barcelona User AI-empowered AI-X Regional
a better life Knowledge
Management
Support system for City of Barcelona User Natural language JRC Regional
classification of citizen processing
inquiries or complaints
IBM Watson: Automated Spanish Tax Authority User Natural language JRC National
assistance in the tax filing processing

275
 Adopt AI Study

Brain4it: monitoring the City Council of Sant Feliu User Planning and JRC Regional
SmartCity de Llobregat scheduling
Sentilo BCN - Plataforma City of Barcelona User Natural language JRC Regional
de Sensors i Actuadors de processing
Barcelona
Sweden AIDA – Interpreting Orebro Municipality User Predictive Analytics, AI-X Regional
detailed plan provisions Simulation and Data
Visualisation
Skatti – Answering system Swedish Tax Agency User Chatbots, AI-X National
about population Intelligent Digital
registration and income tax Assistants, Virtual
return Agents and
Recommendation
Systems
Swedish Land Registry Swedish Land Registry User Natural Language JRC National
(SLR) - Fostering efficiency processing
when dealing with land
registry requests
Automated processes Uddevalla Municipality User Robotics and JRC National
Uddevalla - AI to internal Automation
processes

Member State Policy Name Implementing body Government Policy Type Source Level of Government
Role
MOBILITY
Austria 32nd Amendment to the National Government of Austria Regulator Regulation OECD National
Austrian motor vehicle
act
Automated driving Federal Ministry for Traffic, Regulator Regulation OECD National
regulation Innovation and Technology
Code of Practice: Federal Ministry for Transport, Regulator Regulation OECD National
Automated Innovation and Technology
Belgium Autonomous vehicles – Federal Public Service Mobility Regulator Regulation OECD National
code of practice for and Transport
testing
Innoviris Innoviris Funder Funding Support OECD Regional
Digital Wallonia 4AI Agency of Development Funder Funding Support OECD Regional
FARI Institute Brussels Capital Region Funder Funding Support National Regional
Strategy
Report
Czechia Catalogue of Ministry of Transport Regulator Regulation OECD National
autonomous vehicle

276
 Adopt AI Study

testing areas on public

roads
Denmark Danish Road Directorate Ministry of Transportation Regulator Regulation OECD National
Estonia Traffic load – Machine Public Transport and Traffic User Computer Vision AI-X Regional
vision for collecting Management Transport Board
information on traffic
load
A unique artificial Ulemiste City User Computer Vision JRC Regional
intelligence system now
directs Ülemiste City
drivers to available
parking spots
Finland Road Traffic Act Ministry of Transport and Regulator Regulation OECD National
Communications
AI Business Programme Business Finland Funder From the lab to the National National
market Strategy
Report
Parking chatbot Urban Environment Division User Chatbot JRC Regional
France AI Challenges Directorate General for Funder Funding Support OECD National
Enterprises
Law on mobility Ministry of Ecological Transition Regulator Regulation OECD National
orientation
Rouen Moveo l’Assemblée Générale du Pôle Facilitator Networking National Regional
de compétitivité Strategy
Report
Germany AI Automatic Road Federal Ministry of Transport User Computer vision AI-X National
Condition Detection and Digital Infrastructure
An act amending the Federal Ministry of Transport Regulator Regulation OECD National
road traffic act and the and Digital Infrastructure
compulsory insurance
act
mFund Federal Ministry of Transport Funder Financial Support OECD National
and Digital Infrastructure
Mobility Data Federal Ministry of Transport Facilitator Infrastructure National National
Marketplace/data space and Digital Infrastructure Strategy
mobility Germany Report
Automated and Federal Ministry of Transport Planner National Strategies, OECD National
connected driving and Digital Infrastructure agendas and plans
Real-world Test Field for Federal Ministry of Transport Funder Funding Support OECD National
Digital Mobility and Digital Infrastructure
Ethical guidelines for Federal Ministry of Transport Regulator Regulation OECD National
self-driving vehicles and Digital Infrastructure
Greece ANPR Greece Hellenic Police User Computer visión JRC National

277
 Adopt AI Study

Hungary Robocop – System National Tax and Customs User Computer vision AI-X National
operated by the Police Administration
for Traffic Safety
Automated Processing
Ireland SFI CRT in AI Science Foundation Ireland Funder From the lab to National National
market Strategy
Report
Italy Smart Planner – FBK Research Centre User Computer vision AI-X National
Navigation planning
assistant
Mobilità Integrata Bologna, Trento, Rovereto User Machine Learning JRC Regional
Trentino Trasporti Municipalities
(MITT)
Latvia FITS ITEMS – Road State Public Traffic Safety User Computer vision AI-X National
Safety Directorate
VARIS – Reviews all the Rural Support Service User Searching AI-X National
applicants to Rural
Support Service
Netherlands Experimental law on Ministry of Infrastructure and Regulator Regulation OECD National
self-driving vehicles Water Management
Bridge sensors Province of South Holland User Computer vision JRC Regional
The Smart bridge Province of South Holland User Planning and JRC Regional
scheduling
Automated parking Amsterdam Parking Services User Planning and JRC Regional
control scheduling
Poland Gliwice & Poznan Gliwice & Poznan User Automated JRC Regional
Integrated Public reasoning
Transport Management
System - ITS
AI-based timetables Poznan User Planning and JRC Regional
Scheduling
Wroclaw Intelligent Wroclaw User Planning and JRC Regional
parking system Scheduling
Integrated Public Lublin User Automated JRC Regional
Transport Management reasoning
System - ITS
Portugal Failstopper – Study the Metro de Porto User User AI-X National
compressed air systems
of the vehicle
Slovakia Smart Mobility Lab Ministry of Transport Funder From the lab to OECD National
market
Spain Notice on automatic National Department of Traffic Regulator Regulation OECD National
parking systems

278
 Adopt AI Study

Notice of the National Department of Traffic Regulator Regulation OECD National

authorization to test
automated systems

Member State Policy Name Implementing body Government Role Policy Type Source Level of
Government
HEALTH
Belgium Innoviris Brussels City Region Funder Funding Support OECD Regional
Flanders Care Department of Funder From the lab to market National Regional
Welfare, Public Health Strategy
and Family Report
AI Research Programme Flemish Department of Funder From the lab to market National Regional
Economy, Science and Strategy
Innovation Report
FARI Institute Brussels Region Funder From the lab to market National Regional
Government Strategy
Report
Denmark Innovation Fund Denmark Danish Government Funder Funding Support OECD National
Finland Hyteairo Programme Finnish Institute for Funder From the lab to market National National
Health and Welfare Strategy
Report
AuroraAI Ministry of Finance User Chatbot AI-X National
France Datakalab – COVID-19 Public Transport User Computer Vision AI-X Regional
mask-wearing detection in
some French cities
Breakthrough Innovation The National Portal for Funder Funding Support OECD National
Challenge on AI in Health: eHealth Innovation
3IA Institutes Directorate General for Funder From the lab to the National National
Enterprises market Strategy
Report
Health Data Hub Ministry of Solidarities Facilitator AI Enablers OECD National
and Health
Inria Ministry of National Facilitator Networking OECD International
Education, Advanced
Instruction and
Research
Germany Funding priority: digital Federal Ministry of Funder Funding Support OECD National
innovation for the Health
improvement of patient-
centred care in healthcare

279
 Adopt AI Study

Research on AI Federal Ministry of Funder Funding Support OECD National

technologies in Food and Agriculture
Agriculture, Health
nutrition, food chain and
rural areas
Research data centre Federal Institute for Funder Funding Support OECD National
Drugs and Medical
Devices
C-19 – Interdepartmental ITZBund User Chatbots, Intelligent AI-X National
chatbot on the topic Digital Assistants, Virtual
Corona Agents and
Recommendation
Systems
Hungary Home Quarantine App – Hungarian User User AI-X National
Supporting the use of Government
electronic control for
home quarantine
Ireland Rapid Response R+I Irish Research Council Funder From the lab to market* National National
Programme Strategy
Report
SFI Discover Science Foundation Funder From the lab to market* National National
Ireland Strategy
Report
Italy Italian Institute for AI Ministry for Economic Funder Funding Support OECD National
Development
Lithuania National Research Research Council of Funder Funding Support OECD National
Programme for Healthy Lithuania
Aging
Netherlands Knowledge and Innovation Ministry of Economic User User OECD National
Covenant – Health and Affairs and Climate
Care Policy
National Growth Fund – Economic Affairs and Funder Funding Support National National
regenerative medicine, Climate Policy Strategy
health data infrastructure Report
Portugal CCM-SNS – Verification of Public National Health User Predictive Analytics, AI-X National
medical prescriptions Service Simulation and Data
Visualisation
Data2Help – Optimizing National Institute of User Predictive Analytics, AI-X National
the allocation of resources Medical Emergency Simulation and Data
for a better and faster Visualisation
response to medical
emergencies

280
 Adopt AI Study

Derm.AI – Tele National Health User Predictive Analytics, AI-X National

Dermatological Screening Service Simulation and Data
Visualisation
Sweden Analytic Imaging Swedish Energy User User Other National
Diagnostic Arena Agency

Member Policy Name Implementing Policy Level/area of education Government Source Level of
State body type Role government
EDUCATION
Austria Society for Federal Ministry Platform Research/Industry User OECD National
measurement, for Climate
automation and Action,
robotics Environment,
Energy, Mobility,
Innovation and
Technology
Belgium Smart Flemish Project Schools Facilitator National Regional
Education@schools Government Grants Strategy Report
i-Learn Flemish Agency Portal Schools User National Regional
for Innovation Strategy Report
and
Entrepreneurship
Klascement Flemish Portal Schools User National Regional
Government Strategy Report
Databuzz Flemish Workshop Schools Facilitator National Regional
Government Strategy Report
VubAI Experience Flemish Workshop Industry/Policymakers/NGOs Facilitator National Regional
Government Strategy Report
EluciDATA Lab Flemish Course Industry (tech) Facilitator National Regional
Government Strategy Report
MolenGeek Brussels Capital Course Industry Facilitator National Regional
Region Strategy Report
BeCode Brussels Capital Course Industry Facilitator National Regional
Region Strategy Report
Numeria Brussels Capital Course Industry Facilitator National Regional
Region Strategy Report
Jobnet Flemish Initiative Industry User National Regional
Employment and Strategy Report
Vocational
Training Service
Bulgaria AI Research Centre Ministry of Centre Industry Facilitator OECD National
of Excellence Economy

281
 Adopt AI Study

Denmark Exam control in Danish Ministry Initiative Schools User

schools: fighting of Education
fraud or profiling
students?

Estonia ProgeTiger Information Course Schools Facilitator National National

Technology Strategy Report
Foundation for
Education
University of Tartu´s University of MOOC All Facilitator National National
MOOC Tartu Strategy Report
Finland Elements of AI Ministry of MOOC All User OECD National
Course Economic Affairs
and Employment
France 4 Interdisciplinary Directorate Centre Industry/Higher Education Facilitator OECD National
Institutes of AI General for
Enterprises
Germany AI Campus Federal Ministry Centre Industry/Higher Education Facilitator National National
of Education and Strategy Report
Research
Helmholtz Federal Ministry Centre Higher Education Facilitator National National
Information and data of Education and Strategy Report
science academy Research
Ireland ADAPT Research Science Centre Industry/Higher Education Facilitator National National
Centre Foundation Strategy Report
Ireland
Italy Pierino – Platform for Ministry of Initiative Schools/Higher Education User AI-X National
the extraction and Education
retrieval of online
information
REDOC – Digital Redoc (private) Initiative Schools User JRC National
tutor to make easier
learning the STEM
subject
Respondus: Università Initiative Schools User JRC National
Assessment software Cattolica di
for remote Milano
management of
students
Luxembourg AI Academy learning Digital Academy Course Industry Facilitator National National
course Strategy Report
Malta AI Family Challenge Ministry of Workshop Schools Facilitator National National
Education and Strategy Report
Employment

282
 Adopt AI Study

AI Olympiad Ministry of Workshop Schools Facilitator National National

Education and Strategy Report
Employment
Netherlands National Data Statistics Course Higher Education Facilitator National National
Science Trainee Netherlands Strategy Report
Programme
National online Innovation MOOC All Facilitator National National
course in AI Center for AI Strategy Report
STAP-scheme Ministry of Social Course Industry Facilitator National National
Affairs and Strategy Report
Employment
Portugal Ciencia Viva clubs Directorate Workshop Schools Facilitator National National
General for Strategy Report
Education
MOOC on AI in Directorate MOOC All Facilitator National National
education General for Strategy Report
Education
NAU National Portal All User National National
Foundation for Strategy Report
Scientific
Computation
Slovakia AI courses for the Ministry of Course Industry Facilitator National National
employees of the Education Strategy Report
public sector
Slovenia MIRRI´s Kinit.SK KInIT Institute Platform Industry User OECD National
and Slovak.AI
platform

283
 Adopt AI Study

6.3 Chapter 3 & 4 Annexes

This Annex outlines the organisation of the stakeholder consultation, including
workshop running, collection and analysis of data of the sectoral dialogues.
The online sectoral dialogues aimed to collect original empirical data to better
understand, verify and complement findings and analyses carried within Tasks 1 and 2
of the study, with innovative ideas sourced directly from a wide spectrum of
stakeholders on how to accelerate the uptake and procurement of AI technologies in
the public sector.
The sectoral dialogues also helped understand the problems faced by
stakeholders in key sectors identified, public services need for AI as well as
requirements for adoption in public procurement processes, best practice
implementation, and user needs. The Policy Workshop was used to gather further
insights and practical policy solutions.
6.3.1 Methodology
The organisation of online sectoral dialogues
Organising the online sectoral dialogues involved inviting the participants, finalising
the agendas and shaping the facilitation of the workshops.
In line with the Coordinated Plan on Artificial Intelligence, the Sectoral Dialogue and
Policy Workshops were open, transparent and EU-wide.592 Openness and transparency
have been ensured by:
 launching an open survey which was open to everyone interested in
contributing and expressing their opinion on the topic - the survey was open
from June to September 2022 and gathered a total of 77 complete responses.
 sharing the results of the Sectoral Dialogue and Policy Workshops with all
participants in the workshops.
The dissemination of the online survey took place using multiple dissemination
channels. This dissemination was also used to inform stakeholders about the
upcoming workshops and ask for their expression of interest in participating in them.
Invitations were later sent to those stakeholders that expressed interest in
participating, and the study team created a simple registration portal to facilitate the
registration process and the follow-up with the participants. Following their
registration, participants were informed about the key questions that were to be
addressed during the events and what was expected of them as participants.
The agenda and workshop materials were also finalised in close cooperation with
DG CNCT and were shared with stakeholders ahead of the workshops (see section
6.3.2 for final workshop agendas and materials).
The mapping of stakeholders was done keeping in mind their level of influence and
engagement. The study team sent out invitations well in advance and closely
monitored the registration of participants. Moreover, the study team constantly sent
out invitations to a large number of possible participants, to ensure the required level
of attendance.
After the mapping was finalised, additional invitations were sent throughout August
and September. The workshops were carried out on the following dates:
 Health workshop on 26 September (10:00-12:30 CET)
 E-government workshop on 27 September (10:00-12:30 CET)
 Mobility (transport) workshop on 3 October (10:00-12:30 CET)

592
https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review

284
 Adopt AI Study

 Education workshop on 4 October (10:00-12:30 CET)

 Policy workshop on 17 October (10:30-13:00 CET)
The study team used the following software solutions to facilitate the online
workshops:
 Videoconferencing solution: Microsoft Teams
 Mural: digital whiteboarding where participants could share ideas using sticky
notes, shapes, images and videos and even drag and drop files directly onto the
Mural Board to bring ideas to life.
The virtual meetings were conducted in English and open discussions were encouraged
throughout all the designed sessions. Using digital whiteboard software (Mural), the
level of interactivity was increased and attracted a higher level of engagement from
the participants.
During the Policy workshop, the study team also explored the option of splitting the
participants into three different breakout rooms. This setup enabled both in-depth
specific discussions as well as cross-cutting discussions between stakeholder groups.
Structure of the workshops
The 4 Sectoral Workshops were focused on understanding the problems and needs
of stakeholders from the selected sectors. In these 2.5 hours workshops, facilitators
engaged with workshop participants (including businesses, sectoral organisations, and
public sector buyers), to clarify challenges, identify important organisational
considerations and surface opportunities for overcoming barriers to adopting AI in
public tendering and public services.
Based on the findings of the desk research, the study identified the main challenges
(issues/obstacles) that the Member States’ governments face when considering the
public procurement of AI technologies. Four main axes of challenges that may pose a
barrier to the public procurement and adoption of AI technologies in the public sector
(and private sector when relevant) guided the discussion throughout sectoral
dialogues:
 Procurement process challenges
 Data challenges
 AI technology challenges
 Organisational capacity challenges
 Sectoral and AI technologies-related challenges
The discussion on challenges was closely followed by stakeholders' consultation on
possible recommendations, best practices and sharing of success stories.
The Policy Workshop was focused on discussing policy recommendations and the
audience was formed of stakeholders about all 4 sectors involved in the Sectoral
Workshops. The 2.5 hours workshop built on the problems and issues identified in the
sectoral workshops and facilitated a solutions-oriented discussion aimed at
understanding the needs and priorities of the selected sector stakeholders and their
recommendations for future EU policy actions. Through open discussions facilitated
within breakout groups, the stakeholders also explored, compared and assessed the
feasibility of the potential solutions from the policy-making perspective.
6.3.2 Workshop materials
Information packs have been distributed to participants in advance of the virtual
workshops for them to reflect on the discussion topics and prepare. These contained
the agenda of the workshop with the discussion questions and emerging study findings

285
 Adopt AI Study

for each of the selected sectors and can be accessed under the documents linked
below.

Workshop material Workshop material Workshop material Workshop material

and Agenda - Mobility 3oct2022_final.pdf
and Agenda - E-Goverment 27Sep2022.pdf
and Agenda - Health 26Sep2022.pdf and Agenda - Education 4oct2022_final.pdf

After the finalisation of the online workshops, the resulting Mural boards have been
shared with registered stakeholders. These capture the main points of the discussions
which took place in each of the workshops organised and can be found under the
documents linked below.

Mural board Health Mural board Mural board Mobility Mural board
workshop.pdf E-government workshop.pdf workshop.pdf Education workshop.pdf

Mural board Policy

workshop.pdf

286
 Adopt AI Study

6.4 References
Accenture, 2019. Transforming Public Service with AI. URL:
https://www.accenture.com/_acnmedia/PDF-110/Accenture-Govtech-POV.pdf
Accenture. 2016. Why Artificial Intelligence is the Future of Growth. URL:
http://dl.icdst.org/pdfs/files2/2aea5d87070f0116f8aaa9f545530e47.pdf
AI for Humanity. 2018. For a meaningful Artificial Intelligence (Mission Villani Report).
URL: https://www.aiforhumanity.fr/pdfs/MissionVillani_Report_ENG-VF.pdf
Allied Market Research. 2020. Predictive Analytics Market by Component (Solution and
Services), Deployment (On-premise and Cloud), Enterprise Size (Large Enterprises
and Small & Medium-sized Enterprises), and Industry Vertical (BFSI, Retail, IT &
Telecom, Healthcare, Government, Manufacturing, and Others): Global Opportunity
Analysis and Industry Forecast, 2020–2027. URL:
https://www.alliedmarketresearch.com/predictive-analytics-market
Allied Market Research. 2020. Speech Recognition Market by Deployment Mode (On
Cloud and On-Premise) and End Use (Consumer Electronics, Healthcare, Enterprise,
and Others): Global Opportunity Analysis and Industry Forecast, 2019–2026. URL:
https://www.alliedmarketresearch.com/speech-recognition-market
Artificial Intelligence National Laboratory. URL: https://mi.nemzetilabor.hu/about-us
Austrian Federal Ministry of Education, Science and Research. 2021. Austrian Research
and Technology Report. URL:
https://www.bmbwf.gv.at/en/Topics/Research/Research-in-Austria/Services/FTB.html
Autonomous Systems National Laboratory. URL: https://autonom.nemzetilabor.hu/
BCC Research. 2020. Deep Learning: Global Markets. URL:
https://www.bccresearch.com/market-research/information-technology/deep-
learning-market.html
Boston Consulting Group. 2021. Which Sectors Perform Best in Digital Transformation?
URL: https://www.bcg.com/publications/2021/learning-from-successful-digital-leaders
Campion, A., Gasco-Hernandez, M., Jankin Mikhaylov, S. and Esteve, M., 2020.
Overcoming the challenges of collaboratively adopting artificial intelligence in the
public sector. Social Science Computer Review.
Communication from The Commission to the European Parliament, the European
Council, the Council, the European Economic and Social Committee and the
Committee of the Regions on Artificial Intelligence for Europe. 25 April 2018. URL:
https://ec.europa.eu/digital-single-market/en/news/communication-artificial-
intelligence-europe
Communication from the Commission to the European Parliament, the European
Council, the Council, the European Economic and Social Committee and the
Committee of the Regions. A New Industrial Strategy for Europe. COM/2020/102 final.
URL: https://eur-lex.europa.eu/legal-
content/EN/TXT/?qid=1593086905382&uri=CELEX%3A52020DC0102
Deloitte. 2015. Technology and people: the great job-creating machine. URL:
https://www2.deloitte.com/content/dam/Deloitte/uk/Documents/finance/deloitte-uk-
technology-and-people.pdf
Digitalwallonia.be. 10 March 2020. Digital Wallonia4.ai: l’appel à projets Tremplin IA
est lancé. URL: https://www.digitalwallonia.be/fr/publications/dw4ai-tremplin-ia
Digitalwallonia.be. 29 October 2021. Start IA, explorez le potentiel de la donnée et de
l’IA pour votre entreprise. URL: https://www.digitalwallonia.be/fr/publications/appel-
start-ia-05

287
 Adopt AI Study

Engstrom, D. F., Ho, D. E., Sharkey, C. M., & Cuéllar, M.F. 2020. Government by
Algorithm: Artificial Intelligence in Federal Administrative Agencies. In SSRN Electronic
Journal. URL: https://doi.org/10.2139/ssrn.3551505
European Commission Joint Research Centre. 25 May 2022. AI Watch: Road to the
adoption of AI by the Public Sector. URL: https://op.europa.eu/en/publication-detail/-
/publication/34251428-dc12-11ec-a534-01aa75ed71a1
European Commission, Directorate-General for Communications Networks, Content
and Technology. 2020. European enterprise survey on the use of technologies based
on artificial intelligence: final report. URL: https://data.europa.eu/doi/10.2759/759368
European Commission. 2020. AI Watch Artificial Intelligence in public services –
Overview of the use and impact of AI in public services in the EU. URL:
https://publications.jrc.ec.europa.eu/repository/handle/JRC120399
European Commission. 2021. Digital Economy and Society Index (DESI) 2021. Human
capital.
European Commission. 2021. The 2021 EU Industrial R&D Investment Scoreboard.
URL: https://iri.jrc.ec.europa.eu/sites/default/files/contentype/scoreboard/2021-
12/EU%20RD%20Scoreboard%202021%20FINAL%20online.pdf ;
https://iri.jrc.ec.europa.eu/scoreboard/2021-eu-industrial-rd-investment-scoreboard
European Commission. A European approach to artificial intelligence. URL:
https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-
intelligence
European Commission. Cordis. URL: https://cordis.europa.eu/
European Commission. Industrial applications of artificial intelligence and big data.
URL: https://ec.europa.eu/growth/industry/strategy/advanced-technologies/industrial-
applications-artificial-intelligence-and-big-data_en
European Commission. InvestEU. URL: https://europa.eu/investeu/home_en
European Commission. Supporting policy with scientific evidence. URL:
https://knowledge4policy.ec.europa.eu/home_en
European Investment Bank. 2021. EIB Investment Survey. URL:
https://www.eib.org/attachments/publications/eibis_2021_european_union_en.pdf
European Parliament. 2015. Industry 4.0. Digitalisation for productivity and growth.
URL:
https://www.europarl.europa.eu/RegData/etudes/BRIE/2015/568337/EPRS_BRI(2015
)568337_EN.pdf
European Parliament. 2021. Artificial Intelligence and public services. URL :
https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/662936/IPOL_BRI(2021)
662936_EN.pdf
Eurostat. 2021. Artificial Intelligence in EU enterprises. URL:
https://ec.europa.eu/eurostat/web/products-eurostat-news/-/ddn-20210413-1
Eurostat. 2021. How digitalised are the EU’s enterprises? URL:
https://ec.europa.eu/eurostat/web/products-eurostat-news/-/ddn-20211029-1
Forbes. 2021. A Market to Harness: Speech Recognition Artificial Intelligence (AI)
Innovations On The Rise. URL:
https://www.forbes.com/sites/cindygordon/2021/12/23/a-market-to-harness-speech-
recognition-artificial-intelligence-ai-innovations-on-the-rise/?sh=5fe17773134d
Gartner. 2019. Gartner Survey Shows 37 Percent of Organizations Have Implemented
AI in Some Form. URL: https://www.gartner.com/en/newsroom/press-releases/2019-
01-21-gartner-survey-shows-37-percent-of-organizations-have

288
 Adopt AI Study

Gartner. 2021. The 4 Trends That Prevail on the Gartner Hype Cycle for AI, 2021.
URL: https://www.gartner.com/en/articles/the-4-trends-that-prevail-on-the-gartner-
hype-cycle-for-ai-2021
Global Government Forum. 2021. Making Artificial Intelligence fit for life inside the
government. URL: https://www.globalgovernmentforum.com/making-artificial-
intelligence-fit-for-a-life-inside-government/
Government of Ireland. 2021. AI – Here for Good. A National Artificial Intelligence
Strategy for Ireland. URL: https://assets.gov.ie/152580/b8ad2fa0-9ef2-44da-bab6-
aaf8bb03c898.pdf
Government Offices of Sweden. 2018. National Approach to Artificial Intelligence. URL:
https://www.government.se/4a7451/contentassets/fe2ba005fb49433587574c513a83
7fac/national-approach-to-artificial-intelligence.pdf
Grand View Research. 2019. Machine Learning Market Size, Share & Trends Analysis
Report By Component, By Enterprise Size, By End Use (Healthcare, BFSI, Law, Retail,
Advertising & Media), And Segment Forecasts, 2019 – 2025. URL:
https://www.grandviewresearch.com/industry-analysis/machine-learning-market
Grand View Research. 2019. U.S. Voice Recognition Market Size, Share & Trends
Analysis Report By Vertical (Automotive, Enterprise, Consumer, Banking, Financial
Services & Insurance, Government, Retail, Healthcare, Military, Legal, Education), And
Segments Forecasts, 2019 – 2025. URL:
https://www.grandviewresearch.com/industry-analysis/us-voice-recognition-market
Innoviris Brussels. URL: https://innoviris.brussels/
International Data Corporation (IDC). 2021. European Spending on Artificial
Intelligence Will Reach $22 Billion in 2022, Supported by Strong Investments Across
Banking and Manufacturing. URL:
https://www.idc.com/getdoc.jsp?containerId=prEUR148297521
International Data Corporation (IDC). 2021. Worldwide Artificial Intelligence Spending
Guide. URL:
https://www.idc.com/tracker/showproductinfo.jsp?containerId=IDC_P33198
Kankanhalli, A., Charalabidis, Y., & Mellouli, S. (2019). IoT and AI for smart
government: A research agenda. Government Information Quarterly, 36(2), 304–309.
KBV Research. 2019. Global Natural Language Processing Market By Component
(Solution and Services) By Application (Text Classification, Machine Translation,
Question Answering, Sentiment Analysis, Information Extraction, Automatic
Summarization and Others) By Type (Rule-Based, Statistical and Hybrid) By
Deployment Type (On-premise and Cloud) By Region, Industry Analysis and Forecast,
2019 – 2025. URL: https://www.kbvresearch.com/natural-language-processing-
market/
KBV Research. 2020. Global Computer Vision Market By Product Type (PC-Based and
Smart Camera-Based), By Component (Hardware and Software), By Application
(Quality Assurance & Inspection, Measurement, Identification, Predictive Maintenance,
Positioning & Guidance and 3D Visualization & Interactive 3D Modeling), By Vertical
(Industrial and Non-Industrial), By Region, Industry Analysis and Forecast, 2020 –
2026. URL: https://www.kbvresearch.com/computer-vision-market/
KBV Research. 2021. Global Smart Robots Market By Component (Hardware and
Software), By Mobility (Mobile and Stationary), By Operating Environment (Ground
and Underwater), By End User (Military & Defense, Logistics Management, Inspection
& Maintenance, Field/Agricultural, Healthcare, Industrial, Domestic and Others), By
Region, Industry Analysis and Forecast, 2020 – 2026/ URL:
https://www.kbvresearch.com/smart-robots-market/

289
 Adopt AI Study

Markets and Markets. 2021. Chatbot Market by Component, Type (Rule Based & AI
Based), Application (Customer Service, Customer Engagement & Retention), Channel
Integration, Business Function (ITSM, Finance), Vertical, and Region - Global Forecast
to 2026. URL: https://www.marketsandmarkets.com/Market-Reports/smart-advisor-
market-72302363.html
Markets and Markets. 2021. Predictive Analytics Market with Covid-19 Impact Analysis
by Solution (Financial Analytics, Risk Analytics, Marketing Analytics, Web & Social
Media Analytics), Service, Deployment Mode, Organization Size, Vertical, and Region -
Global Forecast to 2026. URL: https://www.marketsandmarkets.com/Market-
Reports/predictive-analytics-market-1181.html
McBride, K., van Noordt, C., Misuraca, G. and Hammerschmid, G., 2021. Towards a
Systematic Understanding of the Challenges of Procuring Artificial Intelligence in the
Public Sector. (Final version not yet published). URL:
http://dx.doi.org/10.31235/osf.io/un649
McKinsey & Company. 2020. The state of AI in 2020. URL:
https://www.mckinsey.com/business-functions/mckinsey-analytics/our-
insights/global-survey-the-state-of-ai-in-2020
McKinsey & Company. 2021. The state of AI in 2021. URL:
https://www.mckinsey.com/business-functions/mckinsey-analytics/our-
insights/global-survey-the-state-of-ai-in-2021
McKinsey Global Institute. 2019. Reviving innovation in Europe. URL:
https://www.mckinsey.com/featured-insights/innovation-and-growth/reviving-
innovation-in-europe
Mordor Intelligence. 2021. Artificial Intelligence Market - Growth, Trends, Covid-19
Impact, Forecast (2022 - 2027). URL: https://www.mordorintelligence.com/industry-
reports/global-artificial-intelligence-market
Mordor Intelligence. 2021. Threat Intelligence Market – Growth, Trends, Covid-19
impact, and Forecasts (2022- 2027). URL:
https://www.mordorintelligence.com/industry-reports/threat-intelligence-market
Mordor Intelligence. Natural Language Processing (NLP) Market – Growth, Trends,
COVID-19 Impact, and Forecasts (2022 - 2027). URL:
https://www.mordorintelligence.com/industry-reports/natural-language-processing-
market
Mucha, T. and Seppälä, T., 2021. Estimating firm digitalization: A method for
disaggregating sector-level digital intensity to firm-level. MethodsX, 8, p.101233. URL:
https://www.sciencedirect.com/science/article/pii/S2215016121000261
Narrative Science. 2019. Outlook on Artificial Intelligence in the Enterprise. URL:
https://narrativescience.com/wp-content/uploads/2019/02/Research-Report_Outlook-
on-AI-for-the-Enterprise.pdf
O’Reilly. 2021. AI Adoption in the Enterprise 2021. URL:
https://www.oreilly.com/radar/ai-adoption-in-the-enterprise-2021/
OECD. 2018. A taxonomy of digital-intensive sectors. OECD Science, Technology and
Industry Working Papers. URL: https://www.oecd-ilibrary.org/docserver/f404736a-
en.pdf?expires=1647547698&id=id&accname=guest&checksum=21B7A68786505330
6A0ECD363169EF44
OECD. National AI policies & strategies. URL: https://oecd.ai/en/dashboards
Orbis Research. 2022. Global Knowledge Management Software Market Growth
(Status and Outlook) 2022-2028. URL:
https://www.orbisresearch.com/reports/index/global-knowledge-management-
software-market-growth-status-and-outlook-2022-2028

290
 Adopt AI Study

PricewaterhouseCoopers. 2017. Sizing the prize. Global Artificial Intelligence Study:

Exploiting the AI Revolution. URL: https://www.pwc.com/gx/en/issues/data-and-
analytics/publications/artificial-intelligence-study.html
PricewaterhouseCoopers. 2017. Sizing the prize. What’s the real value of AI for your
business and how can you capitalise? URL:
https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-prize-
report.pdf
Smart Mobility Lab. URL: https://smartmobilitylab.sk/
Statista. 2019. Machine Learning Tops AI Dollars. URL: https://lb-aps-
frontend.statista.com/chart/17966/worldwide-artificial-intelligence-funding/
Statista. https://www.statista.com/statistics/1133875/global-voice-recognition-
market-size/
Statistical Analysis System (SAS). N.d. Predictive analytics. What it is and why it
matters. URL: https://www.sas.com/en_au/insights/analytics/predictive-analytics.html
UK Government Digital Service. 2020. A guide to using artificial intelligence in the
public sector. URL:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attach
ment_data/file/979892/A_guide_to_using_AI_in_the_public_sector__Print_version_.p
df
UK Government Digital Service. 2020. Guidelines for AI procurement. URL:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attach
ment_data/file/990469/Guidelines_for_AI_procurement.pdf
Urbike. URL: https://urbike.be/
Wolters Kluwer. 2020. How natural language processing impacts professions. URL:
https://www.wolterskluwer.com/en/expert-insights/how-natural-language-processing-
impacts-professions
World Bank. 2020. Artificial Intelligence in the Public Sector: Maximizing
Opportunities, Managing Risks. URL:
https://openknowledge.worldbank.org/handle/10986/35317
World Economic Forum. 2019. 5 challenges for government adoption of AI. URL:
https://www.weforum.org/agenda/2019/08/artificial-intelligence-government-public-
sector/
World Economic forum. 2020. AI Procurement in a Box: Project overview. URL:
https://www3.weforum.org/docs/WEF_AI_Procurement_in_a_Box_Project_Overview_
2020.pdf
World Economic Forum. 2020. AI Procurement in a Box: AI Government Procurement
guidelines. URL:
https://www2.deloitte.com/content/dam/Deloitte/global/Documents/About-
Deloitte/gx-wef-ai-government-procurement-guidelines-2020.pdf
World Economic Forum. 2020. Don’t fear AI. It will lead to long-term job growth. URL:
https://www.weforum.org/agenda/2020/10/dont-fear-ai-it-will-lead-to-long-term-job-
growth/
World Economic Forum. 2020. How governments can use public procurement to shape
the future of AI regulation – and boost innovation and growth. URL:
https://www.weforum.org/agenda/2020/06/artificial-intelligence-ai-government-
procurement-standards-regulation-economic-growth-covid-19-response/
World Economic Forum. 2022. How deep learning can improve productivity and boost
business? URL: https://www.weforum.org/agenda/2022/01/deep-learning-business-
productivity-revenue/

291
 7 HOW TO OBTAIN EU PUBLICATIONS

7.1 Free publications:

 one copy:

via EU Bookshop (http://bookshop.europa.eu);

 more than one copy or posters/maps:
from the European Union’s representations
(http://ec.europa.eu/represent_en.htm);
from the delegations in non-EU countries
(http://eeas.europa.eu/delegations/index_en.htm);
by contacting the Europe Direct service
(http://europa.eu/europedirect/index_en.htm) or calling 00 800 6 7 8 9 10 11
(freephone number from anywhere in the EU) (*).

(*) The information given is free, as are most calls (though some operators, phone boxes or hotels
may charge you).

7.2 Priced publications:

 via EU Bookshop (http://bookshop.europa.eu).

7.3 Priced subscriptions:

 via one of the sales agents of the Publications Office of the European Union
(http://publications.europa.eu/others/agents/index_en.htm).

Directorate-General for Communications Networks, Content and Technology (CNECT)