Artificial Intelligence and Data Science in The Au

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/319534479
Artiﬁcial Intelligence and Data Science in the Automotive Industry
Article · September 2017
CITATIONS READS
11 1,917
3 authors, including:
Florian Neukart Thomas Bäck

Leiden University Leiden University
58 PUBLICATIONS 269 CITATIONS 433 PUBLICATIONS 19,856 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
DELIVER View project
Process mining for multi-objective online control View project
All content following this page was uploaded by Florian Neukart on 25 December 2017.
The user has requested enhancement of the downloaded file.

Artificial Intelligence and Data Science in the
Automotive Industry
Martin Hofmann1, Florian Neukart2,3, Thomas Bäck3

*1,2
Volkswagen AG, 3Leiden University
*1 *2 *3
martin.hofmann@volkswagen.de; florian.neukart@vw.com; t.h.w.baeck@liacs.leidenuniv.nl
Abstract people, opening the door to a world of fundamentally new

Data science and machine learning are the key technologies when it comes possibilities. In fact, the now already implementable idea of
to the processes and products with automatic learning and optimization to autonomous driving is virtually a tangible reality for many
be used in the automotive industry of the future. This article defines the
drivers today with the help of lane keeping assistance and
terms “data science” (also referred to as “data analytics”) and “machine
learning” and how they are related. In addition, it defines the term
adaptive cruise control systems in the vehicle.
“optimizing analytics“ and illustrates the role of automatic optimization as The fact that this is just the tip of the iceberg, even in the
a key technology in combination with data analytics. It also uses examples automotive industry, becomes readily apparent when one
to explain the way that these technologies are currently being used in the
considers that, at the end of 2015, Toyota and Tesla's
automotive industry on the basis of the major subprocesses in the
automotive value chain (development, procurement; logistics, production,
founder, Elon Musk, each announced investments
marketing, sales and after-sales, connected customer). Since the industry is amounting to one billion US dollars in artificial intelligence
just starting to explore the broad range of potential uses for these research and development almost at the same time. The
technologies, visionary application examples are used to illustrate the
trend towards connected, autonomous, and artificially
revolutionary possibilities that they offer. Finally, the article demonstrates
how these technologies can make the automotive industry more efficient
intelligent systems that continuously learn from data and are
and enhance its customer focus throughout all its operations and activities, able to make optimal decisions is advancing in ways that are
extending from the product and its development process to the customers simply revolutionary, not to mention fundamentally
and their connection to the product.
important to many industries. This includes the automotive
Keywords industry, one of the key industries in Germany, in which
data science, big data, machine learning, automatic optimization, international competitiveness will be influenced by a new
optimizing analytics, automotive industry factor in the near future – namely the new technical and
service offerings that can be provided with the help of data
1 Introduction science and machine learning.
This article provides an overview of the corresponding
Data science and machine learning are now key
methods and some current application examples in the
technologies in our everyday lives, as we can see in a
automotive industry. It also outlines the potential
multitude of applications, such as voice recognition in
applications to be expected in this industry very soon.
vehicles and on cell phones, automatic facial and traffic sign
Accordingly, sections 2 and 3 begin by addressing the
recognition, as well as chess and, more recently, Go
subdomains of data mining (also referred to as “big data
machine algorithms1 which humans can no longer beat. The
analytics”) and artificial intelligence, briefly summarizing
analysis of large data volumes based on search, pattern
the corresponding processes, methods, and areas of
recognition, and learning algorithms provides insights into
application and presenting them in context. Section 4 then
the behavior of processes, systems, nature, and ultimately
provides an overview of current application examples in the
automotive industry based on the stages in the industry's
1
D. Silver et. al.: Mastering the Game of Go with Deep Neural value chain –from development to production and logistics
Networks and Tree Search, Nature 529, 484-489 (January 28,
through to the end customer. Based on such an example,
2016).
JOURNAL TITLE - MONTH YEAR 1

section 5 describes the vision for future applications using application is used for optimizing purposes, admittedly with
three examples: one in which vehicles play the role of an intermediate human step.
autonomous agents that interact with each other in cities, Within this context, another important aspect is the fact that
one that covers integrated production optimization, and one multiple criteria required for the relevant application often
that describes companies themselves as autonomous agents. need to be optimized at the same time, meaning that multi-
Whether these visions will become a reality in this or any criteria optimization methods – or, more generally, multi-
other way cannot be said with certainty at present – criteria decision-making support methods – are necessary.
however, we can safely predict that the rapid rate of These methods can then be used in order to find the best
development in this area will lead to the creation of possible compromises between conflicting goals. The
completely new products, processes, and services, many of examples mentioned include the frequently occurring
which we can only imagine today. This is one of the conflicts between cost and quality, risk and profit, and, in a
conclusions drawn in section 6, together with an outlook more technical example, between the weight and passive
regarding the potential future effects of the rapid rate of occupant safety of a body.
development in this area. Optimzing “What am I Decision
Analytics supposed to do?” support,
2 The data mining process multicriterial
Gartner uses the term “prescriptive analytics“ to describe optimization
the highest level of ability to make business decisions on the Predictive “What will Modelling
basis of data-based analyses. This is illustrated by the Analytics happen?”
question “what should I do?” and prescriptive analytics Diagnostic “Why did it BI, Modelling
supplies the required decision-making support, if a person is Analytics happen?”
still involved, or automation if this is no longer the case. Descriptive “What Business
The levels below this, in ascending order in terms of the use Analytics happened?” Intelligence
and usefulness of AI and data science, are defined as Figure 1: The four levels of data analysis usage within a
follows: descriptive analytics (“what has happened?”), company
diagnostic analytics (“why did it happen?”), and predictive
These four levels form a framework, within which it is
analytics (“what will happen?”) (see Figure 1). The last two
possible to categorize data analysis competence and
levels are based on data science technologies, including data
mining and statistics, while descriptive analytics essentially potential benefits for a company in general. This framework
is depicted in Figure 1 and shows the four layers which
uses traditional business intelligence concepts (data
build upon each other, together with the respective
warehouse, OLAP).
technology category required for implementation.
In this article, we seek to replace the term “prescriptive
The traditional Cross-Industry Standard Process for Data
analytics“ with the term “optimizing analytics.“ The reason
Mining (CRISP-DM)2 includes no optimization or decision-
for this is that a technology can “prescribe” many things,
making support whatsoever. Instead, based on the business
while, in terms of implementation within a company, the
goal is always to make something “better” with regard to understanding, data understanding, data preparation,
modeling, and evaluation sub-steps, CRISP proceeds
target criteria or quality criteria. This optimization can be
directly to the deployment of results in business processes.
supported by search algorithms, such as evolutionary
Here too, we propose an additional optimization step that in
algorithms in nonlinear cases and operation research (OR)
turn comprises multi-criteria optimization and decision-
methods in – much rarer – linear cases. It can also be
making support. This approach is depicted schematically in
supported by application experts who take the results from
Figure 2.
the data mining process and use them to draw conclusions
regarding process improvement. One good example are the
decision trees learned from data, which application experts
can understand, reconcile with their own expert knowledge, 2
and then implement in an appropriate manner. Here too, the https://en.wikipedia.org/wiki/Cross_Industry_Standard_Process_fo

r_Data_Mining
2 JOURNAL TITLE - MONTH YEAR

Business Understanding Data Understanding generate data, and integrate and process it in such a way that
data mining algorithms can be applied to it. In addition,
automatic modeling and automatic optimization are
Data Preparation necessary in order to update models and use them as a basis
Data
for generating optimal proposed actions in online
Modeling
applications. These actions can then be communicated to the
process expert as a suggestion or – especially in the case of
Deployment Evaluation
continuous production processes – be used directly to
Optimization
control the respective process. If sensor systems are also
integrated directly into the production process – to collect
Figure 2: Traditional CRISP-DM process with an data in real time – this results in a self-learning cyber-
additional optimization step physical system 3
that facilitates implementation of the
4
It is important to note that the original CRISP model deals Industry 4.0 vision in the field of production engineering.
with a largely iterative approach used by data scientists to
Multi-criteria optimization
analyze data manually, which is reflected in the iterations
Automatic modeling
between business understanding and data understanding as
Data management
well as data preparation and modeling. However,
Sensor system Control system/actuators
evaluating the modeling results with the relevant application
Machine/system
experts in the evaluation step can also result in having to
start the process all over again from the business Figure 3: Architecture of an Industry 4.0 model for
understanding sub-step, making it necessary to go through optimizing analytics
all the sub-steps again partially or completely (e.g., if
additional data needs to be incorporated). This approach is depicted schematically in Figure 3. Data
The manual, iterative procedure is also due to the fact that from the system is acquired with the help of sensors and
the basic idea behind this approach – as up-to-date as it integrated into the data management system. Using this as a
may be for the majority of applications – is now almost 20 basis, forecast models for the system's relevant outputs
years old and certainly only partially compatible with a big (quality, deviation from target value, process variance, etc.)
data strategy. The fact is that, in addition to the use of are used continuously in order to forecast the system's
nonlinear modeling methods (in contrast to the usual output. Other machine learning options can be used within
generalized linear models derived from statistical modeling) this context in order, for example, to predict maintenance
and knowledge extraction from data, data mining rests on results (predictive maintenance) or to identify anomalies in
the fundamental idea that models can be derived from data the process. The corresponding models are monitored
with the help of algorithms and that this modeling process continuously and, if necessary, automatically retrained if
can run automatically for the most part – because the any process drift is observed. Finally, the multi-criteria
algorithm “does the work.” optimization uses the models to continuously compute
In applications where a large number of models need to be 3
Systems “in which information and software components are
created, for example for use in making forecasts (e.g., sales connected to mechanical and electronic components and in which
forecasts for individual vehicle models and markets based data is transferred and exchanged, and monitoring and control
tasks are carried out, in real-time using infrastructures such as the
on historical data), automatic modeling plays an important Internet.” (Translation of the following article in Gabler
role. The same applies to the use of online data mining, in Wirtschaftslexikon, Springer:
http://wirtschaftslexikon.gabler.de/Definition/cyber-physische-
which, for example, forecast models (e.g., for forecasting systeme.html).
4
product quality) are not only constantly used for a Industry 4.0 is defined therein as “a marketing term that is also
used in science communication and refers to a ‘future project’ of
production process, but also adapted (i.e., retrained)
the German federal government. The so-called ‘Fourth Industrial
continuously whenever individual process aspects change Revolution’ is characterized by the customization and
(e.g., when a new raw material batch is used). This type of hybridization of products and the integration of customers and
business partners into business processes.” (Translation of the
application requires the technical ability to automatically following article in Gabler Wirtschaftslexikon, Springer:
http://wirtschaftslexikon.gabler.de/Definition/industrie-4-0.html).

optimum setpoints for the system control. Human process planning in areas such as production and logistics, among
experts can also be integrated here by using the system as a many others, can only be touched upon within the scope of
suggestion generator so that a process expert can evaluate this article.
the generated suggestions before they are implemented in
the original system. 3. The pillars of artificial intelligence
In order to differentiate it from “traditional” data mining, An early definition of artificial intelligence from the IEEE
the term “big data” is frequently defined now with three Neural Networks Council was “the study of how to make
(sometimes even four or five) essential characteristics: computers do things at which, at the moment, people are
volume, velocity, and variety, which refer to the large better.”5 Although this still applies, current research is also
volume of data, the speed at which data is generated, and the focused on improving the way that software does things at
heterogeneity of the data to be analyzed, which can no which computers have always been better, such as analyzing
longer be categorized into the conventional relational large amounts of data. Data is also the basis for developing
database schema. Veracity, i.e., the fact that large artificially intelligent software systems not only to collect
uncertainties may also be hidden in the data (e.g., information, but also to:
measurement inaccuracies), and finally value, i.e., the value
• Learn
that the data and its analysis represents for a company's
business processes, are often cited as additional • Understand and interpret information
characteristics. So it is not just the pure data volume that • Behave adaptively
distinguishes previous data analytics methods from big data,
but also other technical factors that require the use of new • Plan
methods– such as Hadoop and MapReduce – with • Make inferences
appropriately adapted data analysis algorithms in order to
• Solve problems
allow the data to be saved and processed. In addition, so-
called “in-memory databases” now also make it possible to • Think abstractly
apply traditional learning and modeling algorithms in main
• Understand and interpret ideas and language
memory to large data volumes.
This means that if one were to establish a hierarchy of data 3.1 Machine learning
analysis and modeling methods and techniques, then, in
At the most general level, machine learning (ML)
very simplistic terms, statistics would be a subset of data
algorithms can be subdivided into two categories:
mining, which in turn would be a subset of big data. Not
supervised and unsupervised, depending on whether or not
every application requires the use of data mining or big data
the respective algorithm requires a target variable to be
technologies. However, a clear trend can be observed, which
specified.
indicates that the necessities and possibilities involved in the
use of data mining and big data are growing at a very rapid Supervised learning algorithms
pace as increasingly large data volumes are being collected Apart from the input variables (predictors), supervised
and linked across all processes and departments of a learning algorithms also require the known target values
company. Nevertheless, conventional hardware architecture (labels) for a problem. In order to train an ML model to
with additional main memory is often more than sufficient identify traffic signs using cameras, images of traffic signs –
for analyzing large data volumes in the gigabyte range. preferably with a variety of configurations – are required as
Although optimizing analytics is of tremendous importance, input variables. In this case, light conditions, angles, soiling,
it is also crucial to always be open to the broad variety of etc. are compiled as noise or blurring in the data;
applications when using artificial intelligence and machine nonetheless, it must be possible to recognize a traffic sign in
learning algorithms. The wide range of learning and search rainy conditions with the same accuracy as when the sun is
methods, with potential use in applications such as image shining. The labels, i.e., the correct designations, for such
and language recognition, knowledge learning, control and
5
E. Rich, K. Knight: Artificial Intelligence, 5, 1990

data are normally assigned manually. This correct set of In other words, machine learning is the area of artificial
input variables and their correct classification constitute a intelligence (AI) that enables computers to learn without
training data set. Although we only have one image per being programmed explicitly. Machine learning focuses on
training data set in this case, we still speak of multiple input developing programs that grow and change by themselves
variables, since ML algorithms find relevant features in as soon as new data is provided. Accordingly, processes that
training data and learn how these features and the class can be represented in a flowchart are not suitable candidates
assignment for the classification task indicated in the for machine learning – in contrast, everything that requires
example are associated. Supervised learning is used dynamic and changing solution strategies and cannot be
primarily to predict numerical values (regression) and for constrained to static rules is potentially suitable for solution
classification purposes (predicting the appropriate class), with ML. For example, ML is used when:
and the corresponding data is not limited to a specific
• No relevant human expertise exists
format – ML algorithms are more than capable of
processing images, audio files, videos, numerical data, and • People are unable to express their expertise
text. Classification examples include object recognition
• The solution changes over time
(traffic signs, objects in front of a vehicle, etc.), face
recognition, credit risk assessment, voice recognition, and • The solution needs to be adapted to specific cases
customer churn, to name but a few. In contrast to statistics, which follows the approach of
Regression examples include determining continuous making inferences based on samples, computer science is
numerical values on the basis of multiple (sometimes interested in developing efficient algorithms for solving
hundreds or thousands) input variables, such as a self- optimization problems, as well as in developing a
driving car calculating its ideal speed on the basis of road representation of the model for evaluating inferences.
and ambient conditions, determining a financial indicator Methods frequently used for optimization in this context
such as gross domestic product based on a changing number include so-called “evolutionary algorithms” (genetic
of input variables (use of arable land, population education algorithms, evolution strategies), the basic principles of
levels, industrial production, etc.), and determining potential which emulate natural evolution6. These methods are very
market shares with the introduction of new models. Each of efficient when applied to complex, nonlinear optimization
these problems is highly complex and cannot be represented problems.
by simple, linear relationships in simple equations. Or, to Even though ML is used in certain data mining applications,
put it another way that more accurately represents the and both look for patterns in data, ML and data mining are
enormous challenge involved: the necessary expertise does not the same thing. Instead of extracting data that people can
not even exist. understand, as is the case with data mining, ML methods are
Unsupervised learning algorithms used by programs to improve their own understanding of the
data provided. Software that implements ML methods
Unsupervised learning algorithms do not focus on individual
recognizes patterns in data and can dynamically adjust the
target variables, but instead have the goal of characterizing a
behavior based on them. If, for example, a self-driving car
data set in general. Unsupervised ML algorithms are often
(or the software that interprets the visual signal from the
used to group (cluster) data sets, i.e., to identify
corresponding camera) has been trained to initiate a braking
relationships between individual data points (that can
maneuver if a pedestrian appears in front it, this must work
consist of any number of attributes) and group them into
with all pedestrians regardless of whether they are short,
clusters. In certain cases, the output from unsupervised ML
tall, fat, thin, clothed, coming from the left, coming from the
algorithms can in turn be used as an input for supervised
right, etc. In turn, the vehicle must not brake if there is a
methods. Examples of unsupervised learning include
stationary garbage bin on the side of the road.
forming customer groups based on their buying behavior or
demographic data, or clustering time series in order to group The level of complexity in the real world is often greater
millions of time series from sensors into groups that were 6
Th. Bäck, D.B. Fogel, Z. Michalewicz: Handbook of
previously not obvious. Evolutionary Computation, Institute of Physics Publishing, New
York, 1997.

than the level of complexity of an ML model, which is why, obstacle that needs to be evaded, but is also probably not
in most cases, an attempt is made to subdivide problems into sleeping and a medical emergency is occurring.
subproblems and then apply ML models to these
Vision in biological organisms is regarded as an active
subproblems. The output from these models is then
process that includes controlling the sensor and is tightly
integrated in order to permit complex tasks, such as 7
linked to successful performance of an action .
autonomous vehicle operation, in structured and 8
Consequently, CV systems are not passive either. In other
unstructured environments.
words, the system must:
3.2 Computer vision • Be continuously provided with data via sensors

(streaming)
Computer vision (CV) is a very wide field of research that
merges scientific theories from various fields (as is often the • Act based on this data stream
case with AI), starting from biology, neuroscience, and
Having said that, the goal of CV systems is not to
psychology and extending all the way to computer science,
understand scenes in images – first and foremost, the
mathematics, and physics. First, it is important to know how
systems must extract the relevant information for a specific
an image is produced physically. Before light hits sensors in
task from the scene. This means that they must identify a
a two-dimensional array, it is refracted, absorbed, scattered,
“region of interest” that will be used for processing.
or reflected, and an image is produced by measuring the
Moreover, these systems must feature short response times,
intensity of the light beams through each element in the
since it is probable that scenes will change over time and
image (pixel). The three primary focuses of CV are:
that a heavily delayed action will not achieve the desired
• Reconstructing a scene and the point from which effect. Many different methods have been proposed for
the scene is observed based on an image, an image object recognition purposes (“what” is located “where” in a
sequence, or a video. scene), including:
• Emulating biological visual perception in order to • Object detectors, in which case a window moves
better understand which physical and biological over the image and a filter response is determined
processes are involved, how the wetware works, for each position by comparing a template and the
and how the corresponding interpretation and sub-image (window content), with each new object
understanding work. parameterization requiring a separate scan. More
sophisticated algorithms simultaneously make
• Technical research and development focuses on
calculations based on various scales and apply
efficient, algorithmic solutions – when it comes to
filters that have been learned from a large number
CV software, problem-specific solutions that only
of images.
have limited commonalities with the visual
perception of biological organisms are often • Segment-based techniques extract a geometrical
developed. description of an object by grouping pixels that
define the dimensions of an object in an image.
All three areas overlap and influence each other. If, for
Based on this, a fixed feature set is computed, i.e.,
example, the focus in an application is on obstacle
the features in the set retain the same values even
recognition in order to initiate an automated braking
when subjected to various image transformations,
maneuver in the event of a pedestrian appearing in front of
such as changes in light conditions, scaling, or
the vehicle, the most important thing is to identify the
rotation. These features are used to clearly identify
pedestrian as an obstacle. Interpreting the entire scene – e.g.,
objects or object classes, one example being the
understanding that the vehicle is moving towards a family
aforementioned identification of traffic signs.
having a picnic in a field – is not necessary in this case. In
contrast, understanding a scene is an essential prerequisite if
7
context is a relevant input, such as is the case when R. Bajcsy: Active perception, Proceedings of the IEEE, 76:996-
1005, 1988
developing domestic robots that need to understand that an 8
J. L. Crowley, H. I. Christensen: Vision as a Process: Basic
occupant who is lying on the floor not only represents an Research on Computer Vision Systems, Berlin: Springer, 1995

• Alignment-based methods use parametric object Other research directions include tracking 11 , 12 , contextual
models that are trained on data 9 , 10
. Algorithms scene understanding, 13,14 and monitoring 15 , although these
search for parameters, such as scaling, translation, aspects are currently of secondary importance to the
or rotation, that adapt a model optimally to the automotive industry.
corresponding features in the image, whereby an
approximated solution can be found by means of a 3.3 Inference and decision-making
reciprocal process, i.e., by features, such as This field of research, referred to in the literature as
contours, corners, or others, “selecting” “knowledge representation & reasoning” (KRR), focuses on
characteristic points in the image for parameter designing and developing data structures and inference
solutions that are compatible with the found algorithms. Problems solved by making inferences are very
feature. often found in applications that require interaction with the
With object recognition, it is necessary to decide whether physical world (humans, for example), such as generating
algorithms need to process 2-D or 3-D representations of diagnostics, planning, processing natural languages,
objects – 2-D representations are very frequently a good answering questions, etc. KRR forms the basis for AI at the
compromise between accuracy and availability. Current human level.
research (deep learning) shows that even distances between Making inferences is the area of KRR in which data-based
two points based on two 2-D images captured from different answers need to be found without human intervention or
points can be accurately determined as an input. In daylight assistance, and for which data is normally presented in a
conditions and with reasonably good visibility, this input formal system with distinct and clear semantics. Since 1980,
can be used in addition to data acquired with laser and radar it has been assumed that the data involved is a mixture of
equipment in order to increase accuracy – moreover, a simple and complex structures, with the former having a
single camera is sufficient to generate the required data. In low degree of computational complexity and forming the
contrast to 3-D objects, no shape, depth, or orientation basis for research involving large databases. The latter are
information is directly encoded in 2-D images. Depth can be presented in a language with more expressive power, which
encoded in a variety of ways, such as with the use of laser or requires less space for representation, and they correspond
stereo cameras (emulating human vision) and structured to generalizations and fine-grained information.
light approaches (such as Kinect). At present, the most
Decision-making is a type of inference that revolves
intensively pursued research direction involves the use of
primarily around answering questions regarding preferences
superquadrics – geometric shapes defined with formulas,
between activities, for example when an autonomous agent
which use any number of exponents to identify structures
attempts to fulfill a task for a person. Such decisions are
such as cylinders, cubes, and cones with round or sharp
very frequently made in a dynamic domain which changes
edges. This allows a large variety of different basic shapes
over the course of time and when actions are executed. An
to be described with a small set of parameters. If 3-D
example of this is a self-driving car that needs to react to
images are acquired using stereo cameras, statistical
changes in traffic.
methods (such as generating a stereo point cloud) are used
instead of the aforementioned shape-based methods,
because the data quality achieved with stereo cameras is
11
F. Chaumette, S. Hutchinson: Visual Servo Control I: Basic
poorer than that achieved with laser scans. Approaches, IEEE Robotics and Automation Magazine, 13(4): 82-
90, 2006
12
E. D. Dickmanns: Dynamic Vision for Perception and Control of
Motion, London: Springer, 2007
13
T. M. Straat, M. A. Fischler: Context-Based Vision:
Recognizing Objects Using Information from Both 2D and 3D
Imagery, IEEE Transactions on Pattern Analysis and Machine
9
D. P. Huttenlocher, S. Ulman: Recognizing Solid Objects by Intelligence, 13: 1050-65, 1991
14
Alignment with an Image, International Journal of Computer D. Hoiem, A. A. Efros, M. Hebert: Putting Objects in
Vision, 5: 195-212, 1990 Perspective, Proceedings of the IEEE International Conference on
10
K. Frankish, W. M. Ramsey: The Cambridge Handbook of Computer Vision and Pattern Recognition (CVPR), 2137-44, 2006
15
Artificial Intelligence, Cambridge: Cambridge University Press, H. Buxton: Learning and Understanding Dynamic Scene
2014 Activity: A Review, Vision Computing, 21: 125-36, 2003

Logic and combinatorics • Propositional logic
Mathematical logic is the formal basis for many applications • First-order predicate logic
in the real world, including calculation theory, our legal
• Modal logic
system and corresponding arguments, and theoretical
developments and evidence in the field of research and • Non-monotonic logic
development. The initial vision was to represent every type
Automated decision-making, such as that found in
of knowledge in the form of logic and use universal
autonomous robots (vehicles), WWW agents, and
algorithms to make inferences from it, but a number of
communications agents, is also worth mentioning at this
challenges arose – for example, not all types of knowledge
point. This type of decision-making is particularly relevant
can be represented simply. Moreover, compiling the
when it comes to representing expert decision-making
knowledge required for complex applications can become
processes with logic and automating them. Very frequently,
very complex, and it is not easy to learn this type of
this type of decision-making process takes account of the
knowledge in a logical, highly expressive language. 16 In
dynamics of the surroundings, for example when a transport
addition, it is not easy to make inferences with the required
robot in a production plant needs to evade another transport
highly expressive language – in extreme cases, such
robot. However, this is not a basic prerequisite, for example,
scenarios cannot be implemented computationally, even if
if a decision-making process without a clearly defined
the first two challenges are overcome. Currently, there are
direction is undertaken in future, e.g., the decision to rent a
three ongoing debates on this subject, with the first one
warehouse at a specific price at a specific location.
focusing on the argument that logic is unable to represent
Decision-making as a field of research encompasses
many concepts, such as space, analogy, shape, uncertainty,
multiple domains, such as computer science, psychology,
etc., and consequently cannot be included as an active part
economics, and all engineering disciplines. Several
in developing AI to a human level. The counterargument
fundamental questions need to be answered to enable
states that logic is simply one of many tools. At present, the
development of automated decision-making systems:
combination of representative expressiveness, flexibility,
and clarity cannot be achieved with any other method or • Is the domain dynamic to the extent that a sequence
system. The second debate revolves around the argument of decisions is required or static in the sense that a
that logic is too slow for making inferences and will single decision or multiple simultaneous decisions
therefore never play a role in a productive system. The need to be made?
counterargument here is that ways exist to approximate the • Is the domain deterministic, non-deterministic, or
inference process with logic, so processing is drawing close stochastic?
to remaining within the required time limits, and progress is
being made with regard to logical inference. Finally, the • Is the objective to optimize benefits or to achieve a
third debate revolves around the argument that it is goal?
extremely difficult, or even impossible, to develop systems • Is the domain known to its full extent at all times?
based on logical axioms into applications for the real world. Or is it only partially known?
The counterarguments in this debate are primarily based on
Logical decision-making problems are non-stochastic in
the research of individuals currently researching techniques
nature as far as planning and conflicting behavior are
for learning logical axioms from natural-language texts.
concerned. Both require that the available information
In principle, a distinction is made between four different regarding the initial and intermediate states be complete,
types of logic17 which are not discussed any further in this that actions have exclusively deterministic, known effects,
article: and that a specific defined goal exists. These problem types
are often applied in the real world, for example in robot
16
N. Lavarac, S. Dzeroski: Inductive Logic Programming, Vol. 3:
Non-Monotonic Reasoning and Uncertain Reasoning, Oxford
control, logistics, complex behavior in the WWW, and in
University Press: Oxford, 1994 computer and network security.
17
K. Frankish, W. M. Ramsey: The Cambridge Handbook of
Artificial Intelligence, Cambridge: Cambridge University Press, In general, planning problems consist of an initial (known)
2014

situation, a defined goal, and a set of permitted actions or 3.4 Language and communication
transitions between steps. The result of a planning process is
In the field of artificial intelligence, processing language is
a sequence or set of actions that, when executed correctly,
considered to be of fundamental importance, with a
change the executing entity from an initial state to a state
distinction being made here between two fields:
that meets the target conditions. Computationally speaking,
computational linguistics (CL) and natural language
planning is a difficult problem, even if simple problem
processing (NLP). In short, the difference is that CL
specification languages are used. Even when relatively
research focuses on using computers for language
simple problems are involved, the search for a plan cannot
processing purposes, while NLP consists of all applications,
run through all state-space representations, as these are
including machine translation (MT), Q&A, document
exponentially large in the number of states that define the
summarization, information extraction, to name but a few.
domains. Consequently, the aim is to develop efficient
In other words, NLP requires a specific task and is not a
algorithms that represent sub-representations in order to
research discipline per se. NLP comprises:
search through these with the hope of achieving the relevant
goal. Current research is focused on developing new search • Part-of-speech tagging
methods and new representations for actions and states,
• Natural language understanding
which will make planning easier. Particularly when one or
more agents acting against each other are taken into • Natural language generation
account, it is crucial to find a balance between learning and • Automatic summarization
decision-making – exploration for the sake of learning while
decisions are being made can lead to undesirable results. • Named-entity recognition
Many problems in the real world are problems with • Parsing

dynamics of a stochastic nature. One example of this is • Voice recognition
buying a vehicle with features that affect its value, of which
we are unaware. These dependencies influence the buying • Sentiment analysis
decision, so it is necessary to allow risks and uncertainties to • Language, topic, and word segmentation
be considered. For all intents and purposes, stochastic
• Co-reference resolution
domains are more challenging when it comes to making
decisions, but they are also more flexible than deterministic • Discourse analysis
domains with regard to approximations – in other words,
• Machine translation
simplifying practical assumptions makes automated
decision-making possible in practice. A great number of • Word sense disambiguation
problem formulations exist, which can be used to represent
• Morphological segmentation
various aspects and decision-making processes in stochastic
domains, with the best-known being decision networks and • Answers to questions
Markov decision processes. • Relationship extraction
Many applications require a combination of logical (non-
• Sentence splitting
stochastic) and stochastic elements, for example when the
control of robots requires high-level specifications in logic The core vision of AI says that a version of first-order
and low-level representations for a probabilistic sensor predicate logic (“first-order predicate calculus” or “FOPC”)
model. Processing natural languages is another area in supported by the necessary mechanisms for the respective
which this assumption applies, since high-level knowledge problem is sufficient for representing language and
in logic needs to be combined with low-level models of text knowledge. This thesis says that logic can and should
and spoken signals. supply the semantics underlying natural language. Although
attempts to use a form of logical semantics as the key to
representing contents have made progress in the field of AI
and linguistics, they have had little success with regard to a

program that can translate English into formal logic. To previously unknown texts. 18 This works the same way for
date, the field of psychology has also failed to provide proof lightly supervised and unsupervised learning, such as when
that this type of translation into logic corresponds to the way no annotations have been made by humans and the only data
in which people store and manipulate “meaning.” presented is a text in a language with texts with identical
Consequently, the ability to translate a language into FOPC contents in other languages or when relevant clusters are
continues to be an elusive goal. Without a doubt, there are found in thesaurus data without there being a defined goal. 19
NLP applications that need to establish logical inferences With regard to AI and language, information retrieval (IR)
between sentence representations, but if these are only one and information extraction (IE) play a major role and
part of an application, it is not clear that they have anything correlate very strongly with each other. One of the main
to do with the underlying meaning of the corresponding tasks of IR is grouping texts based on their content, whereas
natural language (and consequently with CL/NLP), since the IE extracts similarly factual elements from texts or is used
original task for logical structures was inference. These and to be able to answer questions concerning text contents.
other considerations have crystallized into three different These fields therefore correlate very strongly with each
positions: other, since individual sentences (not only long texts) can
also be regarded as documents. These methods are used, for
• Position 1: Logical inferences are tightly linked to
example, in interactions between users and systems, such as
the meaning of sentences, because knowing their
when a driver asks the on-board computer a question
meaning is equivalent to deriving inferences and
regarding the owner's manual during a journey – once the
logic is the best way to do this.
language input has been converted into text, the question's
• Position 2: A meaning exists outside logic, which semantic content is used as the basis for finding the answer
postulates a number of semantic markers or primes in the manual, and then for extracting the answer and
that are appended to words in order to express their returning it to the driver.
meaning – this is prevalent today in the form of
annotations. 3.5 Agents and actions
• Position 3: In general, the predicates of logic and In traditional AI, people focused primarily on individual,
formal systems only appear to be different from isolated software systems that acted relatively inflexibly to
human language, but their terms are in actuality the predefined rules. However, new technologies and
words as which they appear applications have established a need for artificial entities
that are more flexible, adaptive, and autonomous, and that
The introduction of statistical and AI methods into the field
act as social units in multi-agent systems. In traditional AI
is the latest trend within this context. The general strategy is
(see also "physical symbol system hypothesis" 20 that has
to learn how language is processed – ideally in the way that
been embedded into so-called “deliberative” systems), an
humans do this, although this is not a basic prerequisite. In
action theory that establishes how systems make decisions
terms of ML, this means learning based on extremely large
and act is represented logically in individual systems that
corpora that have been translated manually by humans. This
must execute actions. Based on these rules, the system must
often means that it is necessary to learn (algorithmically)
prove a theorem – the prerequisite here being that the
how annotations are assigned or how part-of-speech
system must receive a description of the world in which it
categories (the classification of words and punctuation
currently finds itself, the desired target state, and a set of
marks in a text into word types) or semantic markers or
actions, together with the prerequisites for executing these
primes are added to corpora, all based on corpora that have
been prepared by humans (and are therefore correct). In the
18
case of supervised learning, and with reference to ML, it is G. Leech, R. Garside, M. Bryant: . CLAWS4: The Tagging of
the British National Corpus. In Proceedings of the 15th
possible to learn potential associations of part-of-speech International Conference on Computational Linguistics (COLING
tags with words that have been annotated by humans in the 94) Kyoto, Japan, pp. 622-628, 1994
19
K. Spärck Jones: Information Retrieval and Artificial
text, so that the algorithms are also able to annotate new, Intelligence, Artificial Intelligence 141: 257-81, 1999
20
A. Newell, H. A. Simon: Computer Science as Empirical
Enquiry: Symbols and Search, Communications of the ACM
19:113-26

actions and a list of the results for each action. It turned out Within this context, agents are located in an
that the computational complexity involved rendered any environment that is described by a set of possible
system with time limits useless even when dealing with states. Every time an agent executes an action, it is
simple problems, which had an enormous impact on “rewarded” with a numerical value that expresses
symbolic AI, resulting in the development of reactive how good or bad the action was. This results in a
architectures. These architectures follow if-then rules that series of states, actions, and rewards, and the agent
translate inputs directly into tasks. Such systems are is compelled to determine a course of action that
extremely simple, although they can solve very complex entails maximization of the reward.
tasks. The problem is that such systems learn procedures
• Social behavior:
rather than declarative knowledge, i.e., they learn attributes
that cannot easily be generalized for similar situations. In an environment where various entities act, it is
Many attempts have been made to combine deliberative and necessary for agents to recognize their adversaries
reactive systems, but it appears that it is necessary to focus and form groups if this is required by a common
either on impractical deliberative systems or on very loosely goal. Agent-oriented systems are used for
developed reactive systems – focusing on both is not personalizing user interfaces, as middleware, and
optimal. in competitions such as the RoboCup. In a scenario
where there are only self-driving cars on roads, the
Principles of the new, agent-centered approach
individual agent’s autonomy is not the only
The agent-oriented approach is characterized by the indispensable component – car2car
following principles: communications, i.e., the exchange of information
between vehicles and acting as a group on this
• Autonomous behavior:
basis, are just as important. Coordination between
“Autonomy” describes the ability of systems to the agents results in an optimized flow of traffic,
make their own decisions and execute tasks on rendering traffic jams and accidents virtually
behalf of the system designer. The goal is to allow impossible (see also section 5.1, “Vehicles as
systems to act autonomously in scenarios where autonomous, adaptive, and social agents & cities as
controlling them directly is difficult. Traditional super-agents”).
software systems execute methods after these
In summary, this agent-oriented approach is accepted within
methods have been called, i.e., they have no
the AI community as the direction of the future.
choice, whereas agents make decisions based on
their beliefs, desires, and intentions (BDI)21. Multi-agent behavior
• Adaptive behavior: Various approaches are being pursued for implementing

multi-agent behavior, with the primary difference being in
Since it is impossible to predict all the situations
the degree of control that designers have over individual
that agents will encounter, these agents must be
agents.22,23,24 A distinction is made here between:
able to act flexibly. They must be able to learn
from and about their environment and adapt • Distributed problem-solving systems (DPS)
accordingly. This task is all the more difficult if not
• Multi-agent systems (MAS)
only nature is a source of uncertainty, but the agent
is also part of a multi-agent system. Only DPS systems allow the designer to control each individual
environments that are not static and self-contained agent in the domain, with the solution to the task being
allow for an effective use of BDI agents – for
example, reinforcement learning can be used to 22
H. Ah. Bond, L. Gasser: Readings in Distributed Artificial
compensate for a lack of knowledge of the world. Intelligence, San Mateo, CA: Morgan Kaufmann, 1988
23
E. H. Durfee: Coordination for Distributed Problem Solvers,
Boston, MA: Kluwer Academic, 1988
21 24
M. Bratman, D. J. Israel, M. E. Pollack: Plans and Resource- G. Weiss: Multiagent Systems: A Modern Approach to
Bounded Practical Reasoning, Computational Intelligence, 4: 156- Distributed Artificial Intelligence, Cambridge, MA: MIT Press,
72, 1988 1999

distributed among multiple agents. In contrast, MAS • Many agents try to solve the same problem
systems have multiple designers, each of whom can only independently of each other by competing with
influence their own agents with no access to the design of each other
any other agent. In this case, the design of the interaction
Reinforcement learning is one of the approaches being used
protocols is extremely important. In DPS systems, agents
in this context.30
jointly attempt to achieve a goal or solve a problem,
whereas, in MAS systems, each agent is individually 4 Data mining and artificial intelligence in the
motivated and wants to achieve its own goal and maximize
automotive industry
its own benefit. The goal of DPS research is to find
collaboration strategies for problem-solving, while At a high level of abstraction, the value chain in the
minimizing the level of communication required for this automotive industry can broadly be described with the
purpose. Meanwhile, MAS research is looking at following subprocesses:
coordinated interaction, i.e., how autonomous agents can be 1. Development
brought to find a common basis for communication and 2. Procurement
undertake consistent actions. 25
Ideally, a world in which 3. Logistics
only self-driving cars use the road would be a DPS world. 4. Production
However, the current competition between OEMs means 5. Marketing
that a MAS world will come into being first. In other words, 6. Sales, after-sales, and retail
communication and negotiation between agents will take 7. Connected customer
center stage (see also Nash equilibrium). Each of these areas already features a significant level of
complexity, so the following description of data mining and
Multi-agent learning
artificial intelligence applications has necessarily been
Multi-agent learning (MAL) has only relatively recently restricted to an overview.
26,27,28,29
been bestowed a certain degree of attention. The key
problems in this area include determining which techniques 4.1 Development
should be used and what exactly “multi-agent learning”
Vehicle development has become a largely virtual process
means. Current ML approaches were developed in order to
that is now the accepted state of the art for all
train individual agents, whereas MAL focuses first and
manufacturers. CAD models and simulations (typically of
foremost on distributed learning. “Distributed” does not
physical processes, such as mechanics, flow, acoustics,
necessarily mean that a neural network is used, in which
vibration, etc., on the basis of finite element models) are
many identical operations run during training and can
used extensively in all stages of the development process.
accordingly be parallelized, but instead that:
The subject of optimization (often with the use of evolution
• A problem is split into subproblems and individual strategies 31 or genetic algorithms and related methods) is
agents learn these subproblems in order to solve usually less well covered, even though it is precisely here in
the main problem using their combined knowledge the development process that it can frequently yield
OR impressive results. Multi-disciplinary optimization, in which
multiple development disciplines (such as occupant safety
and noise, vibration, and harshness (NVH)) are combined
25
K. Frankish, W. M. Ramsey: The Cambridge Handbook of
Artificial Intelligence, Cambridge: Cambridge University Press,
and optimized simultaneously, is still rarely used in many
2014
26
E. Alonso: Multi-Agent Learning, Special Issue of Autonomous
30
Agents and Multi-Agent Systems 15(1), 2007 L. Busoniu, R. Babuska, B. De Schutter: A Comprehensive
27
E. Alonso: M. d’Inverno, D. Kudenko, M. Luck, J. Noble: Survey of Multi-Agent Reinforcement Learning, IEEE
Learning in Multi-Agent Systems, Knowledge Engineering Transactions on Systems, Man, and Cybernetics – Part C:
Review 16: 277-84, 2001 Applications and Reviews 38: 156-72, 2008
28 31
M. Veloso, P. Stone: Multiagent Systems: A Survey from a “Evolution strategies” are a variant of “evolutionary
Machine Learning Perspective, Autonomous Robots 8: 345-83, algorithms,” which has been developed in Germany. They offer
2000 significantly better performance than genetic algorithms for this
29
R. Vohra, M. Wellmann: Foundations of Multi-Agent Learning, type of task. See also Th. Bäck: Evolutionary Algorithms in
Artificial Intelligence 171:363-4, 2007 Theory and Practice, Oxford University Press, NY, 1996.

cases due to supposedly excessive computation time multiple departments, variants, and model series, as is
requirements. However, precisely this approach offers essential for real use of data in the sense of a continuously
enormous potential when it comes to agreeing more quickly learning development organization. The current situation in
and efficiently across the departments involved on a practice is that department-specific simulation data is often
common design that is optimal in terms of the requirements organized in the form of file trees in the respective file
of multiple departments. system within a department, which makes it difficult to
In terms of the analysis and further use of simulation results, access for an evaluation based on machine learning
data mining is already being used frequently to generate co- methods. In addition, simulation data may already be very
called “response surfaces.” In this application, data mining voluminous for an individual simulation (in the range of
methods (the entire spectrum, ranging from linear models to terabytes for the latest CFD simulations), so efficient
Gaussian processes, support vector machines, and random storage solutions are urgently required for machine-
forests) are used in order to learn a nonlinear regression learning-based analyses.
model as an approximation of the representation of the input While simulation and the use of nonlinear regression models
vectors for the simulation based on the relevant (numerical) limited to individual applications have become the standard,
32
simulation results . Since this model needs to have good the opportunities offered by optimizing analytics are rarely
interpolation characteristics, cross-validation methods that being exploited. Particularly with regard to such important
allow the model's prediction quality for new input vectors to issues as multi-disciplinary (as well as cross-departmental)
be estimated are typically used for training the algorithms. machine learning, learning based on historical data (in other
The goal behind this use of supervised learning methods is words, learning from current development projects for
frequently to replace computation-time-consuming future projects), and cross-model learning, there is an
simulations with a fast approximation model that, for enormous and completely untapped potential for increasing
example, represents a specific component and can be used efficiency.
in another application. In addition, this allows time-
consuming adjustment processes to be carried out faster and 4.2 Procurement
with greater transparency during development. The procurement process uses a wide variety of data
One example: It is desirable to be able to immediately concerning suppliers, purchase prices, discounts, delivery
evaluate the forming feasibility33 of geometric variations in reliability, hourly rates, raw material specifications, and
components during the course of an interdepartmental other variables. Consequently, computing KPIs for the
meeting instead of having to run complex simulations and purpose of evaluating and ranking suppliers poses no
wait one or two days for the results. A response surface problem whatsoever today. Data mining methods allow the
model that has been previously trained using simulations available data to be used, for example, to generate forecasts,
can immediately provide a very good approximation of the to identify important supplier characteristics with the
risk of excessive thinning or cracks in this type of meeting, greatest impact on performance criteria, or to predict
which can then be used immediately for evaluating the delivery reliability. In terms of optimizing analytics, the
corresponding geometry. specific parameters that an automotive manufacturer can
These applications are frequently focused on or limited to influence in order to achieve optimum conditions are also
specific development areas, which, among other reasons, is important.
due to the fact that simulation data management, in its role Overall, the finance business area is a very good field for
as a central interface between data generation and data optimizing analytics, because the available data contains
usage and analysis, constitutes a bottleneck. This applies information about the company's main success factors.
especially when simulation data is intended for use across Continuous monitoring 34 is worth a brief mention as an
example, here with reference to controlling. This monitoring
32
For example: Th. Bäck, C. Foussette, P. Krause: Automatische
Metamodellierung von CAE-Simulationsmodellen (Automatic
is based on finance and controlling data, which is
Meta-Modeling of CAE Simulation Models), ATZ –
34
Automobiltechnische Zeitschrift 117(5), 64-69, 2015. This term is used in a great many ways and is actually very
33
For details: http://www.divis- general in nature. We use it here in the sense of continuous
gmbh.de/fileadmin/download/fallbeispiele/140311_Fallbeispiel_B monitoring of company KPIs that are automatically generated and
MW_Machbarkeitsbewertung_Umformsimulation_DE.pdf analyzed on a weekly basis, for example.

continuously prepared and reported. This data can also be 2003 in combination with a forecast of expected vehicle-
used in the sense of predictive analytics in order to specific sales revenue35.
automatically generate forecasts for the upcoming week or In spare parts logistics, i.e., the provision of spare parts and
month. In terms of optimizing analytics, analyses of the key their storage, data-driven forecasts of the number of spare
influencing parameters, together with suggested optimizing parts needing to be held in stock depending on model age
actions, can also be added to the aforementioned forecasts. and model (or sold volumes) are one important potential
These subject areas are more of a vision than a reality at application area for data mining, because it can significantly
present, but they do convey an idea of what could be decrease the storage costs.
possible in the fields of procurement, finance, and As the preceding examples show, data analytics and
controlling. optimization must frequently be coupled with simulations in
the field of logistics, because specific aspects of the logistics
4.3 Logistics chain need to be simulated in order to evaluate and optimize
In the field of logistics, a distinction can be made between scenarios. Another example is the supplier network, which,
procurement logistics, production logistics, distribution when understood in greater depth, can be used to identify
logistics, and spare parts logistics. and avoid critical paths in the logistics chain, if possible.
Procurement logistics considers the process chain extending This is particularly important, as the failure of a supplier to
from the purchasing of goods through to shipment of the make a delivery on the critical path would result in a
material to the receiving warehouse. When it comes to the production stoppage for the automaker. Simulating the
purchasing of goods, a large amount of historical price supplier network not only allows this type of bottleneck to
information is available for data mining purposes, which can be identified, but also countermeasures to be optimized. In
be used to generate price forecasts and, in combination with order to facilitate a simulation that is as detailed and
delivery reliability data, to analyze supplier performance. As accurate as possible, experience has shown that mapping all
for shipment, optimizing analytics can be used to identify subprocesses and interactions between suppliers in detail
and optimize the key cost factors. becomes too complex, as well as nontransparent for the
A similar situation applies to production logistics, which automobile manufacturer, as soon as attempts are made to
deals with planning, controlling, and monitoring internal include Tier 2 and Tier 3 suppliers as well.
transportation, handling, and storage processes. Depending This is why data-driven modeling should be considered as
on the granularity of the available data, it is possible to an alternative. When this approach is used, a model is
identify bottlenecks, optimize stock levels, and minimize the learned from the available data about the supplier network
time required, for example here. (suppliers, products, dates, delivery periods, etc.) and the
Distribution logistics deals with all aspects involved in logistics (stock levels, delivery frequencies, production
transporting products to customers, and can refer to both sequences) by means of data mining methods. The model
new and used vehicles for OEMs. Since the primary can then be used as a forecast model in order, for example,
considerations here are the relevant costs and delivery to predict the effects of a delivery delay for specific parts on
reliability, all the subcomponents of the multimodal supply the production process. Furthermore, the use of optimizing
chain need to be taken into account – from rail to ship and analytics in this case makes it possible to perform a worst-
truck transportation through to subaspects such as the case analysis, i.e., to identify the parts and suppliers that
optimal combination of individual vehicles on a truck. In would bring about production stoppages the fastest if their
terms of used-vehicle logistics, optimizing analytics can be delivery were to be delayed. This example very clearly
used to assign vehicles to individual distribution channels shows that optimization, in the sense of scenario analysis,
(e.g., auctions, Internet) on the basis of a suitable, vehicle- can also be used to determine the worst-case scenario for an
specific resale value forecast in order to maximize total sale automaker (and then to optimize countermeasures in future).
proceeds. GM implemented this approach as long ago as
35
http://www.automotive-fleet.com/news/story/2003/02/gm-
installs-nutechs-vehicle-distribution-system.aspx

4.4 Production (or directly implementing) actions for optimizing the
relevant target variable (quality in this case) even further.
Every sub-step of the production process will benefit from
This implementation of optimizing analytics, with automatic
the consistent use of data mining. It is therefore essential for
modeling and optimization, is technically available,
all manufacturing process parameters to be continuously
although it is more a vision than a reality for most users
recorded and stored. Since the main goal of optimization is
today.
usually to improve quality or reduce the incidence of
The potential applications include forming technology
defects, data concerning the defects that occur and the type
(conventional as well as for new materials), car body
of defect is required, and it must be possible to clearly
manufacture, corrosion protection, painting, drive trains,
assign this data to the process parameters. This approach
and final assembly, and can be adapted to all sub-steps. An
can be used to achieve significant improvements,
integrated analysis of all process steps, including an analysis
particularly in new types of production process – one
of all potential influencing factors and their impact on
example being CFRP36. Other potential optimization areas
overall quality, is also conceivable in future – in this case, it
include energy consumption and the throughput of a
would be necessary to integrate the data from all
production process per time unit. Optimizing analytics can
subprocesses.
be applied both offline and online in this context.
When used in offline applications, the analysis identifies
4.5 Marketing
variables that have a significant influence on the process.
Furthermore, correlations are derived between these The focus in marketing is to reach the end customer as
influencing variables and their targets (quality, etc.) and, if efficiently as possible and to convince people either to
applicable, actions are also derived from this, which can become customers of the company or to remain customers.
improve the targets. Frequently, such analyses focus on a The success of marketing activities can be measured in sales
specific problem or an urgent issue with the process and can figures, whereby it is important to differentiate marketing
deliver a solution very efficiently – however, they are not effects from other effects, such as the general financial
geared towards continuous process optimization. situation of customers. Measuring the success of marketing
Conducting the analyses and interpreting and implementing activities can therefore be a complex endeavor, since
the results consistently requires manual sub-steps that can multivariate influencing factors can be involved.
be carried out by data scientists or statisticians – usually in It would also be ideal if optimizing analytics could always
consultation with the respective process experts. be used in marketing, because optimization goals, such as
In the case of online applications, there is a very significant maximizing return business from a marketing activity,
difference in the fact that the procedure is automated, maximizing sales figures while minimizing the budget
resulting in completely new challenges for data acquisition employed, optimizing the marketing mix, and optimizing
and integration, data pre-processing, modeling, and the order in which things are done, are all vital concerns.
optimization. In these applications, even the provision of Forecast models, such as those for predicting additional
process and quality data needs to be automated, as this sales figures over time as a result of a specific marketing
provides integrated data that can be used as a basis for campaign, are only one part of the required data mining
modeling at any time. This is crucial given that modeling results – multi-criteria decision-making support also plays a
always needs to be performed when changes to the process decisive role in this context.
(including drift) are detected. The resulting forecast models Two excellent examples of the use of data mining in
are then used automatically for optimization purposes and marketing are the issues of churn (customer turnover) and
are capable of, e.g., forecasting the quality and suggesting customer loyalty. In a saturated market, the top priority for
automakers is to prevent loss of custom, i.e., to plan and
implement optimal countermeasures. This requires
36
C. Sorg: Data Mining als Methode zur Industrialisierung und
Qualifizierung neuer Fertigungsprozesse für CFK-Bauteile in
information that is as individualized as possible concerning
automobiler Großserienproduktion (Data Mining as a Method for the customer, the customer segment to which the customer
the Industrialization and Qualification of New Production
belongs, the customer’s satisfaction and experience with
Processes for CFRP Components in Large-Scale Automotive
Production). Dissertation, Technical University of Munich, 2014. their current vehicle, and data concerning competitors, their
Dr. Hut Verlag.

models, and prices. Due to the subjectivity of some of this trends than drawing quantitative conclusions. Nevertheless,
data (e.g., satisfaction surveys, individual satisfaction within the scope of optimization, it is possible to determine
values), individualized churn predictions and optimal whether an action will have a positive or negative impact,
countermeasures (e.g., personalized discounts, refueling or thereby allowing the direction to be determined, in which
cash rewards, incentives based on additional features) are a actions should go.
complex subject that is always relevant.
Since maximum data confidentiality is guaranteed and no 4.6 Sales, after-sales, and retail
personal data is recorded – unless the customer gives their The diversity of potential applications and existing
explicit consent in order to receive offers as individually applications in this area is significant. Since the “human
tailored as possible – such analyses and optimizations are factor,” embodied by the end customer, plays a crucial role
only possible at the level of customer segments that within this context, it is not only necessary to take into
represent the characteristics of an anonymous customer account objective data such as sales figures, individual price
subset. discounts, and dealer campaigns; subjective customer data
Customer loyalty is closely related to this subject, and takes such as customer satisfaction analyses based on surveys or
on board the question of how to retain and optimize, i.e., third-party market studies covering such subjects as brand
increase the loyalty of existing customers. Likewise, the image, breakdown rates, brand loyalty, and many others
topic of “upselling,” i.e., the idea of offering existing may also be required. At the same time, it is often necessary
customers a higher-value vehicle as their next one and being to procure and integrate a variety of data sources, make
successful with this offer, is always associated with this. It them accessible for analysis, and finally analyze them
is obvious that such issues are complex, as they require correctly in terms of the potential subjectivity of the
information about customer segments, marketing evaluations37 – a process that currently depends to a large
campaigns, and correlated sales successes in order to extent on the expertise of the data scientists conducting the
facilitate analysis. However, this data is mostly not analysis.
available, difficult to collect systematically, and The field of sales itself is closely intermeshed with
characterized by varying levels of veracity, i.e., uncertainty marketing. After all, the ultimate objective is to measure the
in the data. success of marketing activities in terms of turnover based on
Similar considerations apply to optimizing the marketing sales figures. A combined analysis of marketing activities
mix, including the issue of trade fair participation. In this (including distribution among individual media, placement
case, data needs to be collected over longer periods of time, frequency, costs of the respective marketing activities, etc.)
so that it can be evaluated and conclusions can be drawn. and sales can be used to optimize market activities in terms
For individual marketing campaigns such as mailing of cost and effectiveness, in which case a portfolio-based
campaigns, evaluating the return business rate with regard to approach is always used. This means that the optimum
the characteristics of the selected target group is a much selection of a portfolio of marketing activities and their
more likely objective of a data analysis and corresponding scheduling – and not just focusing on a single marketing
campaign optimization. activity – is the main priority. Accordingly, the problem
In principle, very promising potential applications for here comes from the field of multi-criteria decision-making
optimizing analytics can also be found in the marketing support, in which decisive breakthroughs have been made in
field. However, the complexity involved in data collection recent years thanks to the use of evolutionary algorithms
and data protection, as well as the partial inaccuracy of data and new, portfolio-based optimization criteria. However,
collected, means that a long-term approach with careful applications in the automotive industry are still restricted to
planning of the data collection strategy is required. The a very limited scope.
issue becomes even more complex if “soft” factors such as
brand image also need to be taken into account in the data
mining process – in this case, all data has a certain level of 37
One example can be found in this article:
uncertainty, and the corresponding analyses (“What are the http://www.enbis.org/activities/events/current/214_ENBIS_12_in_
Ljubljana/programmeitem/1183_Ask_the_Right_Questions__or_A
most important brand image drivers?” “How can the brand
pply_Involved_Statistics__Thoughts_on_the_Analysis_of_Custom
image be improved?”) are more suitable for determining er_Satisfaction_Data.

Similarly, customer feedback, warranty repairs, and 4.7 Connected customer
production are potentially intermeshed as well, since
While this term is not yet established as such at present, it
customer satisfaction can be used to derive soft factors and
does describe a future in which both the customer and their
warranty repairs can be used to derive hard factors, which
vehicle are fully integrated with state-of-the-art information
can then be coupled with vehicle-specific production data
technology. This aspect is closely linked to marketing and
and analyzed. In this way, factors that affect the occurrence
sales issues, such as customer loyalty, personalized user
of quality defects not present or foreseeable at the factory
interfaces, vehicle behavior in general, and other visionary
can be determined. This makes it possible to forecast such
aspects (see also section 5). With a connection to the
quality defects and use optimizing analytics to reduce their
Internet and by using intelligent algorithms, a vehicle can
occurrence. Nevertheless, it is also necessary to combine
react to spoken commands and search for answers that, for
data from completely different areas – production, warranty,
example, can communicate directly with the navigation
and after-sales – in order to make it accessible to the
system and change the destination. Communication between
analysis.
vehicles makes it possible to collect and exchange
In the case of used vehicles, residual value plays a vital role
information on road and traffic conditions, which is much
in a company’s fleet or rental car business, as the
more precise and up-to-date than that which can be obtained
corresponding volumes of tens of thousands of vehicles are
via centralized systems. One example is the formation of
entered into the balance sheet as assets with the
black ice, which is often very localized and temporary, and
corresponding residual value. Today, OEMs typically
which can be detected and communicated in the form of a
transfer this risk to banks or leasing companies, although
warning to other vehicles very easily today.
these companies may in turn be part of the OEM’s corporate
group. Data mining and, above all, predictive analytics can
5 Vision
play a decisive role here in the correct evaluation of assets,
as shown by an American OEM as long as ten years ago38. Vehicle development already makes use of “modular
Nonlinear forecasting models can be used with the systems” that allow components to be used across multiple
company's own sales data to generate individualized, model series. At the same time, development cycles are
equipment-specific residual value forecasts at the vehicle becoming increasingly shorter. Nevertheless, the field of
level, which are much more accurate than the models virtual vehicle development has not yet seen any effective
currently available as a market standard. This also makes it attempts to use machine learning methods in order to
possible to optimize distribution channels – even as far as facilitate automatic learning that extracts both knowledge
geographically assigning used vehicles to individual auction that is built upon other historical knowledge and knowledge
sites at the vehicle level – in such a way as to maximize a that applies to more than one model series so as to assist
company’s overall sales success on a global basis. with future development projects and organizing them more
Considering sales operations in greater detail, it is obvious efficiently. This topic is tightly intermeshed with that of
that knowledge regarding each individual customer's data management, the complexity of data mining in
interests and preferences when buying a vehicle or, in simulation and optimization data, and the difficulty in
future, temporarily using available vehicles, is an important defining a suitable representation of knowledge concerning
factor. The more individualized the knowledge concerning vehicle development aspects. Furthermore, this approach is
the sociodemographic factors for a customer, their buying restricted by the organizational limitations of the vehicle
behavior, or even their clicking behavior on the OEM's development process, which is often still exclusively
website, as well as their driving behavior and individual use oriented towards the model being developed. Moreover, due
of a vehicle, the more accurately it will be possible to to the heterogeneity of data (often numerical data, but also
address their needs and provide them with an optimal offer images and videos, e.g., from flow fields) and the volume of
for a vehicle (suitable model with appropriate equipment data (now in the terabyte range for a single simulation), the
features) and its financing. issue of “data mining in simulation data” is extremely
38
http://www.syntragy.com/doc/q3-05%5B1%5D.pdf

complex and, at best, the object of tentative research roads, and if they are, they are not widespread. This means
approaches at this time39. that the vehicle as an agent cannot communicate with all
New services are becoming possible due to the use of other vehicles, and that vehicles driven by humans adjust
predictive maintenance. Automatically learned knowledge their behavior based on the events in their drivers’ field of
regarding individual driving behavior – i.e., annual, view. Navigation systems offer support by indicating traffic
seasonal, or even monthly mileages, as well as the type of congestion and suggesting alternative routes. However, we
driving – can be used to forecast intervals for required now assume that every vehicle is a fully connected agent,
maintenance work (brake pads, filters, oil, etc.) with great with the two primary goals of:
accuracy. Drivers can use this information to schedule
• Contributing to optimizing the flow of traffic
garage appointments in a timely manner, and the vision of a
vehicle that can schedule garage appointments by itself in • Preventing accidents
coordination with the driver’s calendar – which is accessible In this scenario, agents communicate with each other and
to the on-board computer via appropriate protocols – is negotiate routes with the goal of minimizing total travel
currently more realistic than the often cited refrigerator that time (obvious parameters being, for example, the route
automatically reorders groceries. distance, the possible speed, roadworks, etc.). Unforeseeable
In combination with automatic optimization, the local events are minimized, although not eliminated completely –
authorized repair shop, in its role as a central coordination for example, storm damage would still result in a road being
point where individual vehicle service requests arrive via an blocked. This type of information would then need to be
appropriate telematics interface, can optimally schedule communicated immediately to all vehicles in the relevant
service appointments in real time – keeping workloads as action area, after which a new optimization cycle would be
evenly distributed as possible while taking staff availability required. Moreover, predictive maintenance minimizes
into account, for example. damage to vehicles. Historical data is analyzed and used to
With regard to the vehicle's learning and adaptation predict when a defect would be highly likely to occur, and
abilities, there is virtually limitless potential. Vehicles can the vehicle (the software in the vehicle, i.e., the agent)
identify and classify their drivers' driving behavior – i.e., makes a service appointment without requiring any input
assign them to a specific driver type. Based on this, the from the passenger and then drives itself to the repair shop –
vehicles themselves can make adjustments to systems all made possible with access to the passenger’s calendar. In
ranging from handling to the electronic user interface – in the event of damage making it impossible to continue a
other words, they can offer drivers individualization and journey, this would also be communicated as quickly as
adaptation options that extend far beyond mere equipment possible – either with a “breakdown” broadcast or to a
features. The learned knowledge about the driver can then control center, and a self-driving tow truck would be
be transferred to a new vehicle when one is purchased, immediately available to provide assistance, ideally
ensuring that the driver’s familiar environment is followed by a (likewise self-driving) replacement vehicle. In
immediately available again. other words, vehicles act:
5.1 Vision – Vehicles as autonomous, adaptive, and • Autonomously in the sense that they automatically
social agents & cities as super-agents follow a route to a destination
Research into self-driving cars is here to stay in the • Adaptively in the sense that they can react to
automotive industry, and the “mobile living room” is no unforeseen events, such as road closures and
longer an implausible scenario, but is instead finding a more breakdowns
and more positive response. Today, the focus of • Socially in the sense that they work together to
development is on autonomy, and for good reason: In most achieve the common goals of optimizing the flow
parts of the world, self-driving cars are not permitted on of traffic and preventing accidents (although the
actual situation is naturally more complex and
39
L. Gräning, B. Sendhoff: Shape Mining: A Holistic Data Mining many subgoals need to be defined in order for this
Approach to Engineering Design. Advanced Engineering to be achieved).
Informatics 28(2), 166-185, 2014.

In combination with taxi services that would also have a significant increase in right-hand-side parking light failures
self-driving fleet, it would be possible to use travel data and can be expected for model x, platform C vehicles delivered
information on the past use of taxi services (provided that from July 2015 onwards. In this case, “significant” means
the respective user gives their consent) in order to send taxis that a statistically verifiable upwards trend (increase in
to potential customers at specific locations without the need reported malfunctions) can be extracted based on vehicle
for these customers to actively request the taxis. In a sales between January 2015 and November 2015. By
simplified form, it would also be possible to implement this analyzing fault chains and repair chains, it is possible to
in an anonymized manner, for example, by using data to determine which events will result in a fault or defect or
identify locations where taxis are frequently needed (as which other models are or will be affected. If the error in
identified with the use of clusters; see also section 3.1, production is caused by a production robot, for example,
“Machine learning”) at specific times or for specific events this can be traced back to a hardware fault and/or software
(rush hour, soccer matches, etc.). error or to an incorrect or incomplete configuration. In the
worst-case scenario, it may even be necessary to update the
If roads become digital as well, i.e., if asphalt roads are
control system in order to eliminate the error. In addition, it
replaced with glass and supplemented with OLED
is not possible to update software immediately, because
technology, dynamic changes to traffic management would
work on patches can only begin after the manufacturer has
also be possible. From a materials engineering perspective,
received and reviewed the problem report. Likewise,
this is feasible:
reconfiguring the robot can be a highly complex task due to
• The surface structure of glass can be developed in the degrees of freedom (axes of rotation) that multi-axis
such a way as to be skid resistant, even in the rain. manipulators have at their disposal. In short, making such
• Glass can be designed to be so flexible and sturdy corrections is time-consuming, demanding, and, in all but
that it will not break, even when trucks drive over ideal scenarios, results in subsequent issues.
it. Artificial intelligence (AI) approaches can be used to
optimize this process at several points.
• The waste heat emitted by displays can be used to
heat roads and prevent ice from forming during One of the areas being addressed by AI research is enabling
winter. systems (where the term “system” is a synonym for “a
whole consisting of multiple individual parts,” especially in
In this way, cities themselves can be embedded as agents in
the case of software or combinations of hardware and
the multi-agent environment and help achieve the defined
software, such as those found in industrial robots) to
goals.
automatically extract and interpret knowledge from data,
5.2 Vision – integrated factory optimization although the extent of this is still limited at present.
By using software to analyze customer and repair shop

Knowledge Data
reports and repair data concerning defects occurring in the
field, we can already automatically analyze whether an
increase in defects can be expected for specific vehicle Action
models or installed parts. The goal here is to identify and

avoid potential problems at an early stage, before large-scale
Figure 4 - Data, knowledge, action
recall actions need to be initiated. Causes of defects in the
field can be manifold, including deficient quality of the In contrast to data, knowledge can form the basis for an
parts being used or errors during production, which, action, and the result of an action can be fed back into data,
together with the fact that thousands of vehicles leave which then forms the basis for new knowledge, and so on.
Volkswagen production plants every day, makes it clear that
If an agent with the ability to learn and interpret data is
acting quickly is of utmost importance. Assuming that at
supplied with the results (state of the world before the
present (November 2015), the linguistic analysis of
action, state of the world after the action; see also
customer statements and repair shop reports shows that a
section 3.5) of its own actions or of the actions of other

agents, the agent, provided it has a goal and the freedom to scenario, we, as humans, would be able to visually
adapt as necessary, will attempt to achieve its goal recognize and interpret the difference between robots that
autonomously. Biological agents, such as humans and are working correctly and robots that are not – and the robot
animals, do this intuitively without needing to actively making the mistake should be able to learn in a similar way.
control or monitor the process of transforming data into The difference here is in the type of perception involved –
knowledge. If, for instance, wood in a DIY project splits digital systems can “see” much better than us in such cases.
because we hammered in a nail too hard at an excessively Even though the internal workings of ML methods
acute angle, our brain subconsciously transforms the angle, implemented by means of software are rarely completely
the material's characteristics, and the force of the hammer transparent during the learning process – even for the
blow into knowledge and experience, minimizing the developer of the learning system – due to the stochastic
likelihood of us repeating the same mistake. components and complexity involved, the action itself is
transparent, i.e., not how a system does something, but what
In the previously discussed, specific area of artificial
it does. These signals need to be used in order to initiate the
intelligence referred to as “machine learning,” research is
learning process anew and to adapt the control system of the
focused on emulating such behavior. Using ML to enable
problematic robot. In the aforementioned case, these would
software to learn from data in a specific problem domain
be the manipulator and effector motion signals of a robot
and to infer how to solve new events on the basis of past
that is working correctly, which can be measured and
events opens up a world of new possibilities. ML is nothing
defined with any desired level of accuracy. This does not
new in the field of data analysis, where it has been used for
require any human intervention, as the system's complete
many years now. What is new is the possibility to compute
transparency is ensured by continuously securing and
highly complex models with data volumes in the petabyte
analyzing the data accrued in the production process.
range within a specific time limit. If one thinks of a
Neither is any human analysis required in the identification
production plant as an organism pursuing the objective of
and transmission of defects from the field. Based on
producing defect-free vehicles, it is clear that granting this
linguistic analyses of repair shop and customer reports,
organism access to relevant data would help the organism
together with repair data, we can already swiftly identify
with its own development and improvement, provided, of
which problems are attributable to production. The
course, that this organism has the aforementioned
corresponding delivery of this data to the relevant agents
capabilities.
(the data situation makes it possible to determine exactly
Two stages of development are relevant in this case: which machine needs to correct itself) allows these agents to
Stage 1 – Learning from data and applying experiences learn from the defects and correct themselves.
In order to learn from data, a robot must not just operate Stage 2 – Overcoming the limitations of programming –
according to static programming, it must also be able to use smart factories as individuals
ML methods to work autonomously towards defined What if the production plant needs to learn things for which
learning goals. With regard to any production errors that even the flexibility of one or more ML methods used by
may occur, this means, first and foremost, that the actions individual agents (such as production or handling robots) is
being carried out that result in these errors will have been insufficient? Just like a biological organism, a production
learned, and not programmed based on a flowchart and an plant could act as a separate entity composed of
event diagram. Assume, for example, that the subcomponents, similarly to a human, who can be addressed
aforementioned parking light problem has not only been using natural language, understands context, and is capable
identified, but that its cause can also been traced back to an of interpreting this. The understanding and interpretation of
issue in production, e.g., a robot that is pushing a headlamp context have always been a challenge in the field of AI
into its socket too hard. All that’s now required is to define research. AI theory views context as a shared (or common)
the learning goal for the corrective measure. Let us also interpretation of a situation, with the context of a situation
assume that the production error is not occurring with robots and the context of an entity relative to a situation being
in other production plants, and that left-hand headlamps are relevant here. Contexts relevant to a production plant
being installed correctly in general. In the best-case include everything that is relevant to production when

expressed in natural language or any other way. The potentially be lost, to respond to financial crises, and to
following simplified scenario helps in understanding the quickly interpret the potential impact of catastrophes or
concept: Let us assume that the final design for a car body is political structures. We already do all this today, and what
agreed upon by a committee during a meeting. we need for it is data. We are not interested in the personal
data of individuals, but in what can be derived from many
“We decided on the following body for the Golf 15 facelift.
individual components. For example, by analyzing over
Please build a prototype accordingly based on the Golf 15,”
1,600 indicators, we can predict how certain financial
says Mr. Müller while looking at the 3-D model that seems
indicators for markets will move and respond accordingly or
to be floating in front of everyone at the meeting and can
we can predict, with a high probability of being correct,
only be seen with augmented reality glasses.
which customer groups find models currently in pre-
In this scenario, use of evolutionary algorithms for production development appealing and then derive
simulation is conceivable, limited to the possible marketing actions accordingly. In fact, we can go so far as
combinations that can actually be built. Provided that the to determine fully configured models to suit the tastes of
required computing power is available and the parameters specific customer groups. With this knowledge, we make
involved have been reduced, this can cut simulation times decisions, adjust our production levels, prepare marketing
from several hours to minutes, making dynamic morphing of plans, and propose fully configured models appropriate for
components or component combinations possible during a specific customer groups in the configurator.
meeting.
Preparing a marketing plan sometimes follows a static
Factory : "Based on the model input, I determined that it process (what needs to be done), but how something is done
will take 26 minutes to adjust the programming of my remains variable. As soon as it is possible to explain to
robots. In order to assemble the floor assembly, tools x1, y1 another person how and why something is being done, this
must be replaced with tools x2, y2 on robots x, y. Production information can also be made available to algorithms.
of the prototype will be completed in 6 hours and 37 Breaking it down into an example, we can predict that one
minutes.” of our competitors opening a new production plant in a
Of course, this scenario is greatly simplified, but it should country where we already have manufacturing operations
still show what the future may hold. In order to understand would result in us having to expect a drop in our sales. We
what needs to be done, the production plant must understand can even predict (within a certain fluctuation range) how
what a car body is, what a facelift is, etc. and interpret the large this expected drop in sales would be. In this case, “we”
parameters and output from a simulation in such a way that refers to the algorithms that we have developed for this
they can be converted into production steps. This conversion specific use case. Since this situation occurs more than once
into production steps requires the training of individual ML and requires (virtually) identical input parameters every
components on the robots or the adaptation/enhancement of time, we can use the same algorithms to predict events in
their programs based on the simulation data, so that all steps other countries. This makes it possible to use knowledge
can be carried out, from cutting sheet metal to assembling from past marketing campaigns in order to conduct future
and integrating the (still fictitious) Golf 15 basic variant. campaigns. In short, algorithms would prepare the
And while this encompasses a wide range of methods, marketing plans.
extending from natural language understanding and When provided with a goal, such as maximizing the benefit
language generation through to planning, optimization, and for our customers while taking account of cost-
autonomous model generation, it is by no means mere effectiveness, algorithm sub-classes would be able to take
science fiction. internal data (such as sales figures and configurator data)
and external data (such as stock market trends, financial
5.3 Vision – companies acting autonomously indicators, political structures) to autonomously generate
When planning marketing activities or customer output for a “marketing program” or “GDP program.” If a
requirements, for example, it is imperative for companies of company were allowed to use its resources and act
all types to monitor how sales change over time, to predict autonomously, then it would be able to react autonomously
how markets will develop and which customers will to fluctuations in markets, subsidize vulnerable suppliers,

and much more. The possibilities are wide-ranging, and advances in this field are being published every week. In a
such scenarios can already be technically conceived today. world where AI systems are able to improve themselves
Continuously monitoring stock prices, understanding and continuously and, for example, manage companies more
interpreting news items, and taking into account effectively than humans, what would be left for humans?
demographic changes are only a few of the areas that are Time for expanding one's knowledge, improving society,
relevant and that, among other things, require a combination eradicating hunger, eliminating diseases, and spreading our
of natural language understanding, knowledge-based species beyond our own solar system.40 Some theories say
systems, and the ability to make logical inferences. In the that quantum computers are required in order to develop
field of AI research, language and visual information are powerful AI systems 41 , and only a very careless person
very frequently used as the basis for understanding things, would suggest than an effective quantum computer will be
because we humans also learn and understand a great deal available within the next 10 years. Then again, only a
using language and visual stimuli. careless person would suggest that it will not. Regardless of
this, and as history has taught us time and time again with
6 Conclusions the majority of relevant scientific accomplishments, caution
Artificial intelligence has already found its way into our will also have to be exercised when implementing artificial
daily lives, and is no longer solely the subject of science intelligence – systems capable of making an exponentially
fiction novels. At present, AI is used primarily in the larger number of decisions in extremely short times as
following areas: hardware performance improves can achieve many positive
things, but they can also be misused.
• Analytical data processing
• Domains in which qualified decisions need to be

made quickly on the basis of a large amount of
(often heterogeneous) data
• Monotonous activities that still require constant

alertness
In the field of analytical data processing, the next few years

will see us transition from exclusive use of decision-support
systems to additional use of systems that make decisions on
our behalf. Particularly in the field of data analysis, we are
currently developing individual analytical solutions for
specific problems, although these solutions cannot be used
across different contexts – for example, a solution
developed to detect anomalies in stock price movements
cannot be used to understand the contents of images. This
will remain the case in the future, although AI systems will
integrate individual interacting components and
consequently be able to take care of increasingly complex
tasks that are currently reserved exclusively for humans – a
clear trend that we can already observe today. A system that
not only processes current data regarding stock markets, but
that also follows and analyzes the development of political
structures based on news texts or videos, extracts sentiments
from texts in blogs or social networks, monitors and predicts 40
F. Neukart: Reverse-Engineering the Mind; see
relevant financial indicators, etc. requires the integration of https://www.scribd.com/mobile/doc/195264056/Reverse-
Engineering-the-Mind, 2014
many different subcomponents – getting these to interact 41
F. Neukart: On Quantum Computers and Artificial Neural
and cooperate is the subject of current research, and new Networks, Signal Processing Research 2(1), 2013
View publication stats

Artificial Intelligence and Data Science in The Au

Uploaded by

Copyright:

Available Formats

Artificial Intelligence and Data Science in The Au

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Artificial Intelligence and Data Science in The Au

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Artiﬁcial Intelligence and Data Science in the Automotive Industry

Article · September 2017

Florian Neukart Thomas Bäck

SEE PROFILE SEE PROFILE

DELIVER View project

Process mining for multi-objective online control View project

The user has requested enhancement of the downloaded file.

Martin Hofmann1, Florian Neukart2,3, Thomas Bäck3

Abstract people, opening the door to a world of fundamentally new

JOURNAL TITLE - MONTH YEAR 1

and then implement in an appropriate manner. Here too, the https://en.wikipedia.org/wiki/Cross_Industry_Standard_Process_fo

2 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 3

4 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 5

3.2 Computer vision • Be continuously provided with data via sensors

6 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 7

8 JOURNAL TITLE - MONTH YEAR

Many problems in the real world are problems with • Parsing

JOURNAL TITLE - MONTH YEAR 9

10 JOURNAL TITLE - MONTH YEAR

• Adaptive behavior: Various approaches are being pursued for implementing

JOURNAL TITLE - MONTH YEAR 11

12 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 13

14 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 15

16 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 17

18 JOURNAL TITLE - MONTH YEAR

By using software to analyze customer and repair shop

models or installed parts. The goal here is to identify and

JOURNAL TITLE - MONTH YEAR 19

20 JOURNAL TITLE - MONTH YEAR

JOURNAL TITLE - MONTH YEAR 21

• Domains in which qualified decisions need to be

• Monotonous activities that still require constant

In the field of analytical data processing, the next few years

22 JOURNAL TITLE - MONTH YEAR

View publication stats

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.