FAI

DEFINATIONS
○ it refers of simulation of humans intelligence in
machines that are programmed to think mimic
human actions
○ Encompasses various aspects: problem
solving , language understanding ,
perception , reasoning

NEED
– Arise from complexity of human / modern
problems
– Desire to automate tasks traditionally
performed by humans
– Can handle vast data , recognise patterns and
decision making
– More valuable In healthcare , transport ,
manufacturing to improve efficiency &
accuracy

DIFFERENT APPROACHES OF AI
Symbolic or rule based AI
– representing knowledge in symbolic form
– Use logical forms to manipulate these
symbols
– Ex: emulate decision making ability of
human expert of specific domain
Connectionist Ai
– Its inspired by structure and function of
human brain
 – Artificial neural network(ANN) contains
interconnected nodes (neurones)
organised in layers
– They learn data through training where
connections betn neurons are adjusted
on IO pairs
Evolutionary Algorithms
– Inspired by process of natural selection
an evolutions
– Techniques: genetic algorithm,
evolutionary & genetic programming
– To evolve soon to optimising and search
programs ex: finding best route to vehicle
Fuzzy Logic
– Representation of uncertainty and
imprecision in Ai System
– Unlike classical binary logic : only deals
with true and false statements
– It assigns degrees of truth to
propositions, making it suitable for
reasoning with vague / ambiguous
information

CLASSIFICATION OF AI
Based On Capability :
□ Narrow Or Weak Ai:
▲ Designed to only perform
specific task / set of tasks within
limited domain
□ General or strong Ai
▲ It posses human lvl intelligence
 ▲
across wide range of tasks
▲ It can understand learn and apply
knowledge in diverse contents
□ Super AI
▲ Emotional Intelligence
▲ Desires and beliefs
▲ Consciousness
Based on Functionality :
□ Reactive AI
▲ They are based on current input
w/o memory or learning
capabilities
▲ dont form internal rs of world or
learn past exp
▲ Ex; IBM chess playing computer ,
which can analyse current board
states but doesn’t learn / adapt
from previous games
□ Memory Ai
▲ It learns from historical data and
past exp to make decisions
▲ It uses memory and historical
data to inform their actions but
their learning capabilities are
limited compared to humans
▲ ex: chatgpt, self driving cars
□ Theory of Mind AI
▲ They can understand and
interpret human emotions,
intentions and beliefs
▲ They have ability to perceive and
respond to human behaviour In
 ▲

socially intelligent manner

▲ Achieving it true potential is a s
till active research
□ Self Aware AI
▲ Achieving Self Aware Ai remains
largely theoretical
▲ Available on paper terms only

ADVANTAGES OF AI
. Human error reduction
. Reduce risks
. Faster decision
. Digital assistance

DISADVANTAGES OF AI
. High production cost
. Risk of unemployment
. Ethical issues

END OF UNIT 1

AGENTS AND ENVIROMENT

AGENT:
– It s anything that can be viewed as:
◆ Perceiving its environmental through
sensors
◆ Acting upon that environment through

ENVIRONMENT
 – It refers everything outside the agent that the
agent can interact with or perceive
◆ Includes physical objects , other agent and
relevant information
◆ Stimuli that agent can sense

AGENT: ARCHITECTURE AND PROGRAM

– Its machinery That the agent executes On:
◇ Device with sensors and actuators
◇ Ex: car and camera

AGENT PRORGAM
– Its implementation of agent function
◆ Agent function is map from the perception
sequence (history of all , that an agent has
perceived till date) to an action
SENSORS
– It’s device that detects the change in
environment and sends the information to
other electronic devices
– An agent observes its environment through
sensors

ACTUATORS:
– They are responsible for moving and
controlling a system
– Hydraulic electrical thermal magnetic

RATIONAL AGENT
. Performance Measure:
◇ It evaluates how well the agent is
achieving its goals and can be defined in
 ◇

various way depending on the application

◇ For instance, self driving car : includes
safety , speed and passenger comfort
. Rationality:
◇ Its not about achieving the best possible
outcome
◇ But also acting optimally given the
information and computational resource
available
◇ An agent is considered is rational if it
dose the right thing given what it knows
◇ Its nothing but status of being
reasonable , sensible and having good
sense of judgement
◇ It’s concerned with expected actions and
results depending upon what agent is
perceived . Performing action with aim of
obtaining useful information is imp

TYPES OF AGENTS
. Simple reflex agents
. Model based reflex agents
. Goal based agents
. Utility based agents
. Learning and hierarchical agents

. Simple Reflex and based reflex Agent

– Ex: thermostat is simple reflex agent that
detects current temperature that tears on and
off its heating system accordingly
– Ex: a chess playing agent that not only
 –
considers the current board state but also
stimulates possible future moves to choose
the best course of action

. Goal Based and Utility based agents

– ex: a delivery robot navigating through a
crowded city to reach its destination while
avoiding obstacles and adhering traffic
rules
– ex: Stock trading agent that decides which
stock to buy or sell based on their predicted
returns and associated skills

3. Learning and hierarchical agents

– ex: autonomous vehicle that learns from
driving experiences to improve its
navigations skills and avoid accidents
– Ex: Ai powered virtual assistant that used
hierarchy modules for natural language
understanding , task planning and
execution

NATURE OF ENVIRONMENT

○ Single agent multi agent

○ Fully observable partially observable
○ Static dynamic
○ Complete incomplete
○ Accessible inaccessible
○ Discrete continuous

PROPERTIES OF AI ENVIRONMENT
● Single multi agent: if the environment doesn’t
change while agent is acting then its static
otherwise its dynamic
● Static dynamic agent: : if the environment doesn’t
change while agent is acting then its static
otherwise its dynamic ex: make (static) traffic
flow (dynamic)
● Discrete continuous agent: if there’s a limited
number of distinct , clearly defined , states of the
environment , states of environment , the
environment is discrete , otherwise its continuous
ex: chess (discrete) self driving car (dynamic )
● Accessible inaccessible : if agent sensory
apparatus can have access to complete states of
environment , then the environment is accessible
to that agent ex: well lit room(accessible) cave
(inaccessible )

TURING TEST
– The interrogator tries to determine which the
person is and which the machine is
– The machines tries to find the interrogator into
believing that it is a person
– If machine succeeds then we can conclude
that the machine has the ability to think

THE END OF UNIT 2

Problem Solving
 Basic Steps To Solve Any Problem Statement

. Defining The Problem

. Formulating Search States (Set Of All
Possibility States)
. Search Algorithms
. Evaluating Solutions

Common Problem Solving Technique (Problem

Solving Strategy)

– Brute Force Speech : Examining all

Posibile solutions
– Heuristics: Rule Based
Shortcuts to Quickly Solve, enhancing
Solutions
– Divide And Conquer: Simplifying
Analysis and Resolutions
– Backtracking: Builds Potential
Solutions and Discarding The Wrong
Ones
– Greedy Algorithms: Aiming For
globally Optimal Solutions

Properties Of Search Algorithms

. Completeness : Guarantees
Solutions , Ensuring Reliability
. Optimality: Finds Best
Solution Available, Maximising
 .

Efficiency in Results
. Time Complexity: The Time Required
To Solve algorithm/Program
. Space Complexity: The Space
Required To Solve Algorithm/ Program

Unformed Search Algorithms

– Depth First Search (DFS): It Uses

Stack Data Structure To Remember
Next Vertex To Visit & Explores As Far
As Possible Along Each Branch Before
Backtracking
– Breadth First Search(BFS): Expands
Shallowest Nodes In The Search Tree
list & It Uses Queue Data Structure To
Keep Track Of All Nodes
– Uniform Cost Search(UCS): Expands
The Node At The Lowest Cost.
Guarantees Solution and maybe
Inefficient in Some Scenarios

◆ Depth First Search (DFS):

– Uses Stack (LIFO) To

maintain Current Path During
Transversal
– Memory Efficiency: Req
less Memory Compared To BFS
– Recursion Friendly: Can
Be Efficiency Implemented Using
Recursion , Simplifying Code
 – Completeness and Optimality: Not
Complete In finding Best Solutions,
Lacks Optimality Sceneries

◆ Binary First Search (BFS):

– Uses Queue(FIFO): To
Manage Nodes and exploration
Efficiency
– Complete Algorithm:
Makes It Reliable Search Algorithm
– Optimal For Unwieghted Graphs:
When Nodes Are Of Equivalent
Weight, BFS IS preferred Choice
– High Memory Usage: BFS
Consumes More Memory Compared to
DFS Due To Queue Data Structures
– Slower In Deep Solutions: If
Solution Lies Deep within The Graph,
BFS Takes More Time Affecting Its
performance

◆ Uniform Cost Search

– Expands least Cost Node First:
Similar To Bfs But with Priorities,
Ensuring Optimal Path finding
Solutions
– Utilises a Priority Queue:
Orders Nodes Based On Their path
Cost, Facilitating Efficient Path
 –

Exploration
– Complete And Optimal Solution:
Guarantees Finding The least Cost
path
– Can Be Slow With Memory-Intensive
Graphs: inefficient For Graphs With
Numerous Low Cost Branches,
Leading To Potential Performance
Issues

Informed Search Algorithms

■ Greedy Search: Focus On Immediate

Benefits, Often leading To SubOptimal
Solutions
■ A* Tree Search: Combines Cost and
Heuristic To Find Shortest path
efficiently
■ A* Graph Search: Enhances A* By
Avoiding Cycles, Ensuring Optimal
Solutions

Greedy Search
– Expanding Nodes: Expands The Node
closest the goal, Guided by the
heuristic Function
– Implementation Surgery: Utilises a
Priority Queue That Priortize Nodes
– Advantages: Generally First Then UFS
Due to the heuristic Function
– Limitations: Not Guaranteed To find
 –
the optimal Solution

A*Tree Search
Priority Queue And Implementation:
Expands The Node Based on The Based On
The f(n) Values, Ensuring Efficient
Exploration
Completeness and Optimality:
Guaranteed To Reach a Solution If One
exists
Memory Intensive Nature: May Req
Significant Memory Resource due to node
Expansion and storage, Impacting
Performance on Large Scale
problems
Heuristic Quality Impact: Performance
Heavily relies on the accuracy and quality on
the accuracy nd quality of heuristic
Employed For estimating Cost to the
goal state

A*Graph Search
uses Priority Queue and lInked List: to
Efficiently Track Visited Nodes and Minimise
revisits
Complete and optimality: Gives Optimality
In Giving Solutions and Reduces the risk of
influenceies
Increased Complexity: Higher Complexity
Compared to the previous tree Search Due to
need of the explored nodes and managing
the closed list
Memory Intensive Nature: Can Demand
Higher Memory Due to maintainance of
closed lists and priority queue for transversal

End Of Unit 3