Webinar Series on Artificial Intelligence and Robotics

AI Fundamentals

Dr. N GANESH M.E.[CSE]., MBA[HR]., M.L.[IPR]., Ph.D[CSE]

Associate Professor – Senior,
School of Computer Science and Engineering and International Relations Coordinator
Vellore Institute of Technology,
Chennai – 600 127
Mail: ganesh.narayanan@vit.ac.in
Mobile: 9841737667
 Artificial Intelligence is...
“… the science and engineering of making intelligent machines” … “[where]
intelligence is the computational part of the ability to achieve goals in the world” (original
definition by John McCarthy who coined the term ‘Artificial Intelligence’ in 1955)

“… the science of making machines do things that would require intelligence if done
by men” (definition offered by A.I. pioneer Marvin Minsky in 1968)

“… the science of making machines smart” (Demis Hassabis, CEO and founder of
DeepMind, now part of Google)
“… anything that makes machines act more intelligently” (IBM‘s definition)

“… a constellation of technologies that extend human capabilities by sensing,

comprehending, acting and learning – allowing people to do much more” (Accenture‘s
definition)
“… getting computers to do tasks that would normally require human intelligence” (
Deloitte’s definition)
“… a computerized system that exhibits behavior that is commonly thought of as
requiring intelligence” US Government definition (NSTC)
• Augmented Intelligence : AI should not attempt to replace human experts but rather
extends human capability and complex tasks
• Make use of Lots of unstructured data available in internet that subject experts need
with evidences and make informed decisions
• Experts scale their capability and let the machine do consuming works
 Based on the Strength, Breadth, and application AI is categorized into:

Narrow or Weak AI: AI with a narrow Super AI or Conscious AI: AI

range of abilities String AI or generalized AI : AI on that surpasses human
par with human capabilities intelligence
Google Search
virtual assistants like Siri, Alexa, Interact and operate on a wide The hypothetical AI that
Cortana variety of independent and doesn’t just mimic or
Facial recognition software unrelated tasks. understand human intelligence
Disease mapping and prediction tools Can learn new tasks to solve new and behaviour; It is where
Manufacturing and drone robots problems by teaching itself new machines become self-aware
Email spam filters strategies. Perform on human and surpass the capacity of
social media monitoring tools for Level Intelligence human intelligence and ability.
dangerous content
Recommendations based on
watch/listen/purchase behaviour
Is AI dangerous? Will robots take over the world?
Self-driving cars AI could be programmed to do something
Doesn’t learn new ones devastating.
AI could be programmed to do something beneficial but
Making decisions based on training develop a destructive method for achieving its goal.
data
 Some incidents of AI in 2023
Privacy: Vacuum Cleaner Robot Took Private Photo Of humans using restrooms That’s
Leaked On Facebook.

Security & Safety: Tesla that hit fire truck in deadly I-680 crash in Walnut Creek was on
autopilot, company says
AI Chatbot Allegedly Pushed Belgian Man To Take His Own Life.
Man crushed to death by robot that confused him for box of vegetables

Intellectual Property: Artist refuses Sony photography award after admitting he used AI

Fairness & Bias: Asian MIT grad asks AI to make her photo more ‘professional,’ gets
turned into white woman
By 2030, 16 Trillion dollars of GDP will
be added by the AI, not only in S/W
domain but virtually every industry and
AI
every aspect of our life

Application
s
AI means different things to different people
AI in Healthcare AI in Data Security
Helping Doctors See Inside the Body Better Anamoly Detection
Detecting Health Problems Early Predicting Threats
Developing Medications Quickly and Cost-Effectively Automated Safety Response
Personalized Treatment Plans
Managing Hospital Functions and Resources AI in Social Media
Smart Suggestions
AI in Gaming Virtual Assistants and Chatbots
Smart Game Characters Sentiment Analysis
Creating Game Worlds with AI Trend Analysis:
Making Games Look and Feel Real:
AI in Travel & Transport
AI in Finance
Optimization of Route
Identifying and Prevention of Fraud Smart Security Screening
Automated Trading Chatbots for Travel Support
Risk Control AI Prevents Breakdowns
AI in Automotive Industry AI in Agriculture
Self-Driving Cars Crop Observation and Control
Advanced Driver Assistance Systems Smart Farming for Efficiency
(ADAS) Automated Farming
Streamlining Production Processes
Voice Recognition:
AI in Robotics AI in E-commerce
Self-Moving Robots Personalized Product Suggestions
Object Recognition and Manipulation Managing Inventory
Collaboration of Humans and Robots Dynamic Pricing

AI in Entertainment AI in Education
Recommendation of Content Education Content Creation
AI as a Creative Assistant Virtual Learning Assistants
Live Event and Performance Enhancements Automated Assessment and Instant Feedback
Customized Learning Routes
 History of AI
• 1950: Allen Turing published “Computer Machinery and
Intelligence” later becomes Test for Machine Intelligence
• 1952: Arthur Samuel developed a program to play checkers,
which is the first to ever learn the game independently.
• 1958: John McCarthy created LISP, the first programming
language for AI research, which is still in popular use to this
day.
• 1961: The first industrial robot Unimate started working on
an assembly line at General Motors in New Jersey
• 1966: Joseph Weizenbaum created the first “chatterbot”
ELIZA, a mock psychotherapist, that used natural language
processing (NLP) to converse with humans
• 1966: Shakey – general purpose mobile robot
 History of AI
AI Winter1 1974-1980:
Overhyped Expectations
Lack of Computational Power
Failure of machine translation
Negative results in Neural Nets
A boom of AI (1980-1987) : AI came back with "Expert System“
• Type of AI program designed to mimic the decision-making abilities of a human expert in a
specific domain
• Uses a combination of rules and data to solve complex problems by reasoning through
bodies of knowledge, primarily represented in the form of "if-then" rules.
AI Winter2 1987-1993:
Expert Systems were limited in scope and often failed to adapt to real-world complexities
The market for expert systems became saturated
Financial Losses

AI winters refer to periods of time when public interest, funding, and

progress in the field of AI significantly decline
History of AI
• 1997: IBM Deep Blue beats world chess champion, Gary Kasparov, and became the first computer
to beat a world chess champion.
• 1999: For two days in May, NASA runs an AI program called autonomously ran Deep Space 1
• 2002: AI entered the home in the form of Roomba, a vacuum cleaner.
• 2004: AI planner to take decisions on board in Mars rover of NASA with an option to turn of planner
when needed
• 2005: A 2nd DARPA Grand Challenge of 212KM with self driving cars. Stanley car won 1st prize led
by Sebastian thrun from Stanford followed by Standstorm and Highlander led by Red Whittaker of
Carnegie Mellon University.
• 2011: IBM's Watson won jeopardy, a quiz show, where it had to solve the complex questions as
well as riddles
• 2014: Chatbot "Eugene Goostman" won a competition in the infamous "Turing test.“
• 2016: Alpha Go of Deepmind which is acquired by google in 2015 beat Lee Sedol in a game
called Go a 19X19 board game
• 2018: The "Project Debater" from IBM debated on complex topics with two master debaters and
also performed extremely well.
• 2018: Google has demonstrated an AI program "Duplex" which was a virtual assistant, and which
had taken hairdresser appointment on call, and lady on other side didn't notice that she was talking
with the machine.
Machine learning
• A broad subset of artificial intelligence that enables computers to learn from data and
experience without being explicitly programmed
• The most common type of ML are regression and classification algorithms
• Supervised and unsupervised
• Has several subsets of AI within it, including neural networks, deep learning, and
reinforcement learning.

Deep Learning
• Deep learning networks are composed of layers of interconnected processing nodes,
or neurons.
• Input layer, intermediate layers the output layer
• The depth of a network is important because it allows the network to learn complex
patterns in the data.
• well-suited for tasks such as image recognition and natural language processing
Robotics
• Robotics systems are a type of AI system that are used to control physical objects in
the world.
• Built with both supervised learning and unsupervised learning.
• Industrial robotics system
• Service robotics system
• Military robotics system

Natural Language Processing

• NLP deals with the understanding and manipulation of human language

• Used in a variety of applications, such as text classification, sentiment analysis, and
machine translation
• Can also be used to create chatbots and personal assistants
• NLP is also used in search engines
Computer Vision
• Teaching computers to see and interpret visual information from images or videos.
• Involves tasks such as object recognition, image segmentation, facial recognition, and
scene understanding.

Expert Systems
• Replicate the decision-making capability of human experts in specific domains
• Uses knowledge and rules programmed into the system to provide expert-level advice
or recommendations.

Planning
• Involves the development of strategies or action sequences to achieve specific goals
• Addresses the challenge of how to enable machines to make decisions and perform
tasks that require a sequence of actions, often in complex and dynamic environments
• Typically involves: Initial State, Goal State, Actions, Transition Model, Constraints
 Thought processes and reasoning
Humans Like

Rationality
Behavior
Thinking humanly: The cognitive modelling approach

• Where researchers aim to understand and replicate human

thought processes.
• Cognitive architectures and models inspired by psychology and
neuroscience are used to create AI systems that simulate human-like
mental processes.
• Cognitive modelling typically involves the following steps:
• Task Analysis
• Develop a representation of the knowledge and information involved
in the task.
• Design algorithms or mechanisms that simulate the inferential
processes humans use to derive conclusions from available
information.
• Execution and Evaluation
• Learning
Thinking rationally: The
“laws of thought” approach
• What are the correct arguments/thought process?
• Making logical inferences and decisions based on
explicit rules and reasoning
• It involves representing knowledge in a formal,
logical manner and using deductive reasoning to
derive conclusions.
• The Greek philosopher Aristotle was one of the first to
attempt to codify “right thinking,” that is, irrefutable
reasoning processes
• There are two main obstacles to this approach.
• Not all intelligent behaviour is mediated by logical
deliberation (reflexes-hand on heated vessel)
• What is the purpose of thinking (looking at the
 Acting humanly: The Turing
Test approach
• Replicating not just cognitive processes but also behavioural aspects such as language
use, facial expressions, and social interactions
• Programming a computer to pass a rigorously applied test need to possess the following
capabilities:
• NATURAL LANGUAGE PROCESSING, KNOWLEDGE REPRESENTATION,
AUTOMATED REASONING, MACHINE LEARNING
• TOTAL TURING TEST: The machine is evaluated not only on its ability to generate
human-like text responses but also on its capabilities in areas such as Vision, Hearing,
Touch, and other Modalities like taste or smell,
• To pass the total Turing Test, the computer will need:
• COMPUTER VISION
• ROBOTICS
 Acting rationally: The rational
agent approach
• Rational behaviour : doing the right thing
• Making decisions and taking actions that lead to optimal
outcomes based on the available information
• Doesn't necessarily mimic human thought processes but
aims for effective problem-solving and goal achievement.
• Game-playing AI, optimization algorithms, and decision-
making systems often strive to act rationally by
considering all possible actions and selecting the one
that maximizes expected utility or achieves specified
goals
Intelligent Agents

• Agents and environments

• Rationality
• PEAS (Performance measure,
Environment, Actuators, Sensors)
• Environment types
• Agent types

Artificial Intelligence a modern approach 22

Agents
An agent is anything that can be viewed as perceiving its environment
through sensors and acting upon that environment through actuators
• Percept Sequence
• Human agent:
– eyes, ears, and other organs for sensors;
– hands, legs, mouth, and other body parts for actuators
• Robotic agent:
– cameras and infrared range finders for sensors
– various motors for actuators
• Physically Grounded Agents
• Intelligent Buildings, Autonomous aircraft
• Softbots
• Expert systems
• IBM Watson
Artificial Intelligence a modern approach 23
 Vacuum-cleaner world

• Percepts: location and contents, e.g., [A,Dirty]

• Actions: Left, Right, Suck, NoOp
• Agent’s function  look-up table
• For many agents this is a very large table

Artificial Intelligence a modern approach 25

Rationality
–The performance measure that defines the criterion
of success.
–The agent’s prior knowledge of the environment.
–The actions that the agent can perform.
–The agent’s percept sequence to date.

26
 Ideal Rational agent:
For each possible percept sequence, does whatever action
is expected to maximize its performance measure on the
basis of evidence so far and built in knowledge
• Ideal rational agent need not to be omniscience(need not to
know everything about future) but able to learn and exhibits
sufficient autonomy
• Acting in order to obtain information
• Bounded Rationality
• Given our state of knowledge
• Chose an optimal action
• Given limited computational resource
 Is vacuum cleaner a rational agent?
Let us assume the following:
• The performance measure awards one point for each clean
square at each time step, over a “lifetime” of 1000 time
steps
• The “geography” of the environment is known a priori but
not the dirt distribution
• The only available actions are Left , Right, and Suck
• The agent correctly perceives its location and whether that
location contains dirt

28
 PEAS : Part
Picking Robot

• Performance measure:
Percentage of parts in correct bins
• Environment: Conveyor belt with
parts, bins
• Actuators: Jointed arm and hand
• Sensors: Camera, joint angle
sensors
Artificial Intelligence a modern approach 29
PEAS : Interactive
English tutor

• Performance measure: Maximize

student's score on test
• Environment: Set of students
• Actuators: Screen display
(exercises, suggestions,
corrections)
• Sensors: Keyboard
30
 PEAS: Performance measure,
Environment, Actuators, Sensors

Pencil Quality check

Automatic parking
system
PEAS
Environment types
• Fully observable (vs. partially observable)
• Deterministic (vs. stochastic)
• Episodic (vs. sequential)
• Static (vs. dynamic)
• Discrete (vs. continuous)
• Single agent (vs. multiagent):

Artificial Intelligence a modern

approach 33
 Fully observable (vs.
partially observable)

• Is everything an agent requires to choose its actions available to it via its sensors?
Perfect or Full information.
• If so, the environment is fully accessible
• If not, parts of the environment are inaccessible
• Agent must make informed guesses about world.

Cross Word Poker Backgammon Taxi driver Part picking robot Image analysis
Fully Partially Fully Partially Fully Fully

34
 Deterministic (vs. stochastic)
• If the next state of the environment is completely determined by the
current state and the action executed by the agent
• Non-deterministic environments
• Have aspects beyond the control of the agent
• Utility functions have to guess at changes in world
Cross Word Poker Backgammon Taxi driver Part
Part picking robot Image analysis
Deterministic Stochastic Stochastic Stochastic Stochastic Deterministic

Artificial Intelligence a modern approach 35

Episodic (vs. sequential):
• Is the choice of current action
• Dependent on previous actions?
• If not, then the environment is episodic
• In non-episodic environments:
• Agent has to plan ahead:
• Current choice will affect future actions

Cross Word Poker Backgammon Taxi driver Part picking robot Image analysis
Sequential Sequential Sequential Sequential Episodic Episodic

Artificial Intelligence a modern approach 36

Static (vs. dynamic):
• Static environments don’t change
• While the agent is deliberating over what to do
• Dynamic environments do change
• So agent should/could consult the world when choosing actions
• Alternatively: anticipate the change during deliberation OR make decision very
fast
• Semidynamic: If the environment itself does not change with the
passage of time but the agent's performance score does.
Cross Word Poker Backgammon Taxi driver Part picking robot Image analysis
Static Static Static Dynamic Dynamic Semi

Artificial Intelligence a modern approach 37

 Discrete (vs. continuous)
Discrete Environments:
• State Space: the set of possible states is finite. Each state is distinct and countable.
• Action Space: The set of possible actions is also finite and well-defined.
Continuous Environments:
• State Space: The set of possible states is infinite, often uncountably so. States are
typically represented by real numbers.
• Action Space: The set of possible actions is also continuous and may involve real-
valued parameters.

Cross Word Poker Backgammon Taxi driver Part picking robot Image analysis
Discrete Discrete Discrete Conti Conti Conti

38
Single agent (vs. multiagent):
• An agent operating by itself in an environment or there are many
agents working together

Cross Word Poker Backgammon Taxi driver Part picking robot Image analysis
Single Multi Multi Multi Single Single

Artificial Intelligence a modern approach 39

Summary.
Observable Deterministic Episodic Static Discrete Agents

Cross Word Fully Deterministic Sequential Static Discrete Single

Poker Fully Stochastic Sequential Static Discrete Multi

Taxi driver Partially Multi

Stochastic Sequential Dynamic Conti

Part picking robot Partially Stochastic Episodic Dynamic Conti Single

Image analysis Fully Deterministic Episodic Semi Conti Single

Artificial Intelligence a modern approach

Agent types
• Types in order of increasing generality:
• Simple reflex agents
• Reflex agents with state/model
• Goal-based agents
• Utility-based agents
• Learning agents

41
Simple reflex agents
• Simple but very limited intelligence.
• Action does not depend on percept history, only on
current percept.
• Environment Should be fully observable
• Infinite loops
• Suppose vacuum cleaner does not observe
location. What do you do given location = clean? Thermostat
Left of A or right on B -> infinite loop. Automatic Door
Light Switch
• Fly buzzing around window or light. Smoke Detector
• Chess – openings, endings
• Lookup table (not a good idea in general)
• 35100 entries required for the entire game
Artificial Intelligence a modern approach 42
Model-based  Works in Partially Observable

reflex agents environments

 Internal State: Maintain a
representation of the environment’s
state over time and helps the agent
understand the current situation
and remember important details that
are not directly observable at every
moment.
 Can make more informed decisions and
exhibit behaviour that is not solely
dependent on the immediate percept.
 Updates its internal model to reflect
the changes
Autonomous in the(Roomba)
Vacuum Cleaner environment

43
• Designed to achieve specific
Goal-based
objectives or goals within a
given environment
agents
• More sophisticated in their
approach and exhibit a level
of planning and reasoning
• It evaluates how far the
current state is from the
desired goal state
• Considers possible actions
and selects those that are
likely to bring it closer to
achieving its goals
• Mechanisms to assess the Chess-Playing Agent
success or failure of their
actions. Route Planning Agent
Automated Warehouse Robot
• Dynamic Goal Adjustment
44
• Goals are not always enough
• Many action sequences get taxi to Utility-based agents
destination
• Consider other things. How fast, how
safe…..
• A utility function maps a state onto a real
number which describes the associated degree
of “happiness”, “goodness”, “success”
• When there are conflicting goals, only some of
which can be achieved (for example, speed and
safety), the utility function specifies the
appropriate tradeoff.
• When there are several goals that the agent can
aim for, none of which can be achieved with Financial Trading Agent
certainty, utility provides a way in which the
likelihood of success can be weighed against the
importance of the goals
45
Learning Learning allows the agent to operate in
initially unknown environments and to
agents become more competent than its initial
knowledge.
Performance element is what was
previously the whole agent
Learning element
 Modifies performance element.
Critic: how the agent is doing
 Problem generator
 Suggests actions to create new and
Spam Filter informative experiences for the
Personal Assistant learning element to improve its
performance.
Artificial Intelligence a modern approach 46
 Interactive Voice Response (IVR) System
Suitable Type: Simple Reflex Agent
Why:
• It reacts to user inputs with pre-defined responses, following simple
condition-action rules.
Why Not Others:
• Model-Based Reflex: Doesn’t maintain an internal state or model of
the world.
• Goal-Based: Doesn't pursue long-term goals, just immediate
responses.
• Utility-Based: Doesn’t calculate utilities.
• Learning: Basic IVR systems don’t learn from interactions.
Home Cleaning Robot

Suitable Type: Model-Based Reflex Agent

Why:
• It needs to keep track of the layout of the room and its own location
to clean efficiently, using internal state and a model of the world.
Why Not Others:
• Simple Reflex: Needs more than simple reactions to obstacles.
• Goal-Based: Doesn't have a specific end goal, just ongoing tasks.
• Utility-Based: Doesn't calculate and maximize utility, just follows a
cleaning path.
• Learning: Some may learn, but basic operation doesn’t require
learning.
Self-Driving Car
Suitable Type: Utility-Based Agent
Why:
• A self-driving car must make decisions based on multiple criteria like safety, speed,
fuel efficiency, and passenger comfort. These criteria can be balanced using a
utility function to maximize overall driving performance.
Why Not Others:
• Simple Reflex: Too simple; can’t handle complex driving scenarios.
• Model-Based Reflex: While it uses a model, it needs more sophisticated decision-
making based on utility.
• Goal-Based: It doesn’t just aim to reach a destination; it also needs to optimize the
journey.
• Learning: Learning is a part of it, but its primary function is making real-time
decisions based on current utility.
Route Planning Agent
Suitable Type: Goal-Based Agent
Why:
• It aims to find the shortest or fastest path to a destination, focusing on
achieving a specific goal.
Why Not Others:
• Simple Reflex: Needs complex planning beyond simple reactions.
• Model-Based Reflex: Requires goal-oriented planning, not just reacting to
current state.
• Utility-Based: Primarily goal-driven (reach destination), not balancing
multiple utilities.
• Learning: Can enhance with learning, but goal-oriented by nature.
Spam Filter
Suitable Type: Learning Agent
Why:
• It improves over time by learning to better classify emails as spam or not
based on user feedback.
Why Not Others:
• Simple Reflex: Needs to adapt and learn from new data.
• Model-Based Reflex: Requires ongoing learning to improve.
• Goal-Based: Doesn’t have a single end goal, continuously learns.
• Utility-Based: Focuses on learning rather than maximizing a utility
function.
 Learning agents(Taxi driver)
• Performance element
• How it currently drives
• Taxi driver Makes quick left turn across 3 lanes
• Critics observe shocking language by passenger and other drivers and informs bad action
• Learning element tries to modify performance elements for future
• Problem generator suggests experiment out something called Brakes on different Road
conditions
• Exploration vs. Exploitation
• Learning experience can be costly in the short run
• shocking language from other drivers
• Less tip
• Fewer passengers

Artificial Intelligence a modern approach 52

Summary
• History of AI
• Basics of AI
• Few Applications of AI
• Agents can be described by their PEAS.
• Environments can be described by several key properties
• A rational agent maximizes the performance measure for their
PEAS.
• The performance measure depends on the agent function.
• The agent program implements the agent function.

Artificial Intelligence a modern approach 53

 Thank You
&
Signing Off

Dr. N GANESH M.E.[CSE]., MBA[HR]., M.L.[IPR]., Ph.D[CSE]

Recognized as Top 2% Scientists in the World by Elsevier and Stanford University, USA
Associate Professor – Senior,
School of Computer Science and Engineering and International Relations Coordinator,
Vellore Institute of Technology,
Chennai – 600 127
Mail: ganesh.narayanan@vit.ac.in
Mobile: 9841737667