IMPORTANT NOTE TO STUDENTS

These notes are strictly prepared in alignment with the textbook “Artificial Intelligence: Beyond
Classical AI” by Reema Thareja and the prescribed syllabus.
You must first refer to the textbook for every topic.
If you still find any concept difficult to understand, then use these notes as your supporting
reference.
If the topic remains unclear even after that, please come to me directly for further explanation.

MODULE 1 – INTRODUCTION TO
ARTIFICIAL INTELLIGENCE
1.1 What is Artificial Intelligence
Artificial Intelligence (AI) is the scientific and engineering discipline dedicated to building
intelligent machines capable of performing tasks that normally require human reasoning,
perception, learning, and problem-solving. It focuses on designing systems that can perceive
their environment, interpret situations, learn from experiences, and act towards achieving
specific goals with minimal human intervention.

Unlike conventional software systems that follow pre-programmed rules, AI systems

demonstrate the ability to improve automatically through experience and data. This shift
from rule-based processing to self-learning systems marks the essence of artificial
intelligence.

The fundamental idea behind AI is that human thought processes can be replicated by
machines using computational models. AI integrates concepts from computer science,
cognitive psychology, neuroscience, linguistics, mathematics, and engineering. Hence, it is
not limited to algorithms alone but represents a multi-disciplinary attempt to simulate human
intelligence in a digital environment.

AI has evolved to encompass both theoretical and applied domains. The theoretical
dimension involves developing logical and mathematical models of reasoning and learning,
while the applied side deals with implementing these models in real-world systems such as
robots, chatbots, autonomous vehicles, and decision-support systems.

At its core, an AI system consists of:

• Input: Data or sensory information gathered from the environment.

• Processing Mechanism: Algorithms that analyze and interpret data.
• Output: Intelligent responses or actions based on reasoning.

Thus, AI strives to enable computers to think, learn, and act intelligently—not merely to
execute instructions, but to adapt and evolve with changing circumstances.

1.1.1 How Does AI Work?

Artificial Intelligence functions through a systematic combination of data processing, model
learning, and intelligent decision-making. The process can be understood through four
essential components:

(a) Data Acquisition and Preparation

AI systems rely on large quantities of data. The data may be structured (tables, databases) or
unstructured (text, audio, video). Data cleaning, normalization, and transformation are crucial
because poor data quality leads to unreliable models.

For instance, a voice assistant such as Alexa is trained using millions of speech samples,
covering different accents and tones, to ensure it understands diverse speakers.

(b) Learning and Model Building

At this stage, the AI system uses machine learning algorithms to identify patterns within
the data. The algorithms iteratively adjust their internal parameters to minimize prediction
errors. This training process enables the AI to “learn” relationships and make informed
decisions in new situations.

(c) Reasoning and Decision-Making

Once trained, the AI system applies reasoning mechanisms—either logical inference or

probabilistic estimation—to draw conclusions. For example, in a medical AI system, if a
patient’s symptoms resemble known cases of pneumonia, the AI can infer the likelihood of
that diagnosis and suggest possible treatments.

(d) Continuous Feedback and Improvement

A hallmark of AI is its ability to improve with feedback. New data constantly refines the
existing model, leading to better performance over time. This property distinguishes AI from
traditional programming, which remains static once coded.

Hence, AI systems function as adaptive entities: they sense their surroundings, interpret data,
learn from patterns, and respond intelligently.

1.1.2 Advantages and Disadvantages of Artificial

Intelligence
Advantages

1. Automation of Repetitive Tasks:

AI enables industries to automate mundane processes—such as data entry, quality
control, and report generation—saving time and reducing human fatigue.
 2. High Accuracy and Precision:
AI-driven systems like robotic surgery tools or predictive maintenance software
achieve remarkable accuracy since they are not affected by fatigue or emotion.
3. Availability and Consistency:
Unlike humans, AI systems can work 24×7 without breaks. This reliability is essential
for banking systems, online customer support, and industrial monitoring.
4. Data Handling and Analysis:
AI can process terabytes of data rapidly, identifying hidden patterns that humans
might overlook. For example, AI in finance detects fraudulent transactions in real
time.
5. Enhanced Decision-Making:
Decision-support systems powered by AI analyze complex datasets and provide
actionable insights, reducing human bias.
6. Personalization:
AI customizes user experiences, as seen in e-commerce recommendations,
personalized learning systems, or digital marketing.

Disadvantages

1. High Development Cost:

The design, training, and maintenance of AI systems demand massive computational
resources and specialized expertise.
2. Job Displacement:
Automation may reduce employment in repetitive or low-skill tasks, causing social
and economic disruptions.
3. Ethical Concerns:
Questions of accountability, transparency, and privacy arise. If an AI makes an
incorrect decision, determining responsibility can be difficult.
4. Lack of Creativity and Emotion:
AI lacks human qualities such as empathy, moral judgment, and imagination. It
operates on logic and data rather than intuition.
5. Bias in Decision-Making:
Since AI learns from human-generated data, it may inherit the biases present in that
data—leading to unfair outcomes in hiring, credit approval, or law enforcement.
6. Dependency:
Over-reliance on AI systems can erode human analytical skills, as people may trust
algorithmic outputs without critical evaluation.

1.2 History of Artificial Intelligence

The development of AI has been marked by cycles of optimism, innovation, setbacks, and
resurgence. Its roots can be traced to both philosophical thought and computational
innovation.

The Early Foundations (1940–1950s)

The conceptual seeds of AI were planted when scientists like Alan Turing proposed that
machines could simulate any human computation. His Turing Test (1950) remains a
fundamental benchmark: if a machine can converse indistinguishably from a human, it may
be considered intelligent.

Early experiments in symbolic logic and mechanical reasoning inspired computer scientists to
believe that machines could replicate aspects of human thought.

Birth of AI as a Discipline (1956)

The term Artificial Intelligence was formally introduced at the Dartmouth Conference,
organized by John McCarthy, Marvin Minsky, Allen Newell, and Herbert Simon. This
event marked AI’s recognition as an independent field of study.

The Formative Decades (1960s–1970s)

AI researchers focused on symbolic reasoning—creating systems that represented knowledge

in the form of logic rules. Programs such as ELIZA (which simulated a psychotherapist) and
SHRDLU (which understood natural language commands in a virtual world) showcased early
natural language understanding.

However, due to limited computing power and data, AI progress was slow. This period also
saw the first wave of AI Winter—a time when interest and funding declined due to unrealistic
expectations.

The Rise of Expert Systems (1980s)

The 1980s witnessed the emergence of expert systems, which encoded human expertise into
rule-based frameworks. MYCIN, for example, assisted doctors in diagnosing bacterial
infections. Such systems demonstrated practical applications of AI in medicine, business, and
engineering.

Machine Learning Revolution (1990s–2000s)

With the rise of machine learning, AI moved from symbolic reasoning to data-driven
approaches. Algorithms could now learn automatically from examples. IBM’s Deep Blue
defeating Garry Kasparov in 1997 symbolized the power of computational intelligence.

Modern AI and the Deep Learning Era (2010–Present)

The combination of massive datasets, cloud computing, and neural networks triggered
explosive AI growth. AI systems today power voice assistants, image recognition,
autonomous vehicles, and generative tools like ChatGPT and DALL·E. The focus has shifted
toward explainability, ethics, and human-AI collaboration.

1.3 Types of Artificial Intelligence

AI can be classified based on capabilities and functionalities.

1.3.1 Weak AI (Narrow AI)

Weak AI is designed for a specific function. It excels in the particular domain for which it
was programmed but cannot adapt beyond it. For instance, Google Translate cannot play
chess, and a chess AI cannot diagnose diseases.
It operates under fixed parameters without consciousness or understanding.
Examples: Voice assistants (Siri, Alexa), spam filters, and recommendation engines.

1.3.2 Strong AI (General AI)

Strong AI, or Artificial General Intelligence (AGI), refers to systems capable of

understanding, reasoning, and learning across diverse domains—similar to human
intelligence. Such systems could perform any intellectual task a person can do.
Although theoretical, research in neural architectures and cognitive modeling continues to
move in this direction.

1.3.3 Reactive Machines

These are the most fundamental AI systems that respond only to current inputs. They have no
memory and cannot use past experiences.
Example: IBM’s Deep Blue could evaluate thousands of chess positions per second but had
no concept of learning or improvement.

1.3.4 Limited Memory

This category represents most AI systems today. They can analyze past data to improve
performance. For example, self-driving cars use sensory data from previous trips to make
driving decisions.

1.3.5 Theory of Mind

Still in the research phase, Theory of Mind AI would possess the ability to understand human
emotions, beliefs, and intentions. This advancement would allow AI to engage in genuinely
empathetic interactions.

1.3.6 Self-Aware AI

The ultimate stage of AI development—machines possessing consciousness, awareness, and

a sense of self. While this remains speculative, it represents the philosophical boundary of AI
research.

1.4 Is Artificial Intelligence the Same as Augmented

Intelligence and Cognitive Computing?
Although all three share the goal of enhancing human capability, their approaches differ
fundamentally:

• Artificial Intelligence emphasizes machine autonomy. It aims to replicate human

intelligence so machines can operate without human guidance.
• Augmented Intelligence focuses on collaboration between humans and machines. It
helps humans make better decisions rather than replacing them. Example: AI systems
 assisting doctors in surgery.
• Cognitive Computing aims to mimic the brain’s functioning using data mining,
natural language processing, and pattern recognition. IBM’s Watson exemplifies this
paradigm by understanding natural language and reasoning through evidence-based
responses.

Thus, while Artificial Intelligence aspires to replace human effort, Augmented Intelligence
seeks to enhance it, and Cognitive Computing attempts to simulate it.

1.5 Machine Learning and Deep Learning

Machine Learning (ML) is a subfield of AI that enables computers to learn automatically
through experience without explicit programming. It revolves around creating algorithms that
can recognize patterns and improve performance over time.

ML involves three major types:

1. Supervised Learning: The algorithm learns from labeled data. (e.g., spam detection)
2. Unsupervised Learning: The system identifies patterns in unlabeled data. (e.g.,
clustering customers)
3. Reinforcement Learning: The system learns by interacting with the environment and
receiving feedback. (e.g., game-playing AI)

Deep Learning (DL), a subset of ML, utilizes Artificial Neural Networks (ANNs) with
multiple layers to model complex non-linear relationships. DL systems mimic the structure of
the human brain and are capable of understanding images, audio, and natural language.

Applications include:

• Speech recognition (Google Voice)

• Image classification (facial recognition)
• Autonomous driving (object detection)
• Language translation (Neural Machine Translation)

Deep Learning represents the core of today’s AI revolution, transforming how machines
interpret and engage with the world.

3.1 Defining Intelligence

The word intelligence originates from the Latin intelligere, meaning “to understand.”
In psychology, intelligence is the mental ability to learn from experience, reason logically,
plan, adapt to new environments, comprehend complex ideas, and solve problems.
In artificial systems, intelligence refers to the capability of a machine to imitate or
simulate these human cognitive processes.

Reema Thareja defines intelligence as “the ability of an entity to perceive its environment,
process information, and act in a goal-directed manner.”
Hence, intelligence is not merely problem solving; it involves perception, reasoning,
learning, memory, and adaptation.
Human intelligence arises from biological neural activity; machine intelligence results from
algorithmic computation.
The goal of AI is not to copy the human brain physically but to replicate the functions that
make human thinking effective.

3.2 Elements and Components of Intelligence

According to the book, any intelligent entity — whether human or artificial — exhibits the
following components:

1. Perception – The process of collecting data from the environment through sensory
mechanisms (eyes, ears, sensors, cameras, microphones).
Machines perceive through sensors; for instance, a robot uses ultrasonic sensors to
detect obstacles.
2. Learning – The acquisition of knowledge or skills through study, experience, or
teaching.
In humans, this occurs via repetition and memory; in machines, through algorithms
such as supervised and unsupervised learning.
Learning forms the foundation of adaptability.
3. Reasoning – The ability to derive conclusions, inferences, or judgments from existing
knowledge.
Machine reasoning often follows logical inference rules such as modus ponens (if A
→ B and A is true, then B is true).
Example: An expert system diagnosing diseases reasons through symptom–disease
relationships.
4. Problem Solving – The act of determining a sequence of actions that transforms an
initial state into a desired goal state.
AI uses algorithms (search, constraint satisfaction, optimization) to achieve this
transformation efficiently.
5. Memory – Storage and retrieval of information.
Machines maintain memory in structured databases or knowledge bases.
Effective memory enables re-use of previous experiences.
6. Adaptation – Adjustment of behaviour based on changes in the environment.
Adaptive systems recalibrate parameters when they detect new conditions; e.g.,
thermostats adjusting temperature automatically.
7. Language Understanding and Communication – The ability to comprehend and
express information symbolically.
Natural-Language Processing (NLP) allows machines to interpret human language for
dialogue systems, translation, or summarization.

Together these components constitute machine intelligence, enabling systems to act

purposefully and learn continuously.

3.3 Human Intelligence vs Machine Intelligence

Although machine intelligence strives to mirror human cognition, they differ fundamentally.

Aspect Human Intelligence Machine Intelligence

Origin Biological, evolved through natural Artificial, engineered through
 selection algorithms
Learning Experience-driven, context-aware, Data-driven, pattern-based,
Mechanism intuitive statistical
Memory Associative and adaptive; prone to Perfect recall; deterministic
forgetting retrieval
Creativity Can generate novel ideas, Limited creativity; works
imagination, and emotion within trained data
Adaptability Can generalize with minimal data Requires retraining with new
data
Error Handling Uses judgment and intuition Relies on mathematical
correction
Energy Human brain ≈ 20 W Machines require high
Efficiency computational power
Ethics & Driven by morality and empathy Lacks consciousness and moral
Emotion sense

While machines excel in speed, accuracy, and data volume, humans surpass in contextual
understanding, empathy, and moral reasoning.
The most effective systems therefore combine both — an approach often termed Augmented
Intelligence.

3.4 Agent and Environment

3.4.1 Definition of Agent

An agent is an entity that perceives its environment through sensors and acts upon that
environment through actuators to achieve specific objectives.
Formally, an agent implements an agent function that maps a sequence of percepts (inputs)
to an action:

where P* is the set of all percept sequences and A is the set of possible actions.

Example:
A self-driving car (agent) perceives its surroundings through cameras, radar, and LiDAR
(sensors) and acts through steering, acceleration, and braking (actuators).

3.4.2 Environment

The environment is everything external to the agent that it interacts with.

It provides the conditions, obstacles, and stimuli that affect the agent’s actions.

Characteristics of Environments (from Thareja Ch 3.4):

1. Observable vs Partially Observable:

 Fully observable if the agent’s sensors access the complete environment state; e.g.,
chess.
Partially observable when only limited information is available; e.g., driving in fog.
2. Deterministic vs Stochastic:
Deterministic if each action leads to a predictable result; stochastic if outcomes
involve randomness.
3. Static vs Dynamic:
Static when the environment does not change while the agent is deliberating; dynamic
when it can alter spontaneously.
4. Discrete vs Continuous:
Discrete environments have a finite number of states and actions (board games), while
continuous environments (robot motion) involve infinite gradations.
5. Single Agent vs Multi-Agent:
In single-agent systems, only one agent acts (solitaire); in multi-agent systems,
several agents coexist or compete (traffic systems, markets).

3.4.3 Types of Agents

Reema Thareja distinguishes four classical categories:

1. Simple Reflex Agents: Act solely on the current percept; implement condition-action
rules (“if traffic light is red → stop”).
No memory of past states.
2. Model-Based Reflex Agents: Maintain an internal model representing aspects of the
world not currently observed, enabling reasoning in partially observable
environments.
3. Goal-Based Agents: Incorporate explicit goals and choose actions that bring them
closer to those goals, using search or planning algorithms.
4. Utility-Based Agents: Beyond goals, they evaluate outcomes in terms of a utility
function measuring performance or satisfaction, selecting actions that maximize
expected utility.

3.5 Problem Solving and Search

Every intelligent action can be formulated as a search problem.
An agent searches through a space of possible states to find a sequence of actions
transforming the initial state into a goal state.

A state space is the set of all possible configurations reachable from the initial state.
A solution is a path through this state space satisfying the goal condition.

The process involves:

1. Defining the Problem: initial state, goal state, actions, transition model, and cost.
2. Searching the Space: exploring possible sequences systematically.
3. Evaluating Paths: selecting the most efficient route based on cost or heuristic.
4. Executing the Plan: applying the derived sequence of actions.

3.6 Uninformed Search Algorithms

Uninformed (or blind) search methods have no domain-specific knowledge; they explore
purely based on the problem definition.

3.6.1 Breadth-First Search (BFS)

• Expands nodes level by level from the root.

• Uses a queue (FIFO) data structure.
• Guaranteed to find the shortest path if all step costs are equal.
• Time and space complexity: O(b^d) where b = branching factor, d = depth.
Example: Finding the shortest route between two cities.

3.6.2 Depth-First Search (DFS)

• Explores as deep as possible along each branch before backtracking.

• Uses a stack (LIFO).
• Memory efficient but may get trapped in infinite paths.
• Completeness: not guaranteed if infinite depth; optimality: not guaranteed.

3.6.3 Uniform-Cost Search (UCS)

• Expands the node with the lowest cumulative path cost g(n).
• Uses a priority queue ordered by path cost.
• Finds the optimal (least-cost) solution for positive costs.
Example: Dijkstra’s shortest-path algorithm.

Uninformed methods are exhaustive and computationally expensive; hence, heuristic

guidance is introduced through informed search.

3.7 Informed (Heuristic) Search Algorithms

Informed search uses additional knowledge — heuristics — to guide exploration toward the
goal efficiently.

A heuristic function h(n) estimates the cost from node n to the goal.
The smaller the estimated cost, the more promising the node.

3.7.1 Pure Heuristic Search

Also called Greedy Search, it selects the node with the minimum heuristic h(n)**, ignoring
the path cost already incurred.
It “looks ahead” optimistically, assuming the best-appearing node leads directly to the goal.

Advantages:

• Faster than blind search.

Disadvantages:
• Not optimal; may overlook cheaper paths because it ignores accumulated cost.
Example: A navigation system choosing the nearest visible route rather than the
shortest actual route.
3.7.2 Best-First Search (Greedy Algorithm)

A general form that combines aspects of both BFS and DFS.

It uses an evaluation function:

The algorithm repeatedly selects the node with the lowest f(n).
Implementation uses a priority queue ordered by h(n).

Procedure (from Thareja Ch 3.7.2):

1. Start with the initial node in the OPEN list.

2. Choose the node n with the smallest h(n).
3. Expand n and generate successors.
4. Insert successors into OPEN according to h(n).
5. Continue until the goal node is selected.

Example:
Suppose a map-based agent must reach city G from city A.
Each node represents a city; h(n) = straight-line distance to G.
The algorithm always chooses the city appearing geographically closest to G until the
destination is reached.

While efficient for small problems, it can become sub-optimal when heuristic estimates are
inaccurate.

4 KNOWLEDGE REPRESENTATION
(KR)
4.1 Introduction
Knowledge Representation is the method of encoding information about the world in a
format that a computer can utilize to solve complex tasks such as reasoning,
understanding, and decision making.

Thareja emphasizes that data alone is not knowledge.

Knowledge arises when data is organized, interpreted, and related to existing facts.
AI systems must therefore represent knowledge in structures that are both computationally
efficient and semantically meaningful.

4.2 Role of Knowledge in AI

AI applications depend heavily on knowledge for:
 • Diagnosing faults or diseases.
• Planning and scheduling.
• Understanding natural language.
• Interpreting sensory information.

The main objectives of knowledge representation are:

1. Expressiveness: ability to describe all relevant aspects of the domain.

2. Efficiency: ease of storage and retrieval.
3. Reasoning Support: facilitating logical inference and learning.

4.3 Knowledge-Based Agent

A knowledge-based agent stores knowledge about its environment in a knowledge base (KB)
— a collection of facts and rules expressed in a formal representation language such as
propositional logic or predicate logic.

Architecture:

1. Knowledge Base: contains facts (e.g., “Rain → Wet Ground”).

2. Inference Engine: derives new facts using logical rules.
3. Percept Sequence: current input from environment.
4. Action Selector: decides the next action based on derived knowledge.

Operation Cycle:

• Tell: insert new percepts into KB.

• Ask: query the KB for information.
• Act: perform an action decided by the inference engine.

Example: In a medical-diagnosis system, symptoms are told to the KB, it infers possible
diseases, and suggests appropriate actions.

4.4 Types of Knowledge

Reema Thareja classifies knowledge into several distinct categories:

1. Declarative Knowledge:
Facts and assertions about objects or relationships.
Example – “Paris is the capital of France.”
2. Procedural Knowledge:
Describes how to perform actions or procedures.
Example – “Steps for solving a linear equation.”
3. Heuristic Knowledge:
Rule-of-thumb strategies based on experience rather than strict logic.
Example – “In chess, control the center of the board early.”
4. Meta-Knowledge:
Knowledge about the use and structure of other knowledge — knowing how to reason
with knowledge.
 Used in expert systems for adaptive reasoning.
5. Structural Knowledge:
Describes the relationships among concepts, forming semantic networks or frames.
Example – hierarchies in ontology: Animal → Mammal → Dog.

A comprehensive AI system integrates all these forms, enabling reasoning at multiple

abstraction levels.