Unit 1: Introduction to Artificial

Intelligence

Course Target: [Link] Engineering Curriculum (Maharashtra)

This document provides a comprehensive, deep-dive exploration into the foundational

concepts of Artificial Intelligence as outlined in the Unit 1 syllabus.

1. Introduction: What is AI?

Artificial Intelligence (AI) is a multidisciplinary branch of computer science focused on

building smart machines capable of performing tasks that typically require human
intelligence. In a formal engineering context, defining AI is usually categorized into four
distinct quadrants based on whether the system's core goal is to emulate human thought/
behavior or to achieve ideal rational thought/behavior:

• Thinking Humanly (Cognitive Modeling): Attempting to understand and replicate

the inner workings of the human mind. This involves building computational models
based on psychological and neurological evidence.

• Acting Humanly (The Turing Test): Proposed by Alan Turing in 1950, this defines
intelligence operationally. If an AI can communicate with a human evaluator so
convincingly that the evaluator cannot tell it is a machine, it passes the test. It
requires capabilities in Natural Language Processing, Knowledge Representation,
Automated Reasoning, and Machine Learning.

• Thinking Rationally (Laws of Thought): Based on formal logic (syllogisms). The goal
is to build systems that can solve any solvable problem using logical deduction.

• Acting Rationally (Rational Agent): An agent is something that acts in an

environment. A rational agent acts to maximize its expected performance measure
given its percepts. This is the most prevalent view in modern AI engineering.
 2. History & Applications

A Brief History

The evolution of AI has seen periods of massive enthusiasm followed by severe funding
cuts, known as "AI Winters."

• The Gestation (1943-1955): Warren McCulloch and Walter Pitts proposed the first
model of artificial neurons. Alan Turing published his seminal paper on machine
intelligence.

• The Birth (1956): The term "Artificial Intelligence" was officially coined by John
McCarthy at the Dartmouth Summer Research Project, which brought together
pioneers like Marvin Minsky, Allen Newell, and Herbert Simon.

• The AI Winters (1970s & 1980s): Progress stalled due to the limited computational
power of the era and the realization that many AI problems suffered from
combinatorial explosion.

• Modern Resurgence (1990s-Present): Fueled by Moore's Law (vastly increased

computational power via GPUs) and the explosion of digital data, probabilistic
methods and Machine/Deep Learning have revolutionized the field.

Key Engineering Applications

• Autonomous Systems: Utilizing computer vision, sensor fusion, and reinforcement

learning for self-driving cars (e.g., Tesla, Waymo) and industrial robotics.

• Natural Language Processing (NLP): Large Language Models (LLMs), machine

translation, and speech recognition engines.

• Expert Systems in Healthcare: Diagnostic AI analyzing radiological scans (MRI, X-

Rays) with accuracy matching or exceeding human experts.

• Optimization & Finance: High-frequency algorithmic trading, dynamic supply chain

optimization, and real-time fraud detection.
 3. Artificial Intelligence as Representation & Search

In classic symbolic AI, almost all problem-solving is reduced to two fundamental

concepts: Representation and Search. Intelligent behavior emerges from representing
the world properly and searching through that representation efficiently.

• Representation: Before an AI can solve a problem, it needs a formal model of the

problem domain. This includes representing objects, relations, rules, and the current
state of the environment. If the representation is poor, the problem becomes
unsolvable.

• Search: Once the problem space is represented, finding a solution means navigating
(searching) through this space from an initial state to a goal state. Because real-world
problems have massive state spaces, AI focuses heavily on informed (heuristic) search
algorithms that can find solutions without checking every single possibility.

4. Production System

A Production System is a type of cognitive architecture used to build rule-based Expert

Systems. It provides a structured way to formalize and apply knowledge.

The Three Core Components of a Production System:

1. Global Database (Working Memory): The central data structure that contains
the current state of the problem and all known factual information.

2. Set of Production Rules: A library of rules structured as IF [Condition]

THEN [Action]. The condition is matched against the database, and the action
specifies how the database should be updated.

3. Control System (Inference Engine): The program that orchestrates the process.
It executes a "Recognize-Act Cycle":
- Match: Find all rules whose conditions are satisfied by the current database.
- Conflict Resolution: If multiple rules match, apply a strategy to pick the best
one.
- Act: Execute the chosen rule's action, updating the database.
 5. Basics of Problem Solving: Problem Representation
Paradigms

To write an algorithm for a problem, an AI engineer must first map the real-world
problem into a mathematical paradigm. Common paradigms include:

• State-Space Paradigm: Treating the problem as a graph where nodes are states and
edges are actions. Finding a solution is finding a path through the graph.

• Problem Reduction (AND/OR Graphs): Breaking a massive problem into a set of

smaller, independent sub-problems that can be solved recursively.

• Constraint Satisfaction Problems (CSP): Rather than finding a path, the goal is to
find a single state that satisfies a strict list of mathematical or logical constraints (e.g.,
scheduling classes in a university without overlapping rooms).

6. Defining Problem as a State Space Representation

This is the most critical formulation for search algorithms. Formally defining a problem
as a state space requires specifying four components:

• Initial State: The starting configuration (e.g., the pieces on a chessboard before the
first move).

• Actions (Operators): A transition model that defines what moves are legally possible
from any given state, and what the resulting new state will be.

• Goal Test: A function that evaluates a given state to determine if it is a target solution.
This can be an explicit state or a property a state must possess.

• Path Cost: A function that assigns a numeric cost to the path taken to reach a state. In
engineering, we generally want to find the optimal path (the one with the lowest cost,
such as the shortest distance or least energy used).
 7. Characteristics of AI Problems

AI problems fundamentally differ from standard software engineering problems (like

sorting an array or querying a database). Their defining characteristics include:

• Combinatorial Explosion: AI problem spaces grow exponentially. For instance, the

game of Go has more possible board states than there are atoms in the observable
universe. Standard algorithms fail here.

• Reliance on Heuristics: Because exhaustive search is physically impossible for large

state spaces, AI relies on heuristics—intelligent "rules of thumb" that guide the search
toward the goal, sacrificing guaranteed optimality for computational feasibility.

• Uncertain and Dynamic Environments: Real-world AI (like robotics) operates with

incomplete information and in environments that change constantly.

• Lack of Explicit Algorithms: In AI, we often know what the goal is (e.g., "translate
this sentence"), but we do not have a hardcoded, step-by-step mathematical algorithm
to achieve it. Instead, we give the system the ability to search for the answer or learn
it from data.