Unit - I

Introduction
 Contents
• The AI Problems
• The Underlying Assumption
• What is an AI Technique?
• The Level of The Model
• Criteria for Success
• Agents and Environments
• Concept of Rationality
• The Nature of Environments
• Structure of Agents
• Problem Formulation
 Introduction
• Artificial Intelligence (AI) is a branch of computer science focused
on creating systems that can perform tasks that typically require
human intelligence. These tasks include:
• Reasoning (e.g., solving puzzles, making decisions)
• Learning (e.g., recognizing patterns, improving from data)
• Perception (e.g., seeing, hearing, or interpreting sensory input)
• Natural Language Understanding (e.g., chatbots, translation)
• Problem-Solving (e.g., optimizing routes, playing games)
• AI is transforming industries by mimicking human-like intelligence
in machines. While it offers immense benefits, responsible
development is crucial to address risks.
 Continues…..
• AI is the science and engineering of making intelligent machines,
especially intelligent computer programs.
• John McCarthy is the father of AI (1956).
• AI is a branch of computer science dealing with the simulation of
intelligent behaviour in computers.
• AI is the study of how to make computers do things which, at the
moment, people do better.
• AI is the study and design of intelligent agents, where an intelligent
agent is a system that perceives its environment and takes actions.
 How AI Works?
• AI systems typically:
• Take Input (e.g., text, images, sensor data).
• Process It (using algorithms like machine learning).
• Produce Output (e.g., predictions, actions, decisions).
• Example:
• Input: A photo of a cat.
• AI Process: A neural network analyzes pixels to detect patterns (ears,
whiskers).
• Output: Correctly labels the image as "cat."
 Real World AI Examples
Application AI Technique Used Example
Voice Assistants NLP (Natural Language Alexa, Siri, Google
Processing) Assistant
Fraud Detection Machine Learning Banks flagging suspicious
transactions
Medical Diagnosis Deep Learning AI detecting tumors in X-
rays
Self-Driving Cars Computer Vision + Tesla Autopilot
Reinforcement Learning
Recommendation Systems Collaborative Filtering YouTube, Amazon
suggestions
 Subfields of AI
• Machine Learning – Learning from data (e.g. spam detection)
• Natural Language Processing (NLP) – Understanding and
generating human language (e.g. chatbots)
• Computer Vision – Interpreting visual data (e.g. face recognition)
• Robotics – Interacting physically with the environment (e.g. drones)
• Expert Systems – Using knowledge bases to make decisions (e.g.
medical diagnosis tools).
 The AI Problems
• AI problems involve tasks that require some form of intelligent
behaviour, such as learning from experience, making decisions, and
adapting to new situations.
• They also include tasks requiring human-like intelligence (e.g.,
reasoning, learning, perception).
• Examples:
• Playing Chess (Decision-making under rules).
• Siri/Alexa (Natural Language Processing).
• Self-driving cars (Perception + Real-time control).
 Types of AI Problems
Type of Problem Description Example
Finding a solution path in a space of
Search Problems Solving a maze, route planning
possible options

Game Playing Competing with an opponent using strategy Chess, Go

Making sequences of decisions to achieve a

Planning Robot delivery planning
goal
Knowledge Representation Storing facts and rules about the world Medical expert systems
Reasoning and Inference Drawing conclusions from known facts Legal advice systems
Improving performance from data or
Learning Problems Predicting stock prices
feedback
Understanding sensory data (e.g. vision,
Perception Problems Facial recognition
sound)
Understanding or generating human
Natural Language Problems Translation, sentiment analysis
language
 Characteristics of AI Problems
• High complexity (many possible states or actions)
• Incomplete or uncertain information
• Dynamic environments (conditions change over time)
• Need for adaptation or learning
 Example: A Self-Driving Car
• AI in a self-driving car must solve:
• Vision (detect lanes, pedestrians)
• Decision-making (when to stop, accelerate)
• Prediction (anticipating other drivers)
• Navigation (planning optimal routes)
• Learning (improving from data over time)
 The Underlying Assumption
• Intelligence can be modeled or replicated computationally.
• Human thought processes can be formalized as algorithms.
• Physical symbol systems (e.g., computers) can exhibit intelligence
(Symbolic AI).
• Intelligence emerges from data-driven learning (Modern ML).
• Example:
• AlphaGo assumes Go strategy can be learned via data + algorithms.
 What is an AI Technique?
• AI techniques are specific methods used to enable machines to
simulate intelligent behaviour.
• They are methods to solve AI problems.
• These techniques fall under different areas of AI, including machine
learning, knowledge representation and reasoning.
• Examples:
• Search Algorithms: Google Maps uses A* for shortest paths.
• Machine Learning: Netflix recommendations (collaborative
filtering).
• Rule-Based Systems: Medical diagnosis bots (e.g., IBM Watson).
 Main AI Techniques
Technique Description
Learning patterns from data automatically (e.g., supervised,
Machine Learning (ML)
unsupervised, reinforcement learning).
A subset of ML using neural networks with many layers for tasks like
Deep Learning
vision and speech.
Natural Language Processing
Enables machines to understand and generate human language.
(NLP)
Expert Systems Use of rules and logic to mimic decision-making of a human expert.
Search Algorithms Used in planning and problem-solving (e.g., A*, BFS, DFS).
Fuzzy Logic Handles reasoning with uncertain or imprecise information.
Genetic Algorithms Optimization based on principles of natural selection.
Knowledge Representation Storing facts and rules about the world in a form AI systems can use.
Computer Vision Analyzing and interpreting visual information from the world.
 The Level of The Model
• The "level" of an AI model usually refers to the degree of abstraction,
complexity, or capability at which the AI operates. They are like layers in a
pyramid.
1. Knowledge Level
• Focus: What the system knows and what goals it tries to achieve.
• Example: An expert system "knows" medical facts and reasons like a doctor.
2. Algorithmic Level
• Focus: How the system processes knowledge to make decisions or learn.
• Example: A neural network’s learning algorithm, or decision trees for
classification.
3. Implementation Level
• Focus: Physical or software implementation of the algorithm (e.g., Python code,
GPU acceleration).
• Example: Using TensorFlow to implement a deep learning model.
 Criteria for Success
• How to measure AI performance?
• Performance: Accuracy, speed, efficiency.
• Generalization: Works on unseen data.
• Robustness: Handles noise/uncertainty.
• Human Alignment: Ethical, interpretable.
• Examples:
• Accuracy: 95% correct cancer detections (Medical AI).
• Speed: Alexa responds in <1 second.
• Safety: Self-driving car avoids collisions 99.9% of the time.
 Agents and Environments
• An agent is anything that perceives its environment through sensors and
acts upon that environment using actuators.
• The environment is everything outside the agent that the agent interacts
with.
• The agent-environment relationship forms the basis of intelligent
behaviour in AI systems.
• In general,
• Agent: Entity that perceives/acts (e.g., robot, software bot).
• Environment: Context the agent operates in (e.g., physical world, virtual
space).
• Interaction: Sensors → Percepts → Actions → Environment feedback.
 Key Characteristics of an Agent
• Autonomous: Can operate without human intervention.
• Goal-driven: Designed to achieve a set of objectives.
• Reactive & Proactive: Responds to changes and can plan ahead.
• Rational: Chooses actions that maximize expected performance.
 Examples

Scenario Agent Environment

Road, traffic,
Self-driving car Car control software
pedestrians
Chess game AI chess engine Chess board
Human voice,
Voice assistant Alexa/Siri
queries, online info
Room layout, dirt,
Robotic vacuum Robot
obstacles
 Concept of Rationality
• An agent is rational if it selects optimal actions to achieve goals given available
knowledge.
• A rational AI agent is not necessarily "intelligent" in a human sense—it simply
follows logical rules to achieve its goals efficiently.
• The key is designing the right performance measure and ensuring the agent acts
optimally within its constraints.
• A rational agent selects actions that maximize its chances of success, based on
its performance measure.
• In AI, rationality refers to the ability of an agent to make the best possible
decision given:
a) Its knowledge (what it knows about the environment).
b) Its goals (what it is trying to achieve).
c) Available actions (what it can do).
 The Nature of Environments
• The nature of the environment determines how an AI agent perceives,
learns, and acts.
• Environments are classified based on different characteristics:
• Fully Observable: The agent can see the complete state of the
environment at all times.
• Example: Chess (the board is fully visible).
• Partially Observable: The agent has limited or noisy information.
• Example: Poker (opponent cards are hidden).
• Deterministic: The next state is predictable based on actions.
• Example: A robot moving in a grid (each move has a fixed outcome).
 Continues…..
• Stochastic (Probabilistic): Outcomes are uncertain.
• Example: Self-driving cars (traffic, weather, and pedestrians behave
unpredictably).
• Episodic: Each action is independent; past actions don’t affect future ones.
• Example: Spam filtering (each email is processed separately).
• Sequential: Current actions affect future decisions.
• Example: Autonomous drones (each movement impacts the next).
• Static: The environment doesn’t change while the agent is deciding.
• Example: Crossword puzzle (the puzzle stays the same while you solve it).
 Continues…..
• Dynamic: The environment changes while the agent is thinking.
• Example: Stock trading (prices fluctuate in real-time).
• Discrete: States and actions are finite and separate.
• Example: Tic-tac-toe (limited moves).
• Continuous: States and actions are infinite and smooth.
• Example: Drone flight control (infinite possible speeds and angles).
• Single-Agent: Only one AI is acting.
• Example: Solitaire (no opponent).
• Multi-Agent: Multiple agents interact (cooperatively or competitively).
• Example: Video game NPCs (AI enemies) or robot soccer teams.
 Examples of Different Environments
Environment Type Example Why?
Fully Observable + All pieces are visible; moves
Chess
Deterministic have fixed outcomes.
Partially Observable + Opponent cards are hidden;
Poker
Stochastic bluffing adds randomness.
Self-Driving Traffic changes in real-time;
Dynamic + Continuous
Cars steering is a smooth motion.
Each email is processed
Email Spam
Episodic + Static independently; spam rules don’t
Filtering
change mid-task.
 Structure of Agents
• An agent in AI is an entity that perceives its environment through
sensors and acts upon it using actuators.
• The structure of an agent defines how it processes inputs (percepts)
to generate actions.
• Every AI agent has:
• Sensors – Detect changes in the environment (e.g., cameras,
microphones, GPS).
• Actuators – Carry out actions (e.g., motors, speakers, robotic arms).
• Decision-Making Mechanism – The "brain" that maps perceptions
to actions.
 Types of Agent Architectures
• Simple Reflex Agents- Uses predefined condition-action rules (if-
then).
• Model-Based Reflex Agents- Maintains an internal model of the
world to handle partial observability.
• Goal-Based Agents- Selects actions to achieve a specific goal.
• Utility-Based Agents- Maximizes a utility function (measures
"happiness" or efficiency).
• Learning Agents- Improves performance over time using feedback.
 Problem Formulation
• Problem formulation is the process of defining an AI task in a structured
way so that an agent can solve it.
• It involves breaking down a real-world problem into components that an
AI system can process.
• Steps:
• Define the Goal: What should the agent achieve?
• Identify the States: Possible configurations of the environment.
• Define Actions: Choices available to the agent.
• Transition Model: How actions change states.
• Goal Test: Check if the current state meets the goal.
• Cost/Reward: Metrics for action success.