Introduction to Artificial Intelligence (AI): Artificial Intelligence (AI) is the field of computer science aimed

at creating machines capable of performing tasks that typically require human intelligence. These include
reasoning, learning, problem-solving, perception, and natural language understanding.
Problems of AI
1. Natural Language Processing (NLP): Understanding and generating human language.
2. Knowledge Representation: Structuring information so that a machine can process and use it effectively.
3. Reasoning and Decision-Making: Drawing conclusions from data and making decisions under
uncertainty.
4. Learning: Enabling systems to improve performance based on experience.
5. Vision: Interpreting visual data from the world.
6. Robotics: Creating machines that can perform physical tasks autonomously.
AI Techniques
 Heuristics: Using rules of thumb for solving complex problems.
 Search Algorithms: Exploring possible solutions to identify optimal results.
 Knowledge Representation: Storing information in logical structures.
 Machine Learning: Enabling systems to identify patterns and learn from data.
 Planning and Optimization: Strategizing actions to achieve specific goals.
The Tic-Tac-Toe game illustrates state-space search, a fundamental AI concept. The game is modeled as:
 States: All possible board configurations.
 Initial State: An empty board.
 Actions: Possible moves for each player.
 Goal State: Winning (three in a row) or drawing (board full).
An Intelligent Agent is an entity that perceives its environment through sensors and acts upon it using actuators
to achieve goals.
Components of an Agent
1. Sensors: Devices or mechanisms to perceive the environment.
2. Actuators: Mechanisms to interact with the environment.
3. Agent Function: Maps perceived inputs to actions.
Types of Agents
1. Simple Reflex Agents: Act based on current percepts only.
2. Model-Based Reflex Agents: Use an internal model to handle partial knowledge of the environment.
3. Goal-Based Agents: Focus on achieving specific goals.
4. Utility-Based Agents: Aim to maximize a utility function for better decision-making.
5. Learning Agents: Improve performance by learning from experiences.
Nature of the Environment
1. Fully Observable vs. Partially Observable: Whether the agent has complete information.
2. Deterministic vs. Stochastic: Predictability of outcomes.
3. Episodic vs. Sequential: Whether actions are independent or affect future actions.
4. Static vs. Dynamic: Whether the environment changes while the agent is deliberating.
5. Discrete vs. Continuous: The nature of states and actions.
Problem-solving in AI involves searching for a sequence of actions to achieve a desired goal state from an initial
state.
Problem Formulation
1. State Space: The set of all possible states of the problem.
2. Initial State: The state where the problem begins.
3. Goal State: The desired outcome.
4. Actions: Possible moves or transitions between states.
5. Path Cost: The cost associated with a sequence of actions.
Production System
A model of computation consisting of:
1. Rules: Conditions and corresponding actions.
2. Control Strategy: Decides the order of rule execution.
3. Working Memory: Holds the current state.
Problem Characteristics
1. Complexity: The size of the state space.
2. Uncertainty: Incomplete or ambiguous information.
3. Dynamicity: Whether the problem changes over time.
Design Issues in Search Programs
1. Representation: Defining states and transitions effectively.
2. Efficiency: Minimizing computational resources.
3. Heuristics: Guiding the search process to prioritize useful paths.
Solving Problems by Searching
Problem-Solving Agents
Agents designed to solve problems by searching for a sequence of actions that lead from an initial state to a goal
state.
1. Steps in Problem Solving:
1. Formulate the Problem: Define states, actions, and goals.
2. Search for a Solution: Explore the state space using search strategies.
3. Execute the Solution: Perform the sequence of actions.
Uniform Search Strategies
1. Breadth-First Search (BFS): Explores all nodes at the current depth level before moving to the next
level. Guarantees the shortest path in an unweighted graph but is memory-intensive.
2. Depth-First Search (DFS): Explores as far as possible along each branch before backtracking. Uses less
memory but is not guaranteed to find the shortest path.
3. Depth-Limited Search (DLS): A variation of DFS that limits the depth of exploration to a predefined
level.
4. Bidirectional Search: Runs two simultaneous searches, one from the initial state and the other from the
goal state, meeting in the middle.
5. Comparison of Strategies: Consider time complexity, space complexity, optimality, and completeness
for each strategy.
Heuristic Search Strategies-Greedy Best-First Search
Uses a heuristic function to evaluate nodes, expanding the most promising one based on a specific heuristic.
A Search*
Combines the actual cost from the start node (g(n)g(n)g(n)) and the estimated cost to the goal (h(n)h(n)h(n))
using f(n)=g(n)+h(n)f(n) = g(n) + h(n)f(n)=g(n)+h(n). It guarantees optimality if h(n)h(n)h(n) is admissible and
consistent.
Memory-Bounded Heuristic Search
1. Hill Climbing Search: Moves to the neighbor with the highest heuristic value. Prone to getting stuck in
local maxima.
2. Simulated Annealing Search: Introduces randomness to escape local maxima, gradually reducing the
probability of selecting worse solutions.
3. Local Beam Search: Maintains multiple states at a time, focusing on the best candidates.
4. Genetic Algorithms: Uses principles of natural selection, crossover, and mutation to evolve solutions.
Constraint Satisfaction Problems (CSPs)
Involve finding values for variables that satisfy constraints. Techniques include:
 Backtracking Search: A depth-first search that assigns values incrementally.
 Local Search: Focuses on optimization, using techniques like hill climbing and simulated annealing.
Games:Two-player games, like chess or tic-tac-toe, involve opponents with conflicting objectives. Adversarial
search focuses on making optimal moves.
Minimax Search Procedure: Evaluates the minimax value of a game tree, where:
 Max nodes aim to maximize the score.
 Min nodes aim to minimize the score.
Alpha-Beta Pruning: Optimizes minimax by eliminating branches that do not affect the outcome, reducing the
number of nodes evaluated.
Additional Refinements
 Iterative Deepening: Combines depth-limited search with breadth-first search advantages.
 Transposition Tables: Store previously evaluated positions to avoid redundant calculations.
Knowledge Representation Issues: Knowledge representation involves encoding information about the world
so that an AI system can use it to solve complex tasks. Challenges include:
1. Expressiveness: Capturing all necessary knowledge without oversimplification.
2. Incompleteness: Representing partial or uncertain information.
3. Ambiguity: Avoiding multiple interpretations of the same knowledge.
4. Complexity: Balancing detail with computational feasibility.
5. Scalability: Handling large amounts of knowledge efficiently.
Representation:Defines how knowledge is structured, stored, and accessed. Examples include:
 Propositional Logic: Represents facts using statements that are either true or false.
 Predicate Logic: Represents relationships and properties using quantifiers and predicates.
Mapping: Involves translating real-world problems into a formal representation. Effective mapping ensures the
system understands the domain correctly.
Approaches to Knowledge Representation
1. Logical Representation: Uses formal logic (propositional, predicate) to infer conclusions.
2. Semantic Networks: Represents knowledge as a graph with nodes (concepts) and edges (relationships).
3. Frame-Based Representation: Uses structured templates to describe objects, attributes, and their values.
4. Production Rules: Encodes knowledge as "if-then" rules for reasoning.
5. Ontologies: Defines a formal representation of a set of concepts and their relationships in a domain.
Issues in Knowledge Representation
1. Representation Trade-offs: Balancing richness and simplicity.
2. Inference Mechanisms: Ensuring the reasoning process is efficient and correct.
3. Handling Uncertainty: Using probabilistic models or fuzzy logic to deal with incomplete or ambiguous
information.
4. Dynamic Knowledge: Adapting to changes in knowledge over time.
5. Domain-Specific Knowledge: Customizing representation for particular applications.
Predicate Logic: Predicate logic extends propositional logic by introducing predicates, quantifiers, and variables
to represent complex relationships and generalizations.
Representing Simple Facts in Logic
1. Facts: Statements like "John is a teacher" are represented as
Teacher(John)Teacher(John)Teacher(John).
2. Relationships: "John teaches Mary" is represented as Teaches(John,Mary)Teaches(John,
Mary)Teaches(John,Mary).
Representing Instance and ISA Relationships
1. Instance Relationship: Representing specific entities as instances of a class, e.g.,
Cat(Tom)Cat(Tom)Cat(Tom).
2. ISA Relationship: Representing subclass relationships, e.g., ∀x(Dog(x)→Animal(x))\forall x
(Dog(x) \rightarrow Animal(x))∀x(Dog(x)→Animal(x)).
Computable Functions and Predicates
1. Functions: Represent transformations, e.g., Father(John)=JackFather(John) =
JackFather(John)=Jack.
2. Predicates: Evaluate relationships or properties, e.g., GreaterThan(5,3)GreaterThan(5,
3)GreaterThan(5,3).
Resolution: A rule of inference used for automated theorem proving. It combines clauses to deduce new
information.
Natural Deduction: A reasoning system based on applying inference rules (e.g., modus ponens,
universal instantiation) to derive conclusions.
Probabilistic Reasoning
Probabilistic reasoning deals with uncertainty by representing and reasoning about degrees of belief.
Representing Knowledge in Uncertain Domains
1. Use probabilities to represent uncertain events or states, e.g., P(Rain)=0.7P(Rain) =
0.7P(Rain)=0.7.
Semantics of Bayesian Networks
1. A graphical model that represents dependencies among variables using directed acyclic graphs.
2. Nodes: Represent random variables.
3. Edges: Represent conditional dependencies.
4. Inference: Calculate probabilities of unknown variables given known ones.
Dempster-Shafer Theory
1. A mathematical theory for reasoning with uncertainty.
2. Combines evidence from multiple sources to calculate belief and plausibility measures.
Fuzzy Sets and Fuzzy Logic
1. Represents imprecise knowledge using degrees of membership rather than binary values.
2. Fuzzy Sets: Allow partial membership, e.g., "Tall" may have degrees like
Tall(John)=0.8Tall(John) = 0.8Tall(John)=0.8.
3. Fuzzy Logic: Extends classical logic to handle reasoning with fuzzy sets.
1. Natural Language Processing (NLP)
Introduction
 Definition: NLP enables machines to understand, interpret, and respond to human language in a
meaningful way.
 Applications: Machine translation, chatbots, sentiment analysis, speech recognition, and information
retrieval.
Syntactic Processing
 Focuses on analyzing the grammatical structure of sentences.
 Parsing: Identifying sentence components (subject, predicate) using:
o Dependency Parsing: Establishes dependency relationships between words.
o Constituency Parsing: Breaks sentences into sub-phrases.
Semantic Analysis
 Assigns meaning to words and sentences.
 Lexical Semantics: Analyzing word meanings and relationships (synonyms, antonyms).
 Semantic Role Labeling: Identifying the role played by a word in a sentence (agent, action, object).
Discourse and Pragmatic Processing
 Discourse: Understanding language context across sentences to maintain coherence.
 Pragmatics: Interpreting language based on situational context, such as speaker intent or tone.
2. Learning
Forms of Learning
1. Supervised Learning: Training on labeled data (e.g., regression, classification).
2. Unsupervised Learning: Discovering patterns in unlabeled data (e.g., clustering).
3. Reinforcement Learning: Learning by interacting with an environment and receiving feedback in the
form of rewards or penalties.
Inductive Learning
 Deriving general rules from specific examples.
 Example: Identifying patterns in data to generate hypotheses.
Learning Decision Trees
 A model where decisions are made by traversing a tree structure.
 Example: Deciding whether to play tennis based on weather conditions.
Explanation-Based Learning
 Focuses on understanding why a specific example belongs to a concept by analyzing its underlying
structure.
Learning Using Relevance Information
 Identifies features that are most relevant to the task.
 Example: Feature selection techniques in machine learning.
Neural Net Learning
 Utilizes artificial neural networks to model complex patterns.
 Example: Deep learning for image recognition.
Genetic Learning
 Inspired by natural evolution.
 Uses genetic algorithms involving selection, crossover, and mutation to optimize solutions.
3. Expert Systems
Representing and Using Domain Knowledge
 Knowledge Representation: Encodes domain expertise into rules or ontologies.
 Example: Representing medical diagnosis knowledge using "if-then" rules.
Expert System Shells
 Software frameworks for building expert systems.
 Includes inference engines and knowledge base templates.
 Example: MYCIN (medical diagnosis system).
Knowledge Acquisition
 Process of gathering and encoding knowledge into the expert system.
 Methods: Interviews with experts, data mining, and literature review.