Comprehensive Introduction to Artificial Intelligence and Intelligent Agents

Institutional Mission, Vision, and Core Values

• Mission: Christ University serves as a nurturing ground for the holistic development of individuals, enabling them to make effective contributions to society within a dynamic environment.
• Vision: Excellence and Service.
• Core Values:
  • Faith in God
  • Moral Uprightness
  • Love of Fellow Beings
  • Social Responsibility
  • Pursuit of Excellence
• Department: Department of AIML & DS.
• Faculty: Dr. Raghavendra S.

Course Syllabus and Learning Objectives

• Unit 1: Introduction
  • Topics: Definition, Future of Artificial Intelligence, Characteristics, Typical Intelligent agents, Problem solving approach, Problem Solving methods, example problems, and searching for solutions.
  • Objective (CO1): Interpret the basic concepts of Artificial Intelligence and Production Systems.
• Unit 2: Searching Technique
  • Topics: Informed Heuristics Search Strategies, Heuristic function, Greedy best-first search, $A^{*}$ Algorithms, Local search algorithms, optimization problems, Hill-climbing Search, Local beam Search, Genetic algorithm, and Searching with partial observations.
  • Objective (CO2): Outline and apply searching techniques.
• Unit 3: Game Playing and Planning
  • Topics: Games, Optimal decisions in games, Min-Max algorithm, Alpha-Beta Pruning, Imperfect real-time decision, Stochastic Games, Planning, The Blocks world, components of planning system, and Goal stack Planning.
  • Objective (CO3): Demonstrate the concept of Game Playing and Planning.
• Unit 4: Knowledge Representation and Inference
  • Topics: First order logic (Syntax and semantics), Using first order logic, Inference in First order logic, unification, Unification and lifting, forward chaining, and backward chaining.
  • Objective (CO4): Apply the concept of knowledge representation and inference using First Order Logic.
• Unit 5: Learning
  • Topics: Learning from Examples, Forms of Learning, Supervised learning, Learning Decision Trees, Regression and classification with linear models, Artificial Neural Network, and Logical formulation of learning.
  • Objective (CO5): Summarize the concepts of Learning.

Assessment Components

• CIA 1:
  • Component 1: Online Course Certification.
  • Component 2: Programming Assignment.
• CIA 3:
  • Component 1: Closed Book Test on Knowledge Representation and Inference.
  • Component 2: Introduction to Modern AI (Cisco Certified free course).

Fundamentals of Artificial Intelligence

• Etymology:
  • Artificial: Man-made.
  • Intelligence: Thinking power.
• Core Definition: Artificial Intelligence is a branch of computer science focused on creating intelligent machines that behave like humans, think like humans, and possess the ability to make decisions.
• Philosophical Nature: AI is a broad field rooted in philosophy. It is viewed as a general philosophical problem with various manifestations and diverse solutions across multiple fields.
• Origins: AI is a relatively new field. Work began shortly after World War II, and the term was officially coined in $1956$.
• Key Figures:
  • John McCarthy: Creator of LISP.
  • Marvin Minsky: Built the first Neural Network (NN) Learning Machine.
  • Allen Newell and Herbert A. Simon.
• Scope: AI encompasses generic fields like Learning and Perception, as well as specific applications including chess, mathematical theorem proving, poetry writing, driving, and medical diagnosis.

Historical Evolution of AI

• Developmental Eras:
  • $1943$-$1952$: Maturation of Artificial Intelligence (Evolution of artificial neurons in $1943$, Turing Machine in $1950$).
  • $1952$-$1956$: The Birth of AI (Dartmouth Conference in $1956$).
  • $1956$-$1974$: Golden years and early enthusiasm (First chatbot ELIZA in $1966$, first intelligent robot WABOT-$1$ in $1972$).
  • $1974$-$1980$: The first AI winter.
  • $1980$-$1987$: A boom of AI (Rise of Expert Systems).
  • $1987$-$1993$: The second AI winter.
  • $1993$-$2011$: Emergence of intelligent agents (IBM Deep Blue beats world chess champion in $1997$, Roomba in $2002$, IBM Watson wins quiz show in $2011$).
  • $2011$-Present: Deep learning, big data, and artificial general intelligence (Google Now in $2012$, Chatbot Eugene Goostman passes Turing test in $2014$, Amazon Echo in $2015$).
• Paradigmatic Phases:
  • Phase 1: Logic-based AI or Symbolic AI, characterized by combinatorial algorithms and search.
  • Phase 2 (Late $90$s to early $2010$s): Probabilistic AI, featuring the Bayesian network invented by Judea Pearl.
  • Current Phase (Post-$2010$s): Neural models, which have overtaken previous methodologies.

Applications, Advantages, and Disadvantages

• Application Domains: Finance, Data Security, Social Media, Automotive, Gaming, Astronomy, Healthcare, Transport, Agriculture, Robotics, Entertainment, E-commerce, and Education.
• Strategic Justifications for AI:
  • Solving real-world problems with specialized software/devices.
  • Creating personal virtual assistants.
  • Building robotic systems.
  • Establishing a foundation for other open technologies.
• Goals of AI:
  • Replicating human intelligence.
  • Solving knowledge-intensive tasks.
  • Integrating perception and action.
  • Creating systems that exhibit human intelligence.
• Multidisciplinary Foundations: AI comprises Computer Science, Sociology, Psychology, Mathematics, Neuron Science, Philosophy, and Biology.
• Advantages:
  • High accuracy with reduced error rates.
  • High reliability and speed.
  • Deployment as digital assistants and public utilities.
  • Suitability for high-risk areas.
• Disadvantages:
  • High implementation costs.
  • Inability to think outside the box.
  • Absence of feelings and emotions.
  • Increasing human dependency on machines.
  • Lack of original creativity.

Definitions and Dimensions of AI

• AI is defined across two dimensions, resulting in four distinct approaches:
  • Thinking Humanly: Systems that think like humans (Human-centered approach based on observations of behavior).
  • Thinking Rationally: Systems that think rationally (Rationalist approach combining mathematics and engineering).
  • Acting Humanly: Systems that act like humans.
  • Acting Rationally: Systems that act rationally.
• Key Components related to modern AI implementation:
  • Machine Learning: Algorithms that improve performance as they are exposed to more data over time; a subset of AI using statistical methods.
  • Deep Learning: A subset of Machine Learning utilizing multilayered neural networks to learn from vast amounts of data.
  • Cognitive Computing.
  • Natural Language Processing (NLP).
  • Computer Vision.

Functional Components and Characteristics of AI

• Core Functionalities:
  • Reasoning: Processes enabling judgment, decision-making, and prediction. Includes Inductive reasoning (specific to general) and Deductive reasoning (general to specific).
  • Learning: Gaining knowledge through study, practice, instruction, or experience.
  • Problem Solving/Decision Making: Moving from a present situation to a desired solution via a specific path.
  • Perception/Sensing: Acquiring, interpreting, and organizing sensory information.
  • Language Understanding: Interpersonal intelligence used to comprehend and use verbal or written language.
• Key Characteristics:
  • Situatedness: Performing actions based on the specific situation.
  • Autonomy: Acting correctly without human intervention.
  • Adaptivity: Taking initiative based on internal goals.
  • Sociability: Interacting with other agents.

Intelligent Agents: Principles and Terminology

• Agent Definition: Anything that perceives its environment through sensors and acts upon it through actuators.
  • Human Agent: Sensors (eyes, ears, organs), Actuators (hands, legs, vocal tract).
  • Robotic Agent: Sensors (cameras, infrared range finders), Actuators (various motors).
  • Software Agent: Sensors (keystrokes, file contents, network packets), Actuators (screen displays, writing files, sending packets).
• Core Terminologies:
  • Percept: Perceptual input at any given instant.
  • Percept Sequence: The complete history of everything the agent has perceived.
  • Agent Function: An abstract mathematical description mapping a percept sequence to an action.
  • Agent Program: The concrete implementation of the agent function running on a physical architecture.
  • Equation: $Agent = Architecture + Program$.
• Rationality: A rational agent strives to do the "right thing" to maximize success based on what it perceives and the actions it can perform.
  • Performance Measure: An objective criterion for success (e.g., for a vacuum: dirt cleaned, time taken, electricity used).
  • Rationality Factors: Performance function, prior knowledge, available actions, and percept sequence.

Task Environments and PEAS Framework

• PEAS analysis is essential for agent design:
  • Performance: Success criteria.
  • Environment: The external world of operation.
  • Actuators: Means of acting.
  • Sensors: Means of perceiving.
• Example: Automated Taxi Driver
  • Performance: Safe, fast, legal, comfortable trip, maximized profits.
  • Environment: Roads, traffic, pedestrians, weather, customers, signals, GPS coverage.
  • Actuators: Steering, accelerator, brake, signals, horn, display.
  • Sensors: Cameras, LiDAR, Sonar, speedometer, GPS, odometer, accelerometer, engine sensors, keyboard.
• Example: Internet Shopping Agent
  • Performance: Price, quality, appropriateness, efficiency, delivery time, user satisfaction.
  • Environment: WWW, vendors, shippers, mobile/web interfaces, regional currencies.
  • Actuators: Form filling, display to user, following URLs, adding to cart, HTTP requests.
  • Sensors: HTML pages, images, buttons, APIs ($JSON/XML$), cookies, user queries.
• Example: Part-picking Robot
  • Performance: Percentage of parts in correct bins.
  • Environment: Conveyor belt with parts, bins.
  • Actuators: Jointed arm and hand.
  • Sensors: Camera, joint angle sensors.

Properties of Task Environments

• Fully Observable vs. Partially Observable: Whether the agent has complete access to the state of the environment at all times.
• Single Agent vs. Multiagent: Whether the agent operates alone or interacts with others (Competitive or Co-operative).
• Deterministic vs. Stochastic: Whether the next state is perfectly predictable from the current state and action.
• Episodic vs. Sequential: Whether the current decision is independent of or dependent on previous actions.
• Static vs. Dynamic: Whether the environment changes while the agent is deliberating.
• Discrete vs. Continuous: Whether states and actions are distinct/finite or vary smoothly across a range.
• Known vs. Unknown: Whether the agent knows the rules/physics of the environment.

Structure of Agent Programs

• Table-Driven Agent: Uses a lookup table of actions indexed by percept sequences.
  • Space Requirement: For a percept set $P$ and lifetime $T$, the table contains $|P|^T$ entries, which is often physically impossible to store or create.
• Simple Reflex Agent: Selects actions based solely on the current percept using condition-action rules (IF-THEN). It has no memory and may enter infinite loops.
• Model-Based Reflex Agent: Maintains an internal state (model) to track unobserved aspects of the world. It understands how the world evolves and how its own actions affect it.
• Goal-Based Agent: Uses goal information to evaluate future actions. It employs searching and planning to find sequences of actions that reach desirable states.
• Utility-Based Agent: Uses a utility function to maximize happiness or preference. It allows for trade-offs (e.g., speed vs. safety) and characterizes success when multiple paths to a goal exist.
• Learning Agent: Improves performance over time. Comprises four parts:
  • Learning Element: Makes improvements.
  • Performance Element: Selects external actions.
  • Critic: Provides feedback to the learning element.
  • Problem Generator: Suggests new actions for exploration.

Problem Solving and Search

• Problem-Solving Agents: A type of goal-based agent using atomic representations (states are indivisible wholes).
• Phases of Problem Solving:

1. Goal Formulation: Based on current situation and performance measures.
2. Problem Formulation: Formalizing actions and states.
3. Search: Looking for a sequence of actions to reach the goal.
4. Execution: Carrying out the found actions.

• Well-Defined Problem Components:
  • Initial State: Starting condition (e.g., "In(Arad)").
  • Actions: Applicable moves in a given state (e.g., "Go(Sibiu)").
  • Transition Model: Result of an action (e.g., "RESULT(In(Arad), Go(Zerind)) = In(Zerind)").
  • Goal Test: Checking if a state is the goal.
  • Path Cost: Numeric cost assigned to a path ($c(s, a, s')$).
• Search Algorithms:
  • Uninformed Search: No info beyond problem definition.
  • Informed Search: Uses state-specific information to solve efficiently.

Standard Problem Examples

• Vacuum World:
  • States: $n \times 2^n$ states ($8$ for $2$ locations).
  • Actions: Left, Right, Suck.
  • Goal Test: All squares clean.
  • Path Cost: $1$ per action.
• $8$-Puzzle:
  • States: $9!$ permutations ($362,880$), but only half are reachable ($181,440$).
  • Actions: Move blank Left, Right, Up, Down.
  • Path Cost: $1$ per move.
• Robotic Assembly:
  • States: Real-valued coordinates of joint angles.
  • Actions: Continuous motions.
• $8$ Queens Problem:
  • Initial State: Empty board.
  • Actions: Add a queen to an empty square.
  • Goal Test: $8$ queens on board with none under attack.