ISC 4322/6300 – GAM 4322 Artificial Intelligence University Of Houston- Victoria Computer Science Department ISC 4322/6300 – GAM 4322 Artificial Intelligence Lecture 1 The Foundations of AI and Intelligent Agents

Contacts Instructor: Dr. Alireza Tavakkoli Office: UC 101C Mon. 5:00 – 6:00 pm Office hours: Wed. 3:00 – 4:00 pm or by appointment Email: tavakkolia@uhv.edu URL: http://www2.uhv.edu/tavakkolia/teaching/ISC4322-6300/index.html Or Blackboard

Class Goals Introduce state-of-the-art algorithms in artificial intelligence Show how these algorithms apply to real-world problems Provide practice in applying algorithms by solving “real” problems (Games) After this course, you should be able: Evaluate claims about intelligent systems in an informed way Have a great algorithmic toolbox to help you design adaptive systems Build intelligent agents for interesting agents Contribute to AI research

Structure of Course Part I: Decision-Making in Deterministic Environments Single agent: Dynamic programming, informed search techniques, randomized search, genetic algorithms, constraint satisfaction, planning Multi-agent: Game theory, mini-max search and approximations Part II: Decision-Making in Stochastic Environments Single agent: Bayesian Networks, Hidden Markov Models, Kalman and Particle Filters, decision and utility theory, Markov Decision Processes Multi-agent: Expectimax search, stochastic game theory Part III: Learning in Unknown Environments Supervised learning: Decision Trees, Support Vector Machines, Neural Networks Unsupervised learning: reinforcement learning

What is Expected Exams Assignments Grading Policy One mid-term, one Final Assignments 3-4 Homeworks, 2-3 Programming Assignment Projects can be complete by pairs of students. Late assignments: (2n)2% Penalty Grading Policy Homework 15% <87 A Projects 25% 75-87 B Midterm 25% 62-74 C Final 30% 50-61 D Participation 5% 50< F

What is AI? Humanly vs. Rationally Thinking vs. Acting “The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning” (Bellman,1978) “The study of mental faculties through the use of computational models.” (Winston, 1992) “AI is concerned with rational action… and studies the design of rational agents. A rational agent acts so as to achieve the best expected outcome.” (S.R. & P.N., 1995) “The art of creating machines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990)

What is AI? Humanly vs. Rationally Thinking vs. Acting 1 “The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning” (Bellman,1978) “The study of mental faculties through the use of computational models.” (Winston, 1992) 1 “AI is concerned with rational action… and studies the design of rational agents. A rational agent acts so as to achieve the best expected outcome.” (S.R. & P.N., 1995) “The art of creating machines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990)

Acting Humanly Turing Test (1950) Capabilities that the computer would need to possess: Natural language processing Knowledge representation Automated reasoning Machine learning Alan Turing (1912-1954) The total Turing test requires also computer vision and robotics The quest for “artificial flight” succeeded when the Wright brothers and others stopped imitating birds and learned about aerodynamics.

What is AI? Humanly vs. Rationally 2 Thinking vs. Acting “The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning” (Bellman,1978) “The study of mental faculties through the use of computational models.” (Winston, 1992) “AI is concerned with rational action… and studies the design of rational agents. A rational agent acts so as to achieve the best expected outcome.” (S.R. & P.N., 1995) “The art of creating machines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990)

Thinking Humanly Simon and Newel came up with the General Problem Solver – GPS (1961). GPS can solve (in principle) every problem that can be formalized symbolically, e.g. Towers of Hanoi Herbert Simon (1916-2001) Allen Newel (1927-1992) First system to separate knowledge of problems from its strategy on how to solve them “Over Christmas, Allen Newell and I created a thinking machine.” Herbert Simon, 1961 Main concern by Simon and Newel: Is the trace of GPS’ reasoning the same as the trace of a human? But this if the focus of…. Cognitive science: Requires experimental investigation of actual humans or animals

What is AI? Humanly vs. Rationally Thinking 3 vs. Acting “The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning” (Bellman,1978) “The study of mental faculties through the use of computational models.” (Winston, 1992) “AI is concerned with rational action… and studies the design of rational agents. A rational agent acts so as to achieve the best expected outcome.” (S.R. & P.N., 1995) “The art of creating machines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990)

Thinking Rationally “Socrates is a man; all men are mortal; therefore, Socrates is mortal.” Logic-based AI Logic provides: A syntax for describing the world and relations among objects A way to achieve logical/correct inferences Create intelligent systems that reason using the syntax and the laws of logic By 1965 there were programs that could in principle, solve any solvable problem described in logical notation. Aristotle (384-322 BC) But… How to represent knowledge in logical notation when we are less than 100% certain? Computational explosion: A few hundred of facts exhaust the computational resources of any computer when it attempts logical inference without some sort of guidance

What is AI? Humanly vs. Rationally Thinking vs. Acting 4 “The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning” (Bellman,1978) “The study of mental faculties through the use of computational models.” (Winston, 1992) 4 “The art of creating machines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990) “AI is concerned with rational action… and studies the design of rational agents. A rational agent acts so as to achieve the best expected outcome.” (Russell and Norvig, 1995)

Acting Rationally Logical inference is part of intelligence. It does not cover everything: e.g., might be no provably correct thing to do, but still something must be done e.g., reflex actions can be more successful than slower, carefully deliberated ones What is a rational action? One that achieves the best outcome (or in the case of uncertainty… the best expected outcome) Stuart Russell It is this objective that requires: Natural language processing Knowledge representation Automated reasoning Machine Learning Computer Vision Robotics “Human behavior is well adapted for one specific environment and is the product, in part, of a complicated and largely unknown evolutionary process that is still far from producing perfection.” Peter Norvig

The Foundations of AI Philosophy Logic, methods of reasoning, mind as physical system foundations of learning, language, rationality Mathematics Formal representation and proof algorithms, computation, (un)decidability, (in)tractability, probability Economics utility, decision theory Neuroscience physical substrate for mental activity Psychology phenomena of perception and motor control, experimental techniques Computer building fast computers engineering Control theory design systems that maximize an objective function over time Linguistics knowledge representation, grammar

AI History – Modern Era 1943 McCulloch & Pitts: Boolean circuit model of brain 1950 Turing's "Computing Machinery and Intelligence" 1956 Dartmouth meeting: "Artificial Intelligence" adopted 1952—69 Look, Ma, no hands! 1950s Early AI programs, including Samuel's checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine 1965 Robinson's complete algorithm for logical reasoning 1966—73 AI discovers computational complexity Neural network research almost disappears 1969—79 Early development of knowledge-based systems 1980-- AI becomes an industry 1986-- Neural networks return to popularity 1987-- AI becomes a science 1995-- The emergence of intelligent agents

State of the Art Deep Blue defeated the reigning world chess champion Garry Kasparov in 1997 Proved a mathematical conjecture (Robbins conjecture) unsolved for decades No hands across America (driving autonomously 98% of the time from Pittsburgh to San Diego) During the 1991 Gulf War, US forces deployed an AI logistics planning and scheduling program that involved up to 50,000 vehicles, cargo, and people NASA's on-board autonomous planning program controlled the scheduling of operations for a spacecraft Proverb solves crossword puzzles better than most humans

Where are we now? SKICAT: automatically classifies data from space telescopes and indentifies interesting objects in the sky. 94% accuracy, way better than human (decision trees) Deep Blue: the first computer program to defeat human champion Garry Kasparov (minimax search + alpha-beta-pruning + optimizations) Pegasus, Jupiter, etc.: speech recognition systems (Hidden Markov Models) HipNav: a robot hip-replacement surgeon (planning algorithms) DARPA Grand/Urban Challenge: autonomous driving 98% of the time from Pittsburgh to San Diego (filtering and planning algorithms) Deep Space 1: NASA spacecraft that did an autonomous flyby an asteroid (logic- based AI) Credit card fraud detection and loan approval (decision trees and neural networks) Chinook: the world checker’s champion (game theory) Spam Assassin and other spam detectors (naïve Bayes learning) Soccer playing Aibo robots (reinforcement learning)

Where are we now?

Intelligent Agents Agent ? How to fully describe an AI problem: Percepts Environment Sensors How to fully describe an AI problem: Performance measure Environment Actuators Sensors ? Actions Actuators Intelligent (rational) agent For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge of the environment the agent has Definition encompasses both physical and virtual/software agents

Vacuum-Cleaner World Performance Measure Environment Actions Percepts Award One Point per Clean Cell Over 1000 Time Step Environment Two Cell World Actions Left, Right, Suck, NoOp Percepts Location and Contents, e.g., [A,Dirty] Is this a rational agent?

Rational Agents Rationality is distinct from omniscience (all-knowing with infinite knowledge) Agents can perform actions in order to modify future percepts so as to obtain useful information (information gathering, exploration) An agent is autonomous if its behavior is determined by its own experience (with ability to learn and adapt)

PEAS PEAS: Performance measure, Environment, Actuators, Sensors Must first specify the setting for intelligent agent design Consider, e.g., the task of designing an automated taxi driver Agent Type Performance Measure Environment Actuators Sensors Taxi Driver Safe Fast Legal Comfortable Profitable Roads Other Traffic Pedestrians Customers Steering Accelerator Brake Signal Horn Display Cameras Sonar Speedometer GPS Odometer Engine Sensors Keyboard

Properties of Environments Fully observable vs. partially observable: If the agent has always access to the complete state of the environment, then the environment is fully observable Partial observability due to: noisy, inaccurate sensors or sensor limitations Deterministic vs. stochastic: If the next state of the world is completely determined by the current state and the agent’s action, then the environment is deterministic. If the environment is deterministic except from the actions of other agents, then it is called strategic. Episodic vs. sequential: In an episodic environment the agent’s experience does not depend on the actions taken in previous episodes. In sequential environments, the current decision could affect all future decisions.

Properties of Environments Static vs. dynamic: If the environment changes while the agent is not acting, then the environment is dynamic. Discrete vs. continuous: Continuous values can appear in: The state of the environment, time, the percepts or actions of the agent. Single agent vs. multiagent: When are the other entities in the world considered agents? Competitive environments: When they maximize an performance measure that depends on the agent’s actions Cooperative environments: When they communicate or follow common rules so as to satisfy a common performance measure

Agent Functions and Programs How does the inside of a rational agent work? An agent is completely specified by the agent function mapping percept sequences to actions Agent = Architecture + Program Agent Program vs. Agent Function Program takes the percept and returns the action Function takes the entire percept history One agent function (or a small equivalence class) is rational Aim: find a way to implement the rational agent function concisely

A Table Driven Agent Drawbacks: Huge table A Table Driven Agent Program Function TABLE-DRIVEN-AGENT (percept) returns action static percepts, a sequence, initially empty table, a table of actions, initially fully specified append percept to the end of percepts action Lookup (percepts, table) return action Drawbacks: Huge table Take a long time to build the table No autonomy Even with learning, needs a long time to learn the table entries

Reflex Vacuum Agent Program Function REFLEX-VACUUM-AGENT ([location, status]) returns action if status = Dirty then return Suck else if location = A then return Right else if location = B then return Left Sensing input: [Cell Id, Dirty or Not Dirty] Actions: [Clean, move Left, move Right, NoOp] Advantages Small Table Cutting Down the Table Size by Eliminating Percept History Action Does Not Depend on the Location

How do agents work? ? Agent Percepts Environment Actions Sensors Actuators

Reflex Agents Agent Percepts Environment Actions Sensors Actuators What the world is like now? Condition-action (if-then) rules What action should I do now? Actions Actuators

Model-based Reflex Agents Environment Sensors Percepts State What the world is like now? How the world evolves? What my actions do? Condition-action (if-then) rules What action should I do now? Actions Actuators

Goal-based Agents Agent Percepts Environment Actions Search & Planning Sensors Percepts State What the world is like now? How the world evolves? What my actions do? What it will be if I do action A? What action should I do now? Goals Actions Actuators Search & Planning

Utility-based Agents Agent Percepts Environment Actions Sensors Percepts State What the world is like now? How the world evolves? What it will be if I do action A? What my actions do? How happy I will be in the state? Utility What action should I do now? Actions Actuators Utility Function: X (state space)  

Learning Agents Agent Performance standard Percepts Environment State Sensors Percepts feedback changes Learning element Performance Element knowledge learning goals Problem generator Actions Actuators

Intelligent Agents

Types of Agents

Question? AI: A Modern Approach : Chapters 1 & 2