Bil682 - Yapay Anlayış Artificial Intelligence Güz 2011 Dr. Nazlı İkizler Cinbiş Slides mostly adapted from AIMA

What is AI? A system is rational if it does the right thing given what it knows Thinking humanly The exiting new effort to make computers think…machines with minds, in the full and literal sense (Haugeland, 1985) The automation of activities that we associate with human thinking, activities such as decision making, problem solving, learning… (Bellman, 1978) Thinking rationally The study of mental faculties through the use of computational models (Charniak and McDermott, 1985) The study of the computations that make it possible to perceive, reason and act (Winston, 1992) Acting humanly The art of creating machines that perform functions that require intelligence when performed by people (Kurzweil, 1990) The study of how to make computers do things at which, at the moment, people are better (Rich and Knight, 1991) Acting rationally Computational Intelligence is the study of the design of intelligent agents (Poole et. al, 1998) AI … is concerned with intelligent behaviour in artifacts (Nilson, 1998)

Acting humanly: Turing Test Alan Turing (1950) "Computing machinery and intelligence": Operational test for intelligent behavior: the Imitation Game The computer passes the test if a human interrogator, after posing some written questions, cannot tell whether the written responses come from a person or not. If the response of a computer to an unrestricted textual natural-language conversation cannot be distinguished from that of a human being then it can be said to be intelligent. Suggested major components of AI: Natural Language Processing, Knowledge Representation, Automated Reasoning, Machine Learning Total Turing Test: requires Computer Vision and Robotics as well

Thinking humanly: cognitive modeling In order to say that a given program thinks like a human, we must have some way of determining how humans think Requires scientific theories of internal activities of the brain -- How to validate? Requires 1) Cognitive Science: Predicting and testing behavior of human subjects (top-down) or 2) Cognitive Neuroscience: Direct identification from neurological data (bottom-up) Both approaches (roughly, Cognitive Science and Cognitive Neuroscience) are now distinct from AI

Thinking rationally: "laws of thought" Aristotle: what are correct arguments/thought processes? Several Greek schools developed various forms of logic: notation and rules of derivation for thoughts; may or may not have proceeded to the idea of mechanization Formalize “correct” reasoning using a mathematical model(e.g. of deductive reasoning). Direct line through mathematics and philosophy to modern AI : logicist tradition hopes to build logic programs to create intelligent systems Problems: 1.Informal knowledge may not be represented in formal terms of logic. 2.Solving problems in principle is different than solving in practice.

Acting rationally Rational behavior: doing the right thing –The right thing: that which is expected to maximize goal achievement, given the available information –Doesn't necessarily involve thinking – e.g., blinking reflex –But thinking should be in the service of rational action –Entirely dependent on goals –Irrational != insane, irrationality is a suboptimal action –Rational != successful

Rational agents An agent is an entity that perceives its environment and is able to execute actions to change it. Abstractly, an agent is a function from percept histories to actions: [f: P*  A ] For any given class of environments and tasks, we seek the agent (or class of agents) with the best performance

Rational agents Agents have inherent goals that they want to achieve (e.g. survive, reproduce). A rational agent acts in a way to maximize the achievement of its goals. True maximization of goals requires omniscience and unlimited computational abilities. In real world, usually lots of uncertainty Usually, we’re just approximating rationality  design best program for given machine resources and available knowledge

Foundations of AI PhilosophyLogic, methods of reasoning, mind as physical system foundations of learning, language, rationality MathematicsFormal representation and proof algorithms, computation, (un)decidability, (in)tractability, probability Economicsutility, decision theory Neurosciencephysical substrate for mental activity Psychology phenomena of perception and motor control, experimental techniques Computer Sciencehardware, algorithms, computational complexity theory, fast computers Control theorydesign systems that maximize an objective function over time Linguisticsknowledge representation, grammar, syntax, semantics Many older disciplines contribute to AI:

Abridged history of AI 1943 McCulloch & Pitts: Boolean circuit model of brain 1950 Turing's "Computing Machinery and Intelligence" 1956Dartmouth meeting: "Artificial Intelligence" adopted 1952—69Look, Ma, no hands! 1950sEarly AI programs, including Samuel's checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine 1965Robinson's complete algorithm for logical reasoning 1960s Work in the sixties at MIT lead by Marvin Minsky and John McCarthy –Development of LISP symbolic programming language –SAINT: Solved freshman calculus problems –ANALOGY: Solved IQ test analogy problems –SIR: Answered simple questions in English –STUDENT: Solved algebra story problems –SHRDLU: Obeyed simple English commands in theblocks world

Abridged history of AI 1966—73AI discovers computational complexity Neural network research almost disappears 1969—79Early development of knowledge-based systems AI becomes an industry Neural networks return to popularity AI becomes a science The emergence of intelligent agents

Early Limitations Hard to scale solutions to toy problems to more realistic ones due to difficulty of formalizing knowledge and combinatorial explosion of search space of potential solutions. Limitations of Perceptron demonstrated by Minsky and Papert (1969). Slide from R. Mooney

Knowledge is Power: Expert Systems Discovery that detailed knowledge of the specific domain can help control search and lead to expert level performance for restricted tasks. First expert system DENDRAL for interpreting mass spectrogram data to determine molecular structure by Buchanan, Feigenbaum, and Lederberg (1969). Early expert systems developed for other tasks: –MYCIN: diagnosis of bacterial infection (1975) –PROSPECTOR: Found molybendum deposit based on geological data (1979) –R1: Configure computers for DEC (1982) Slide from R. Mooney

AI Industry Development of numerous expert systems in early eighties. Estimated $2 billion industry by Japanese start “Fifth Generation” project in 1981 to build intelligent computers based on Prolog logic programming. MCC established in Austin in 1984 to counter Japanese project. Limitations become apparent, prediction of AI Winter –Brittleness and domain specificity –Knowledge acquisition bottleneck Slide from R. Mooney

Rebirth of Neural Networks New algorithms (e.g. backpropagation) discovered for training more complex neural networks (1986). Cognitive modeling of many psychological processes using neural networks, e.g. learning language. Industrial applications: –Character and hand-writing recognition –Speech recognition –Processing credit card applications –Financial prediction –Chemical process control Slide from R. Mooney

What Can AI Do? Quiz: Which of the following can be done at present? Play a decent game of table tennis? Drive safely along a curving mountain road? Drive safely along busy traffic? Buy a week's worth of groceries on the web? Discover and prove a new mathematical theorem? Converse successfully with another person for an hour? Perform a complex surgical operation? Unload a dishwasher and put everything away? Translate spoken Chinese into spoken English in real time? Write an intentionally funny story? Slide from Dan Klein

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 Proverb solves crossword puzzles better than most humans

18 State of the Art NASA's on-board autonomous planning program controlled the scheduling of operations for a spacecraft –Sojourner, Spirit, and Opportunity explore Mars. –NASA Remote Agent in Deep Space I probe explores solar system. DARPA grand challenge: Autonomous vehicle navigates across desert and then urban environment. iRobot Roomba automated vacuum cleaner, and PackBot used in Afghanistan and Iraq wars. Spam filters using machine learning. Question answering systems automatically answer factoid questions. Usable machine translation thru Google.

Recent Times General focus on learning and training methods to address knowledge-acquisition bottleneck. Shift of focus from rule-based and logical methods to probabilistic and statistical methods (e.g. Bayes nets, Hidden Markov Models). Increased interest in particular tasks and applications –Data mining –Machine Learning –Intelligent agents and Internet applications(softbots, believable agents, intelligent information access) –Scheduling/configuration applications

Fields of AI Machine Learning Computer Vision Speech Recognition Natural Language Processing Data Mining Information Retrieval Game programming Robotics, Planning …

About this course Discussion of recent research problems Theory and Applications Hands-on Experience

About this course: Resources Course Book: Artificial Intelligence: A Modern Approach, 3/E Stuart Russell Peter Norvig ISBN-10: ISBN-13: Publisher: Prentice Hall Format: Cloth; 1152 pp Published: 12/01/2009 Book on Learning: Machine Learning, Tom Mitchell, McGraw Hill, Additional Readings: Research papers

Course Contents Search –Uninformed, Informed, Constraint Satisfaction Game Playing Learning –Supervised Learning, Bayesian Learning –Decision Trees, Adaboost –Neural Nets –Hidden Markov Models –Reinforcement Learning Applications and Research Problems on –Computer Vision –Natural Language Processing –…

Evaluation Paper presentation %20 –Presentation %15 –Peer review %5 Participation %10 –Participation to presentations Project %35 –Intermediate report %10 –Final report %25 Final Exam %35

Project Each student should decide on a research project related to a field of AI. Project Deadlines –Project proposals due : –Intermediate project report due: –Final report due:

Reading Assignment Alan Turing’s “Computing Machinery and Intelligence” (1950) –First vision of AI