Ai 01 introduction
4 likes1,803 views
This document discusses the history and foundations of artificial intelligence. It covers early developments in the 1940s-1950s that led to the birth of AI as a field at the 1956 Dartmouth conference. It describes successes and challenges in the 1960s-1970s, the rise of knowledge-based systems and expert systems in the 1970s, and AI becoming an industry in the 1980s. The return of neural networks in the 1980s-1990s is also summarized. The document outlines different approaches to defining and pursuing AI, including systems that think like humans, think rationally, act like humans, and act rationally. It lists philosophy, mathematics, neuroscience, and other disciplines as foundations of AI.
Ai 01 introduction
- 1. 1 Artificial Intelligence Andreas Zell After the Textbook: Artificial IntelligenceAfter the Textbook: Artificial Intelligence, A Modern Approach by Stuart Russel and Peter Norvig (3rd Edition) • Textbook: Artificial Intelligence, A Modern Approach, by Stuart Russel and Peter Norvig (3rd Edition 2009) 0. Organizational Issues (3 Edition, 2009) • http://aima.cs.berkeley.edu/ • The lecture will follow this book very closely • All graphics and drawings not attributed otherwise are taken from this book • My office hours are • Thu. 13:30 -15:00 in A310, Sand 1 • Email: andreas.zell@uni-tuebingen.de 2Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 2. 2 • Research assistants: • Sebastian Otte (Sebastian Otte@uni-tuebingen de room A318) 0. Organizational Issues (Sebastian.Otte@uni tuebingen.de, room A318) • Sebastian Buck (Sebastian.Buck@uni-tuebingen.de, room A311) • Tutoring (Übungen) • Thursdays, 12:00 – 13:00, A104 • You are allowed to eat sandwiches during this timeg (non-smelling, no chips, must clean up desk afterwards) 3Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) • Sebastian Thrun (Stanford Univ.) and Peter Norvig (Google, Inc.) were offering their AI course CS221: Introduction to Artificial 0. Stanford AI Course CS221 course CS221: Introduction to Artificial Intelligence in the fall semester 2011, free for everybody on the Web. • The course is based on the same book (AI:MA) by Stuart Russel and Peter Norvig • It is now offered via Udacity the company of• It is now offered via Udacity, the company of Prof. Thrun, but still free • See https://www.udacity.com/course/cs271 4Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 3. 3 We can group definitions of Artificial intelligence in the following 4 groups: 1 Systems that think like humans 1.1 What is Artificial Intelligence? 1.Systems that think like humans 2.Systems the think rationally 3.Systems that act like humans 4.Systems that act rationally Historically, all four approaches to AI have been followed 5Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) Definitions of Artificial Intelligence Systems that think like humans • “The exciting new effort to k t thi k Systems that think rationally • “The study of mentalmake computers think … machines with minds, in the full and literal sense.” (Haugeland, 1985) • “The automation of activities that we associate with human The study of mental faculties through the use of computational models.” (Charniak & McDermott, 1985) • “The study of computations that make it thinking, … such as decision making, problem solving, learning …” (Bellman, 1978) computations that make it possible to perceive, reason, and act.” (Winston, 1992) 6Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 4. 4 Definitions of Artificial Intelligence Systems that act like humans • “The art of creating hi th t f Systems that act rationally • “Computationalmachines that perform functions that require intelligence when performed by people.” (Kurzweil, 1990) • “The study of how to make computers do Computational intelligence is the study of the design of intelligent agents.” (Poole et al., 1998) • “AI … is concerned with intelligent behaviour inp things at which, at the moment, people are better.” (Rich and Knight, 1991) intelligent behaviour in artifacts.” (Nilsson, 1998) 7Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) • Turing test, proposed by Alan Turing (1950) • Human interrogator is connected to either another human or a machine in another room. Acting like Humans: The Turing Test • Interrogator may ask text questions via a computer terminal, gets answers from the other room. • If the human interrogator cannot distinguish the human from the machine, the machine is said to be intelligent. 8Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 5. 5 • Natural language processing, to enable it to communicate efficiently, • Knowledge representation to store what it knows or Capabilities of an AI system • Knowledge representation, to store what it knows or hears, • Automated reasoning, to use the stored information to answer questions and draw new conclusions • Machine learning, to adapt to new circumstances and detect and extrapolate patterns • Computer vision, to perceive objects, • Robotics, to manipulate objects and move around. 9Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) • Problem: need to determine how humans think • By introspection (but is this really how we think?) • By psychological experiments Thinking humanly: Cognitive Science • Cognitive Science brings together • Computer models from AI, and • Experimental techniques from psychology to try to construct precise and testable theories of the workings of the human mind. • Some AI researchers (e.g. Newell & Simon with their General Problem Solver) tried to compare the trace of program reasoning with human reasoning steps. 10Zell: Artificial Intelligence (after.)
- 6. 6 • Based on the syllogisms of the greek philosopher Aristotle Logical formalism: Predicate logic Thinking rationally: The laws of thought • Logical formalism: Predicate logic • Programs as logic derivation, Theorem proving • Logic programming: Prolog as computer language (program is logic specification) • Japanese “5th generation computer project”, ca. $1982-1988, ca. 450 Million $ Govt. funding • PIM, parallel inference machines, parallel Prolog • http://en.wikipedia.org/wiki/Fifth_generation_computer 11Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) • An agent is just something that acts. • A rational agent is one that acts so as to achieve the best outcome or under uncertainty the best Acting rationally: The rational agent the best outcome or, under uncertainty, the best expected outcome. • Here the emphasis is not on correct inference (as in logic programming), but on reasonable actions, given that information may be uncertain, incomplete or inconsistent and actions mustincomplete or inconsistent and actions must frequently be taken in a limited amount of time. • The Russel/Norvig book advocates this approach to artificial intelligence. 12Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 7. 7 • Philosophy • Mathematics 1.2 Foundations of Artificial Intelligence • Economics • Neuroscience • Psychology • Computer Science and Computer Engineering • Control Thery and CyberneticsControl Thery and Cybernetics • Linguistics (read the subsection in Russel/Norvig) 13Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) 1.2 Foundations of Artificial Intelligence • Philosophy Logic, methods of reasoning, mind as physical system foundations of learning, rationality • Mathematics Formal representation and proof, algorithms,p p , g , 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 engineeringengineering • Control theory design systems that maximize an objective function over time • Linguistics knowledge representation, grammar
- 8. 8 Gestation of Artificial Intelligence (1943 – 1955) • 1943 McCulloch & Pitts: Boolean circuit model of neurons, History of Artificial Intelligence 1 • 1949 Donald Hebb: Hebbian learning rule • 1950 Turing's “Computing Machinery & Intelligence” • 1950 Marvin Minsky SNARC (first neural computer) Birth of artificial Intelligence (1956) • 1956 Dartmouth meeting (John McCarthy, Marvin1956 Dartmouth meeting (John McCarthy, Marvin Minsky, Claude Shannon, N. Rochester, Alan Newell, Herbert Simon, Arthur Samuel, Oliver Selfridge …): name “Artificial Intelligence” 15Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) Early enthusiasm, great expectations (1952 – 1969) • 1956 Newell & Simon: Logic Theorist (LT) • … Newell & Simon: General Problem Solver History of Artificial Intelligence 2 (GPS) • 1952 - 56 A. Samuel: checkers program (played better than its creator) • 1958 John McCarthy: LISP, MIT AI Lab • 1959 David Gelernter: Geometry Theorem Prover • 1965 J. A. Robinson: Resolution, a complete algorithm for logical reasoning • 1970 Patrick Winston: blocks world learning theory • 1972 Terry Winograd: SHRDLU • 1975 David Waltz, Vision & constraint propagation
- 9. 9 A dose of reality (1966 – 1973) • 1966 - 73 AI systems do not scale up well • Neural network research almost disappears History of Artificial Intelligence 3 pp • Automatic translation fails Knowledge-based systems (1969 – 1979) • 1969 B. Buchanan: Dendral (expert system to infer the molecular structure from MS data) • … Ed Feigenbaum et al. Heuristic Programming Project (Stanford U.) • 1975 Shortliffe, Feigenbaum, Buchanan: MYCIN (expert system to diagnose blood infections) • 1976 Newell & Simon: Physical Symbol Systems Hypothesis 17Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) AI becomes an industry (1980 - 1988) • 1982 Digital Equipment, R1, expert system to configure VAX computers History of Artificial Intelligence 4 to configure VAX computers • Teknowledge (expert system company) • Intellicorp (expert system company) • … many industrial expert systems projects The return of neural networks (1986 – 2000) • 1982 John Hopfield Hopfield networks• 1982 John Hopfield, Hopfield networks • 1985 Rumelhart, Hinton, Williams: Backpropagation, • 1986 Rumelhart, McClelland: PDP books 18Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.)
- 10. 10 AI becomes a science (1987 – 2000) • Scientific method: hypothesis – rigorous empirical experiment results History of Artificial Intelligence 5 empirical experiment – results • Neats vs. scruffies • Probabilistic framework • HMMs (Hidden Markov models) • Bayesian networks The emergence of intelligent agents (1995 – 2005) • SOAR, complete agent architecture (Newell, Laird, Rosenbloom) 19Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) • Autonomous planning and scheduling • NASA’s Remote agent program (scheduling of operations on a spacecraft), 2000 1.4 The State of the Art • Game playing • TD-Gammon (Tesauro, IBM) plays Backgammon at world champion level (1992) • Chinook becomes world champion in checkers (1994) • 9 Men‘s Morris (Mühle) solved (Gasser, 1996) http://library.msri.org/books/Book29/files/gasser.pdf • 8x8 Checkers is solved by Chinook (2007)8x8 Checkers is solved by Chinook (2007) • Deep Blue (IBM) defeats human world champion Garry Kasparov in Chess (1997) • In Go (19x19) AI programs are still inferior to top humans, but catching up (in June 2011, Zen19D reached 5 dan) Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) 20
- 11. 11 • Autonomous Cars • Autonomous cars by Uni der BW • Autonomous cars by DaimlerBenz 1.4 State of the Art • Autonomous cars by DaimlerBenz • ALVINN neural network in NavLab (CMU) drives 2850 miles across USA 98% autonomously • Stanley (S. Thrun, Stanford & VW) wins DARPA Grand Challenge (2005) 212 k th h(2005) over 212 km through Mojave Dessert • BOSS (R. Whitacker, CMU & General Motors) wins DARPA Urban Challenge (2007) Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) 21 • Stanford’s Shelley (Audi TTS) trying to race Pikes Peak (2010) • http://www.youtube.com/watch? 1.4 State of the Art v=hbmf0bB38h0&feature=relmfu • Google autonomous cars (Toyota Prius), 2010, used for Google streetmap, travel for 200.000 miles until mid 2011. Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) 22
- 12. 12 • Diagnosis, Help systems, Scheduling • Medical diagnosis programs perform at expert level • Microsoft uses AI technology for help system in 1.4 State of the Art Microsoft uses AI technology for help system in Windows 7 and Office programs • US Army used DART during Persian Gulf war • Robotics • Many mobile robots with AI technology in research • Limited use of AI concepts in commercial robots Language understanding• Language understanding • IBM’s WATSON wins Jeopardy competition against strongest humans (2011) http://www.youtube.com/watch?v=qO1i7-Qx00k Zell: Artificial Intelligence (after Russel/Norvig, 3rd Ed.) 23