AI History CE417: Introduction to Artificial Intelligence Sharif - PowerPoint PPT Presentation

AI History CE417: Introduction to Artificial Intelligence Sharif University of Technology Spring 2017 Soleymani

Ancient History  The intellectual roots of AI and intelligent machines (human-like artifacts) in mythology  Mechanical devices behaving with some degree of intelligence. 2

Modern History  By emerging modern computers, it became possible to create programs performing difficult intellectual tasks.  From these programs, general tools are constructed which have applications in a wide variety of everyday problems.  Emerging computing programmable devices (electronic computers) was a major breakthrough to make intelligent systems. 3

Early Successes Predictions that AI would eventually do almost anything AI Timeline  1943 McCulloch & Pitts: Boolean circuit model of brain  1950 Turing's "Computing Machinery and Intelligence “ paper  1956 Dartmouth meeting: "Artificial Intelligence" term coined  1952-69 Early AI progress, great expectations  1965 Robinson's complete algorithm for logical reasoning  1966-73 AI discovers computational complexity Dark Age Neural network research almost disappears  1969-79 Early development of knowledge-based systems Crawl back  1980-- AI becomes an industry  1986-- Neural networks return to popularity  1987-- AI becomes a scientific method Industrial & Scientific Age  1995-- The emergence of intelligent agents  2001-- AI on very large datasets 4

Periods in AI (briefly)  Early period - 1950 ’ s & 60 ’ s (mostly based on search)  Game playing (brute force), theorem proving (symbol manipulation), biological models (neural networks)  Symbolic application period - 70 ’ s  Early expert systems, use of knowledge  Commercial period - 80 ’ s  knowledge/ rule bases  Scientific & Industrial period - 90 ’ s and early 21 st Century  Rapid advance due to greater use of solid mathematical methods and rigorous scientific standards  Real-world applications 5

The Gestation of AI (1943-1956) Neural Network  The first AI work: Modeling of Neurons  Warren McCulloch & Walter Pitts, 1943  Any computable function could be computed by some network of connected neurons  Learning neural network (Hebbian rule): updating rule for modifying the weights of connection between neurons  Donald Hebb, 1949  First neural network computer (SNARC)  Marvin Minsky & Dean Edmonds (undergraduate students at Harvard), 1950  Minsky studied universal computation in neural networks during his PhD at Princeton  Later, Minsky proved theorems showing limitations of NN 6

The Gestation of AI (1943-1956) Turing  Alan Turing (1950)  “ Computing Machinery and Intelligence ” (1950) paper includes a complete vision of AI  Turing introduced the Turing test, machine learning, genetic algorithms, and reinforcement learning fields  First Chess Player Program  Claude Shannon & Alan Turing, 1950s 7

The birth of AI (1956)  John McCarthy organized a 2 month workshop at Dartmouth College  McCarthy (Stanford), Minsky (MIT), Simon & Newell (CMU), Samuel (IBM)  “ every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. ”  Achieved no new breakthroughs but AI was dominated by these people and their students and colleagues for the next 20 years  “ Artificial Intelligence ” name was chosen by McCarthy during workshop 8

The birth of AI (1956)  Why AI becomes a separate field: AI duplicates human faculties like creativity, self-improvement, and  language use Methodology: a branch of computer science and the only filed trying to  build machines functioning autonomously in complex, changing environments  Newell and Simon from CMU presented the most general program  Logic Theorist (LT) as a reasoning program (proved many mathematical theorems) 9

Early enthusiasm, great expectations (1952- 1969) - “ Look, Ma, no hands! ”  Many successes (in a limited way) in early years of AI  In few years computers from doing just arithmetic to machines did anything remotely clever  General Problem Solver (GPS) – CMU (Simon & Newell, 1960)  Imitated human thinking  Geometry Theorem Prover – IBM (Gelenter, 1959)  proved theorems that many students of mathematics would find tricky  Checkers Player Machines (Arthur Samuel, 1952)  Using game tree search & Reinforcement Learning  McCarthy, MIT, 1958  LISP ,Time Sharing,Advice Taker (the first complete AI system) 10

Early enthusiasm, great expectations (1952- 1969) - “ Look, Ma, no hands! ”  McCarthy (logic) vs. Minsky (anti-logical outlook)  Minsky ’ s group chose limited problems known as microworlds appeared to require intelligence to solve.  e.g. closed form calculus integration problems, geometric analogy problems that appear in IQ tests, blocks world  NN of McCulloch-Pitts flourished  Enhancing learning byWidrow (1960, 1962) rules  Perceptron by Rosenblatt (1962) and convergence theorem 11

A dose of realty (1966-1973)  Herbert Simon, 1957  The power of AI will increase so rapidly that in a visible future, the range of problems they can handle will be coextensive to that of human.  Predictions did not come true  Problems ( Early systems turned out to fail on wider selections or more difficult problems)  Most of early programs contained little or no knowledge of subject matter  1966, “ There is no Machine Translation for general scientific text and there would be no in immediate prospect. ”  Intractability of problems ( “ Combinatorial Explosion ” )  Failed to prove theorems involving more than a dozen of facts  Lighthill report, 1973  Cancellation of almost all AI research in G.B.  Fundamental limitations on basic structures used to generate intelligent behavior 12

Knowledge based systems: The key to power (1969-1979)  First decade of AI research  General purpose search mechanisms (weak methods – general but cannot scale up)  Alternative – more powerful, domain specific knowledge  DENDRAL, 1969 - Inferring molecular structure  MYCIN, 1971 - Diagnosis of blood infections with 450 rules  Natural language understanding  Shrdlu – Blocks world  Schank,Yale  Demands for workable knowledge representation schemes (Prolog, PLANNER, Minsky ’ s idea of frames) 13

AI becomes industry 1980-present  R1 Expert System at DEC, 1982  Configure orders for new computer systems  Saving $40 million per year  The Fifth Generation Project, 1981 (Japanese)  10 year plan to build intelligent computers running Prolog  Counter attacks in U.S. and G.B.  From a few million dollars in 1980 to billions of dollars in 1988  Expert systems, vision systems, robots, software and hardware specialized for these purposes 14

The return of neural networks 1986-present  Reinvention of BACK-PROPAGATION  First in 1969, then in 1986.  Connectionist  Connectionist vs. Symbolic  Symbolism: manipulating knowledge of the world as explicit symbols (e.g., words), where these symbols have clear relationships to entities in the world  Connectionism: embodying knowledge by assigning numerical conductivities or weights to connections inside a network of nodes 15

AI adopts the scientific method 1987- present  It is more common to build on existing theories than to propose brand-new ones  To base claims on rigorous theorems (rather than intuition) and hard experimental evidence (real applications rather than toy examples)  Early isolation of AI from the rest of computer science has been abandoned ( Neats defeated Scruffies )  Samples of revolutions  HMM for speech recognition and machine translation  Baysian network for uncertain knowledge representation and reasoning  NN became comparable to corresponding techniques (e.g. statistics) 16

Emergence of Intelligent Agents 1995-present  “ Whole Agent ”  Reorganizing previously isolated subfields of AI  Influential founders of AI have expressed discontent with the progress of AI  AI should put less emphasis on creating ever-improved version of applications that are good at a specific task  AI should return to its roots “ machines that think, that learn, and that create ” (Human-level AI or HLAI)  Artificial General Intelligence (AGI), 2007  Universal algorithm for learning and acting in any environment 17

Large Data Sets 2001-present  Data became more important than algorithm  Word-sense disambiguation  Performance increasing yield from using more data exceeds any difference in algorithm choice  Filling in holes of a photograph  Poor when 10000 photos available while excellent when 2000000 photos in collection  Knowledge bottleneck  Learning with enough data instead of hand-coded knowledge engineering 18