Artificial Intelligence 人工智慧. Lecture 1 February 22, 2012 洪國寶

Transcription

1 Artificial Intelligence 人工智慧 Lecture 1 February 22, 2012 洪國寶 1

2 Outline Course information Motivations What is Artificial Intelligence A brief history of Artificial Intelligence Outline of the course 2

3 Course information (1/6) Instructor: Professor Gwoboa Horng Basic assumption Little is assumed of the students except a general background in computing. I will cover the main aspects in enough detail for the students to understand the gist. Course web page: Password: ailab 3

4 Course information (2/6) Textbook Artificial Intelligence Illuminated Ben Coppin, Jones and Bartlett Publishers, Inc ( 開發圖書公司 ) catalog/ / Try Google search (keywords: artificial intelligence illuminated ) 4

5 Course information (3/6) 參考書籍 1. Artificial Intelligence (3rd Edition) By E. Rich, B. Knight, and S. Nair, McGraw-Hill, Common LISP. The Language. Second Edition By Guy Steele ml/cltl/clm/clm.html 5

6 Course information (4/6) The objective of this course is to learn fundamental techniques of artificial intelligence. We will cover a wide range of AI techniques, algorithms, and methodologies, including Searching algorithms, Knowledge representations, LISP, Machine learning: neural network, and genetic algorithms, Planning, and Computer understanding of language and images. 6

7 Course information (5/6) This class is Not a lab or programming course But there will be (LISP) programming assignments. Neither a math course, But strong math background will help. 7

8 Course information (6/6) Grading (Tentative) Homework/Quiz 15% Project 15% (Presentation and/or paper required) Midterm exam 25% (Apr. 18, 2012) Final exam 25% (June 20, 2012) Class participation 20% 8

9 Outline Course information Motivations What is Artificial Intelligence A brief history of Artificial Intelligence Outline of the course 9

10 Why Study Artificial Intelligence? Gain insight into human intelligence by considering computational models of intelligence. Gain the ability to create programs that perform functions normally thought to require intelligence. Improve our own problem-solving skills by taking to lessons learned in AI. Work with neat technology. 10

11 Why Study Artificial Intelligence? Science Search engines Labor Appliances Medicine/ Diagnosis What else? 11

12 Why Study Artificial Intelligence? Engineering: To get machines to do a wider variety of useful things e.g., understand spoken natural language, recognize individual people in visual scenes, find the best travel plan for your vacation, etc. Cognitive Science: As a way to understand how natural minds and mental phenomena work e.g., visual perception, memory, learning, language, etc. Philosophy: As a way to explore some basic and interesting (and important) philosophical questions e.g., what is consciousness, etc. 12

13 Why Study Artificial Intelligence? World wide web (content, commerce, community, context) Web 1.0 World of static pages. (Read Only) Web 2.0 Dynamic and integrated with media like sound and video. (Read and Write) Web 3.0 Intelligence driven + Web AI. (Read Write and execute) Web 3.0 = AI + Web 2.0 add a layer of meaning on top of the existing Web that would make it less of a catalog and more of a guide provide the foundation for systems that can reason in a human fashion 13

14 Outline Course information Motivations What is Artificial Intelligence A brief history of Artificial Intelligence Outline of the course 14

15 A paper by Turing Computing Machinery and Intelligence Published in Mind: A Quarterly Review of Psychology and Philosophy, in I propose to consider the question, Can machines think? And if so, how? And if not, why not? 15

16 What is Artificial Intelligence? A more difficult question is: What is intelligence? This question has puzzled philosophers, biologists and psychologists for centuries. Intelligence has many faces, like creativity, solving problems, pattern recognition, classification, learning, induction, deduction, building analogies, optimization, surviving in an environment, language processing, knowledge, and many more. Artificial Intelligence is easier to define, although there is no standard, accepted definition. (next slide) 16

17 What is Artificial Intelligence? A simple definition might be as follows: Artificial intelligence is the study of systems that act in a way that to any observer would appear to be intelligent. One definition of artificial intelligence is offered by the AAAI -- "the scientific understanding of the mechanisms underlying thought and intelligent behavior and their embodiment in machines". ( ) In fact, Artificial Intelligence techniques are often used to solve complex problems. This may lead us to another definition of Artificial Intelligence, as follows: Artificial Intelligence involves using methods based on the intelligent behavior of humans and other animals to solve complex problems. 17

18 Task domains of AI Mundane tasks Formal tasks Expert tasks 18

19 Mundane Tasks Perception Vision Speech Natural Language Understanding Generation Translation Commonsense Reasoning 19

20 Formal Tasks Games Chess Checkers Othello Go Mathematics Logic Geometry Calculus Proving properties of programs Checkers is solved (Science) 20

21 Expert Tasks Engineering Design Fault Finding Manufacturing planning Medical Diagnosis Medical Image Analysis Financial Stock market predictions 21

22 Key attributes AI differs from more conventional computer science areas in: Viewpoint (plausible and logic reasoning instead of qualitative calculations) Symbolic manipulation (instead of numeric orientation) Knowledge-based design (has a line of reasoning & can explain itself) Heuristic or approximate problem-solving approach 22

23 Key attributes AI encompasses many different areas such as: Expert systems Game playing Automatic programming Deduction and theorem proving Knowledge representation Search Learning Planning Natural language processing Robotics Vision and Scene understanding 23

24 Strong AI and Weak AI There are two entirely different schools of Artificial Intelligence: Strong AI: This is the view that a sufficiently programmed computer would actually be intelligent and would think in the same way that a human does. Weak AI: This is the use of methods modeled on intelligent behavior to make computers more efficient at solving problems. This course is concerned with Weak AI. Strong AI is currently the stuff of science fiction, although there are many that believe that machines will indeed be capable of real thought at some point in the future. 24

25 Strong Methods and Weak Methods Not to be confused with Strong AI and Weak AI. Strong methods use knowledge about the world to solve problems. Weak methods use logic and other symbolic systems. Strong method systems rely on weak methods, as knowledge is useless without a way to handle that knowledge. Weak methods are in no way inferior to strong methods they simply do not employ world knowledge. 25

26 Alan Turing and the 1950s Alan Turing is often seen as the father of Artificial Intelligence. He invented the Turing Test, designed to determine if a computer system can be called an artificial intelligence or not, based on whether it can fool a human into thinking it is human too. No system has yet passed the Turing Test. We will discuss the Turing test in Lecture 2. Around this time, in the 1950 s, systems were being developed that could play checkers, engage in conversation and solve other problems. The term Artificial Intelligence was coined in 1956 by John McCarthy. 26

27 Outline Course information Motivations What is Artificial Intelligence A brief history of Artificial Intelligence Outline of the course 27

28 AI History pre-turing Test Greek mythology: Hephaestus, idea of intelligent robots. 15th century: da Vinci drafted robot design 16th century: the Maharal of Prague s Golem 17th century: Descartes animals are complex machines 19th century: Charles Babbage s Analytical Engine 1940 s: Isaac Asimov Three Laws of Robotics : McCulloch and Pitts, model neurons 28

29 AI history The dawn 1950 Allen Turing invented Turing test Newell and Simon s Logic Theorist (LT) found a shorter proof of a well known theorem in Principia Mathematica 1956 Dartmouth conference: the term artificial intelligence was coined by John McCarthy 29

30 AI history The romantic period Early enthusiasm, great expectations 1956 Samuel s checker player was demonstrated on TV, creating strong impression 1958 In MIT AI Lab memo no.1, McCarthy invented LISP 1959 Gelernter constructed the Geometry Theorem Prover 1965 Robinson discovered the resolution method Continued in the next slide. 30

31 AI history The romantic period 1963 Slagle s SAINT program (closed form integration problems) 1967 Bobrow s STUDENT program (algebra story problems) 1968 Evan s ANALOGY (geometric analogy problems) 1968 Raphael s SIR (semantic information retrieval, questions and answers in restricted subset of English) Continued in the next slide. 31

32 AI history The romantic period Early successes in many areas: Natural language STUDENT (1967), SHRDLU (1972) Knowledge representation semantic network (1968), frame (1975), conceptual dependencies and script (1977), production systems Search techniques (heuristics) Vision constraint propagation Neural network and learning Planning Fahlman s planner (1974) 32

33 AI history a dose of reality Big promises, great expectations, all fell short Methods that worked well for simple examples turned to fail on wider applications and more difficult problem. Why? Early programs were knowledge poor (the best known example is ELIZA program) Combinatorial explosion and intractability ( scaling up was once widely thought as simply a matter of faster hardware and larger memories) The fact that a program can find a solution in principle does not mean that the program contains any of the mechanisms needed to find it in practice 33

34 AI history awakening Old methods of relying on general-purpose search mechanism have been called weak methods. The new approach: knowledge is power! The expert system era: MYCIN PROSPECTOR (1979) successfully found a site containing a large molybdenum deposit New area of research: knowledge engineering, uncertainty reasoning 34

35 AI history booming industry 1982 R1, the first successful commercial ES that saved DEC about $40 million a year 1981 Japan launched the 5th generation project to build intelligent computers MCC was found in the US to counter the Japanese project Funding reinstated LMI, TI, Symbolics, and Xerox built LISP machines Nearly every major US corporation had its own AI group The AI industry went from a few million in sales in 1980 to $2 billion in

36 AI history crude reality, again Many corporations and research groups found that building a successful expert system involved much more than simply buying a reasoning system and filling it with rules Support for AI research and business quickly shrinking LMI (Lisp Machines Inc.) declared bankruptcy th generation project declared a failure 1993 Symbolics declared bankruptcy Another AI winter 36

37 AI history - recap Continued in the next slide. 37

38 AI history - recap The 1950s to the 1990s During this time, the optimism of the 1950s was replaced with realism. Artificial Intelligence replaced as its goal the building of an intelligent robot with the goal of using heuristics and other techniques to solve complex problems. 38

39 AI history late 90 s The Deep Blue chess program beats the current world chess champion, Garry Kasparov. Web crawlers and other AI-based information extraction programs become essential in widespread use of the world-wide-web. 39

40 AI history 21 st century 200x - Interactive robot pets (a.k.a. "smart toys") become commercially available. Sony AIBO (Artificial Intelligence robot) PLEOrb (a dinosaur robot ) 2011 IBM Watson 40

41 AI history current status See AI in the news for history in the making! 41

42 AI history current status The dream is still there, but no more unrealistic expectations Return to well founded and existing theories Hardware improvement improves speech recognition, chess programs, computer vision, and robotics More work on real-world problems rather than microworlds AI techniques will proved to be viable in well defined and limited area. Domain knowledge is the key. 42

43 Outlook for AI AI has a long way to go. It may well be 100 years before any computer is as intelligent as a human. No single discovery will suffice. Real AI will require hundreds of research breakthrough. AI will continue to advance. Progress will continue but no sudden and final success Useful applications of AI are feasible now. AI can enhance human performance. The best overall system will often be a combination of AI programs and humans. 43

44 AI in the 21 st Century AI is everywhere. Fuzzy logic is used in elevators, washing machines and cars. Intelligent agents are used in many software applications. Robots explore other worlds, and toy robots play with children (and some adults). Expert systems diagnose diseases and recommend remedies. Computer games use AI. 44

45 Outline Course information Motivations What is Artificial Intelligence A brief history of Artificial Intelligence Outline of the course 45

46 Outline of the course Part 1: Introduction to Artificial Intelligence Chapter 1: A Brief History of AI Chapter 2: Artificial Intelligence: Uses and Limitations Chapter 3: Knowledge Representation Lisp Part 2: Search Chapter 4: Search Methodologies Chapter 5: Advanced Search Chapter 6: Game Playing Part 3: Logic Chapter 7: Propositional and Predicate Logic Chapter 8: Inference and Resolution for Problem Solving Chapter 9: Rules and Expert Systems Midterm Exam Continued in the next slide. 46

47 Outline of the course (Cont.) Part 4: Machine Learning Chapter 10: Introduction to Machine Learning Chapter 11: Neural Networks Chapter 12: Probabilistic Reasoning and Bayesian Belief Networks Chapter 13: Artificial Life: Learning through Emergent Behavior Chapter 14: Genetic Algorithms Part 5: Planning Chapter 15: Introduction to Planning Chapter 16: Planning Methods Part 6: Advanced Topics Chapter 17: Advanced Knowledge Representation Chapter 18: Fuzzy Reasoning Chapter 19: Intelligent Agents Chapter 20: Understanding Language Chapter 21: Machine Vision If time permitted Final Exam 47

48 Questions? 48