Artificial Intelligence An Introduction by ojf14405

VIEWS: 28 PAGES: 52

									Artificial Intelligence : An Introduction

       for CS570 Artificial Intelligence




            Jin Hyung Kim
      KAIST Computer Science Dept.
                 Definition of AI
   Automation of activities that we associate with human
    thinking, activities such as decision-making, problem
    solving, learning, … (Bellman, 1978)
   Study of how to make computers do things at which, at
    the moment, people are better (Rich & Knight, 1991)
   A Branch of Computer Science that is concerned with
    the automation of intelligent behavior (Luger &
    Stubblefield, 1993)
   The study of mental faculties through the use of
    computational models (Charniak & McDormott, 1995)
    Computer Science Body of Knowledge
                                                              Discrete Structure

                                                              Foundat'n of Progamming

                      31                                      Algorithm and Complexity
                                       43
            16                                                Architecture & Organization

                                                              OS
        10
                                                              Net-Centric Computing
       10              Total 280                  38
   3                                                          Programming Languages
        8              Core Hours
                                                              HCI

         21                                  31               Grophics & Visual Computing

                                                              Intelligent Systems
              15
                      18                                      Information Management
                                  36
                                                              Social and Professional
                                                              Issues
Source : IEEE/ACM Computing Curricula 2001 Computer Science   S/W Engineering
Computing Disciplines, before and after 1990s
        AI : Engineering Definition
   Study of how to make machine do things which require
    intelligence when human do
     things requiring intelligence ?


   Making computer MORE smart

   Making thinking computer
     Can machine think ?


   Focus on how good it performs
 AI : Cognitive Scientific Definition
 Studying   intelligence by computational means

 Programmed Human intelligence
 Artificial Mind


   Focus on how similar it works as human
         Intelligent System



                             Perception,
        Flexibility         Recognition,
Aims   Automation     via  Understanding
       Optimization       Making decisions,
                               Acting
          Examples of AI systems
   Language Translation systems
   Natural Language Question answering systems
   Diagnosis Expert systems
   Avionic Expert systems vs. fly-by-wire
   Space shuttle mission planning
   Robots in factory, Auto-navigation robots
   Intelligent Traffic control system
   OCR, Handwriting Recognition System
   Speech Recognition System
   …
 Categorization of AI definitions

Systems that think like       Systems that think
       human                      rationally

 Systems that act like    Systems that act rationally
       human
                   Go Playing Programs
   Selecting next move
      By analysis of all alternative moves
      By Analysis of Board Pattern (rule-based)

   Which one is better ?
   Which one can be better ?

   Engineering (mathematical)
        How well does it perform ?
        Performance is the key concern. Don’t care of what method used

   Cognitive Scientific
        How similarly does it do as human ?
        Simulation of Behavior
      Acting Humanly : Turing Test
   Turing (1950) “Computing Machinery and
    Intelligence”
     Can machine think ?  Can machine behave
      intelligently ?
     Operational test of intelligent behavior : imitation
      game
   Predicted that by 2000, a machine might
    have 30% chance of fooling a lay person in 5
    minutes
Imitation Game
               Issues on Turing Test
   Intelligent as much as Human
     Is dog intelligent ?
   Searle’s Chinese Room argument
     Strong AI and Weak AI
   “ELIZA - a friend you could never have before”
     http://www-ai.ijs.si/eliza-cgi-bin/eliza_script
     Imitation of Client-centered Rogerian Therapy

   Suggested major component of AI : knowledge,
    reasoning, language, understanding, learning

   Any man-made system passed Turing Test ?
Thinking Rationally : Laws of Thought

   Normative (or prescriptive ) rather than descriptive
   Several school of Greek schools developed various
    forms of logic, notation and rules of derivation of
    thoughts
    Mathematics and Philosophies of modern AI
   Problems
     Not all intelligent behavior is mediated by logical deliberation
     What is purpose of thinking ? What thought should I have ?
   Rational Behavior : doing the right thing
     “right” – expected to maximize goal achievement given
      available information
Hype Cycle (Boom-Bust-Build)



            Science
            Fiction
                      Hangover
                                 Productivity



Curiosity
The Hype Cycle of Emerging Technologies




         ※ 자료 : Gartner, 2002




                                 ※ 자료 : Gartner, 2002
Approaches to Intelligent system
        development
               
   Knowledge-based Approach
               
      Data Driven Approach
               
    Knowledge-base Systems
                   
 Represent Human knowledge as symbol
              combination
                   
Knowledge Acquisition and Representation
                   
   Logic, Expert System, Fuzzy Logic
                   
       Data Driven Approach
                      
Extract common characteristics from collected
                 examples
                      
                  Training
                      
Statistical Methods, Artificial Neural Network
            Generality vs Power
 Aims Powerful and general solutions
 General Problem Solver
     Early attempt : failed
     Complexity : Toy Problems Only
   Specialized Approach to get Power
     Knowledge Based Approach
     “Practical” Expert Systems
                     State of the Art
   Which of the following can be done at present ?
       Play a decent game of table tennis
       Drive along a curvy mountain road
       Drive in the center of Seoul city
       Play decent game of Go
       Discover and prove a new mathematical theorem
       Write an intentionally funny story
       Give competent legal advice in a specialized area of law
       Translate spoken Korean into spoken Japanese in real time
    Axes of AI Research

           Theory




                       Methodology


System

                    Application
     Major research areas (Methodology)
 Symbolic   Programming
 Knowledge Representation
 Search & Planning
 Automated Reasoning
 Machine Learning, knowledge Discovery
 Artificial Neural Net
 Genetic Algorithm
 …...
     Major research areas (Applications)
 Natural Language Understanding
 Image, Speech and pattern recognition
 Uncertainty Modeling
 Expert systems
 Virtual Reality
 …..
          Symbolic Programming
 Program as Representation of world
 Symbol as basic element of representation
     atom, property, relationship
 Symbolic Expression as method of combination
 LISP for Symbolic programming
 PROLOG for logic programming
 Object-Oriented Concept
        Knowledge Representation
 What  kind of Knowledge needed for Problem
  solving ?
 Structure of knowledge ?
     declarative vs procedural
   Representation techniques ?
     explicit vs (implicit + inference)
     logic, frame, object-oriented, semantic net, script
 Knowledge      acquisition and update
                 Search Theory
 An Optimization method
 Analyze alternative cases and select one
 Cope with Exponential complexity, NP classes
     Try likely one first (Heuristic Search)
     Utilize local information (Hill Climbing Method)
     Optimal solution vs good solution
   Genetic Algorithm, Simulated Annealing
     Stochastic search
            Automated Reasoning
 Qualitative    Reasoning
     Utilization of qualitative knowledge such as
   Non-monotonic Reasoning
     Ostrich flys ?
   Plausible Reasoning
     Information fusion under uncertainty
   Case-based Reasoning
     Utilization of Experience
                 Machine Learning
   Performance improvement by experience
     How much of knowledge required to start learning ?
     Method of acquiring new knowledge and merging it to existing
      knowledge-base
     Role of teacher
     Role of examples and experience
   Parameter Adjustment
   Inductive learning
   Computational Learning Theory
     Quality of generalization capability in terms of Training data
   Used in Practice such as Data Mining
                      Data Mining
Knowlegre extraction for decision making
       Data                Information         Decision
                           / knowledge         Making
     인구통계                A상품 구매자의           광고전략은 ?
     Point of Sale        80%가 B상품도 구
                                              상품의 진열
     ATM                  매한다
                                              최적의 예산 할당
     금융통계                미국시장의 자동차
                                               은?
     신용정보                 구매력이 6개월간
                           감소                 시장점유의 확대
     문헌                                       방안은 ?
                          A상품의 매출 증가
     첩보자료                                    고객의 이탈 방지
                           가 B상품의 2배
     진료기록                                     책은 ?
                          탈수 증상을 보이면
     신체검사기록               위험                 처방은 ?
              Neural Network
 Computational   model of Neurons
   Power comes from Connection of simple processing
    element - connectionism



       X1
             w1

       X2    w2
                    S              F(X1, X2, …, Xn)
       .
       .     wn
       .
       Xn
                Neural Network
   learning = link weigh adjustment
     Error-back-propagation : supervised learning
     Any Functional Mapping is learnable
   Strong at Sensory Data Processing
     Symbolic Grounding


   Old Horse on the race again
     Massive parallelism, graceful degradation
      Neural Network Classifier
Job(1/0)

                                   good
      age


    Salary                         medium


 #mouth                            bad


   Debt

             Input   Hidden   Output
             layer    layer    layer
               Genetic Algorithm
 Computational   model of life evolution
 Stochastic optimization technique
     Initial chromosome creation
     New generations are made (cross over, mutation)
     survival of the fittest
   Base of artificial life research
     study evolution of life, by simulation
                History of AI
 50years of rise and fall of New technologies
 after invention of computer

   – Logic                   – Fuzzy Theory
   – Optimization            – Neural Netwrok
   – Proabilistic Modeling   – Genetic Algorithm
   – Search theory           – Chaos theory
   – Rule-based system       – Artificial life
   – Expert systems          – .....
                        AI Prehistory
   Philosophy
      Logic, methods of reasoning, mind as physical system, foundations of
       learning, language, rationality
   Mathematics
      Formal representation of proof, algorithms, computation, decidability,
       tractability, probability
   Psychology
      Adaption, phenomena of perception and motor control, experimental
       techniques
   Linguistics
      Knowledge representation, grammar
   Neurosicence : Physical substrate for mental activity
   Control Theory : homeostatic systems, stability, optimal designs
           Potted History of AI (I)
   1943 : McCulloch & Pitts : Boolean Circuit model of
    Brain
   1950 : Turing’s “Computing Machinery and
    Intelligence”
   1950s : Early AI programs – Samuel’s checker
    program, Newell & Simon’s Logic Theorist
   1956 : Dartmouth meeting “Artificial Intelligence”
    adopted
   1965 : Robinson’s algorithm for logical reasoning
            Potted History of AI (II)
   1966-74 : AI discovers computational complexity
   1969-79 : Early development of knowledge-based systems
   1980-88 : Expert systems industry booms, AI Programming Machine
   1983 – 1993 : Japan initiated 5th generation computer project
   1988-93 : Expert systems industry burst : “AI Winter”
   1985-95 : Neural Network back to the race
   1988 : Resurgence of probabilistic and decision-theoretic methods,
         Rapid increase of technical depth of mainstrean AI
         “Nouville AI : Alife, Genetic Algorithm, Soft computing
                  AI Success Story

 Evans ANOLOGY
 Symbolic Algebra
     Macsyma (http://www.macsyma.com/)

            x4                         1 3
        (1  x 2 ) 5 2 dx  arcsin x  tan (arcsin x)  tan(arcsin x)3
                                       3

 Chess Program DEEP BLUE defeat Gary Kasparov
  (1996)
 Automatic Theorem Proving contest (1999)
AI Success Story (Planning)
       MARVEL (Schwuttke, 1992)
         Real-time Space shuttle Mission planning
       Berth assignment (KAL, 1997)
       Unmanned Vehicle
         Ground and air
       Pathfinder Rover, 1996
       Asimo – a walking robot
Autonomous Land Vehicle
 (DARPA’s GrandChallenge contest)
AI Success Story (Language Processing)

 PEGASUS (Zue, 1994)
    Spoken Natural language for airline reservation
    Limited context, free representation
 Japanese-Korea     Hotel reservation(KT, 1995)
 Chatter   Bot
    자연언어로 대화 (typing)하는 회사소개 에이젼트 등
 Many   machine translation
    일한 실용화 완료, 영한 - 시제품
AI Success Story : Medical expert systems
               Programs listed by Special Field
   Antibiotics & Infectious       Gynecology
    Diseases                       Imaging Analysis
   Cancer
                                   Internal Medicine
   Chest pain
   Dentistry                      Intensive Care
   Dermatology                    Laboratory Systems
   Drugs &                        Orthopedics
    Toxicology
                                   Pediatrics
   Emergency
   Epilepsy                       Pulmonology & Ventilation
   Family Practice                Surgery & Post-Operative
   Genetics                        Care
   Geriatrics                     Trauma Management
    Pattern Recognition Applications
   Handwriting and document recognition
     forms, postal mail, historic documents
     PDA pen recognition
 Signature, biometrics (finger, face, iris, etc.)
 Gesture, facial expression
     As a Human computer intertraction
 EEG, EKG, X-ray
 Trafic monitoring, Remote Sensing
 Smart Weapon – guided missile, target homing
Automatic Target Recognizer
Postal Address Recognition
Handwriting Understanding




         전자 펜으로 수식 입력       수식 인식
次世代 PC : e-Book, Tablet PC, PDA, M-phone
Ubiquitous
 전자교실
BioInformatics / Protein Structure Analysis
              Contribution of AI
 PracticalAI OCR, ICR, Symbolic Algebra,
  Machine Translation, Many Expert systems,
  Planning systems
 New concepts and Ideas to other fields of
  computer sciences
   Programming Language – OO, functional language,
    logic-based
   DataBase
   Operating System
                   Future of AI
   Making AI Easy to use
     Easy-to-use Expert system building tools
     Web auto translation system
     Recognition-based Interface Packages
   Integrated Paradigm
     Symbolic Processing + Neural Processing
   AI in everywhere, AI in nowhere
     AI embedded in all products
     Ubiquitous Computing, Pervasive Computing

								
To top