Artificial Intelligence - Introduction by ecg16852


									Artificial Intelligence -

                      Mahdi Shafiei
              Faculty of Computer Science
                 Dalhousie University

           Acknowledgement: Based on the slides
           provided by Russell and Norvig
Course Information
   Course homepage:

   Check out course homepage for schedule, lecture notes, tutorials,
    assignment, grading, office hours, etc.
   Location and times
     –   Mostly Teaching Lab 2, please check out the summer course
         schedule for the class location
     –   Mondays, Tuesdays, Wednesdays and Thursdays, 10:00-11:30am
         and 13:00-14:30pm.
   Main text: David MacKay, Information Theory, Inference, and
    Learning Algorithms, 2003
   Grading: Class participation (10%), 4 assignments (40%),
   CSCI-4150
     –   Course Project (25%), Final exam – August 8, (25%)
   CSCI-2140
     –   Final exam – August 4, (50%)
Course Overview

   Probability concepts
    –   Random variables
    –   Information theory
    –   Representation of Uncertainty
   Probabilistic reasoning
   Bayesian networks
Why study AI?

                          Search engines

            Appliances        What else?
What is AI?
   Views of AI fall into four categories:

         Thinking humanly            Thinking rationally

           Acting humanly             Acting rationally
Acting humanly: Turing Test
   Alan Turing's 1950 article Computing Machinery and
    Intelligence discussed conditions for considering a
    machine to be intelligent
     – ―Can machines think?‖  ―Can machines behave
     – The Turing test (The Imitation Game): Operational
       definition of intelligence.
Acting humanly: Turing Test

• Computer needs to possess: Natural language processing,
  Knowledge representation, Automated reasoning, and Machine

• Are there any problems/limitations to the Turing Test?
What tasks require AI?

   ―AI is the science and engineering of making
    intelligent machines which can perform tasks
    that require intelligence when performed by
    humans …‖

 What      tasks require AI?
What tasks require AI?

   Tasks that require AI:
    –   Solving a differential equation
    –   Brain surgery
    –   Inventing stuff
    –   Playing Chess
    –   What about walking?
    –   What about grabbing stuff?
    –   What about pulling your hand away from fire?
    –   What about watching TV?
    –   What about day dreaming?
What would a computer need to pass the Turing test?

   Natural language processing: to communicate with
   Knowledge representation: to store and retrieve
    information provided before or during interrogation.
   Automated reasoning: to use the stored information
    to answer questions and to draw new conclusions.
   Machine learning: to adapt to new circumstances
    and to detect and extrapolate patterns.
Acting humanly: Turing Test
   Turing (1950) "Computing machinery and intelligence":
   "Can machines think?"  "Can machines behave intelligently?"
   Operational test for intelligent behavior: the Imitation Game

   Predicted that by 2000, a machine might have a 30% chance of
    fooling a lay person for 5 minutes
   Anticipated all major arguments against AI in following 50 years
   Suggested major components of AI: knowledge, reasoning,
    language understanding, learning
   ―artificial flight‖: imitating birds or learning about aerodynamics
Thinking Humanly: Cognitive Science

   1960 ―Cognitive Revolution‖: information-processing
    psychology replaced behaviorism
   Cognitive science brings together theories and
    experimental evidence to model internal activities of the
     –   What level of abstraction? ―Knowledge‖ or ―Circuits‖?
     –   How to validate models?
          Predicting and testing behavior of human subjects (top-down)
          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 (~ 450 B.C.) attempted to codify ―right
    What are correct arguments/thought processes?

   E.g., ―Socrates is a man, all men are mortal;
    therefore Socrates is mortal‖
   Initiation of the Logic

   Several Greek schools developed various forms of
    notation plus rules of derivation for thoughts.
Thinking Rationally: Laws of Thought

 Problems:
  1) Uncertainty: Not all facts are certain
   (e.g., the flight might be delayed).

  2) Resource   limitations:
    -   Not enough time to compute/process
    -   Insufficient memory/disk/etc
    -   Etc.
Acting Rationally: The Rational Agent
   Computer Agent
     –    Autonomous control, perceiving the environment, adopting to change
          and …
   Rational Agent
     –    Acts so as to achieve the best (expected) outcome
   Rational behavior: Doing the right thing!
   The right thing: That which is expected to maximize the expected
   Provides the most general view of AI because it includes:
     –    Correct inference (―Laws of thought‖)
     –    Uncertainty handling
     –    Resource limitation considerations (e.g., reflex vs. deliberation)
     –    Cognitive skills (NLP, knowledge representation, ML, etc.)
   Advantages:
     1)   More general
     2)   Its goal of rationality is well defined
AI prehistory

   Philosophy       Logic, methods of reasoning,
                     mind as physical system
                     foundations of learning, language, rationality
   Mathematics      Formal representation and proof algorithms,
                     computation, (un)decidability, (in)tractability,
   Economics        utility, decision theory
   Neuroscience     physical substrate for mental activity
   Psychology       phenomena of perception and motor control,
                     experimental techniques
   Computer         building fast computers
   Control theory   design systems that maximize an objective
                     function over time
   Linguistics      knowledge representation, grammar
State of the art
   Game Playing: Deep Blue defeated the reigning world chess
    champion Garry Kasparov in 1997
   Theorem Proving: Proved a mathematical conjecture (Robbins
    conjecture) unsolved for decades
   Autonomous Control: No hands across America (driving
    autonomously 98% of the time from Pittsburgh to San Diego)
   Logistics Planning: 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
   Autonomous planning and scheduling: NASA's on-board
    autonomous planning program controlled the scheduling of
    operations for a spacecraft
   Language Understanding and Problem Solving: Proverb solves
    crossword puzzles better than most humans
   Robotics: robot assistance for microsurgery
State of the art
   Which of the following can be done at present?
    –   Play a decent game of table tennis
    –   Drive along a curving mountain road
    –   Drive in the center of Cairo
    –   Buy a week’s worth of groceries at Sobey’s
    –   Buy a week’s worth of groceries on the Web
    –   Play a decent game of bridge
    –   Discover and prove a new mathematical theorem
    –   Write an intentionally funny story
    –   Give competent legal advice in a specialized area of law
    –   Translate spoken English into spoken Swedish in real time
    –   Perform a complex surgical operation
Statistical Artificial Intelligence

   Principles, methods, and algorithms for
    learning and prediction on the basis of past
   Already everywhere:
    –   speech recognition
    –   hand-written character recognition
    –   information retrieval
    –   operating systems, compilers
    –   fraud detection, security, defense applications
    –    ...

   A classification problem:
    –    predict the grades for students taking this course
       Key steps:
    1.   Data
    2.   Assumptions
    3.   Representation
    4.   Estimation
    5.   Evaluation
    6.   Model selection
   A classification problem:
    –     predict the grades for students taking this course
       Key steps:
    1.    data: what ―past experience‖ can we rely on?
    2.    assumptions: what can we assume about the students or
          the course?
    3.    representation: how do we ―summarize‖ a student?
    4.    estimation: how do we construct a map from students to
    5.    evaluation: how well are we predicting?
    6.    model selection: perhaps we can do even better?

   The data we have available (in principle):
    –   names and grades of students in past years ML
    –   academic record of past and current students
   ―training data‖

   ―test data‖

   There are many assumptions we can make
    to facilitate predictions
    –   the course has remained roughly the same over
        the years
    –   each student performs independently from others
   Academic records are rather diverse so we might limit
    the summaries to a select few courses
    For example, we can summarize the      student (say
    Peter) with a vector

   where the grades correspond to (say) 1806, 6041, and
   The available data in this representation
   Given the training data

    we need to find a mapping from ―input vectors‖ x to
    ―labels‖ y encoding the grades for the ML course.
   Possible solution (nearest neighbor classier):
    1.   For any student x find the ―closest‖ student xi in the training
    2.   Predict yi, the grade of the closest student

   How can we tell how good our predictions
    –   we can wait till the end of this course...
    –   we can try to assess the accuracy based on the
        data we already have (training data)
   Possible solution:
    –   divide the training set further into training and test
    –   evaluate the classifier constructed on the basis of
        only the smaller training set on the new test set
Model selection

   We can refine
    –   the estimation algorithm (e.g., using a classier
        other than the nearest neighbor classier)
    –   the representation (e.g., base the summaries on a
        different set of courses)
    –   the assumptions (e.g., perhaps students work in
        groups) etc.
    We have to rely on the method of evaluating
    the accuracy of our predictions to select
    among the possible refinements
Types of learning problems
A rough (and somewhat outdated) classification of learning

   Supervised learning, where we get a set of training inputs and
     –   classification, regression

    Unsupervised learning, where we are interested in capturing
    inherent organization in the data
     –   clustering, density estimation

    Reinforcement learning, where we only get feedback in the
    form of how well we are doing (not what we should be doing)
     –   planning
Supervised learning : classification

   Example: digit recognition
    (8x8 binary digits)

   We wish to learn the
    mapping from digits to
Representations / Assumptions
Supervised Learning: Regression
Unsupervised learning: data organization

   The digits again

   We'd like to understand the generation process of examples
    (digits in this case)
Abridged history of AI
   1943      McCulloch & Pitts: Boolean circuit model of brain
   1950      Turing's "Computing Machinery and Intelligence"
   1956      Dartmouth meeting: "Artificial Intelligence" adopted
   1952—69   Look, Ma, no hands!
   1950s     Early AI programs, including Samuel's checkers
              program, Newell & Simon's Logic Theorist,
              Gelernter's Geometry Engine
   1965      Robinson's complete algorithm for logical reasoning
   1966—73   AI discovers computational complexity
              Neural network research almost disappears
   1969—79   Early development of knowledge-based systems
   1980--    AI becomes an industry
   1986--    Neural networks return to popularity
   1987--    AI becomes a science
   1995--    The emergence of intelligent agents

To top