Bring Class List
-----------------------------------------------------------------
* Ask about emails
* Ask about text book
* Who did read Chapter 1?
-----------------------------------------------------------------
GOAL of AI -
    understand and build intelligent entities
    (how it work? how can it be built?)

Human-level AI: Build machines/computer systems that behave/act
like human (John McCarthy)

What a five years old kid can do?
    - if he/she falls, it will get up
that is not easy for a robot to do?
    - NASA Mars-Explorer falls into a hole and cannot
    get out

What can human do but cannot yet make the computers do?
    - translation from English into XXX-language
    - reacting in unforeseenbar situations
    - ....
-----------------------------------------------------------------
Why it is interesting?
    - it is strongly believed that this is a possible quest
    - it is a new field with a lot of opportunities
    - it inherits many ideas, viewpoints, and techniques from
    very old fields such as philosophy, mathematics, linguistic,
    psychology, ....
      POSSIBLE ONLY WITH COMPUTERS
    - it is diversified
-----------------------------------------------------------------
1. What is AI?

    Characterizations:
        - thinking and reasoning vs.
        - behavior

    Answer: There are too many answers (see the book)
    * acting humanly        - Turing Test approach
    - thinking humanly      - Cognitive model approach
    - think rationally      - Laws of thought approach
    * acting rationally     - Rational agent approach

-----------------------------------------------------------------
A: Acting humanly: Machine needs to act like human

Needs to have a method of checking when something is 'intelligent'

a. Turing Test
    The setting:

                        |
       computer    <----|----> interrogator
                        |
                    teletype

    Pass if the interrogator cannot tell if it were a human or
    a machine is at the other end

    A number of sub-fields
        - Natural Language Processing
        * Knowledge Representation
        * automated reasoning
        - machine learning

b. Total Turing test
    includes video signal for the interrogator to test the perceptual
    ability of the subject at the other end
        - machine vision
        - robotics
-----------------------------------------------------------------

B. Thinking humanly: Machine needs to follow the thought process of
human in order to get to the right result vs. gets to the right result

a. Must learn how the brain works?
    Different emphases: how a result is obtained vs. getting the right answer

--> cognitive science: computer models from AI and experimental techniques
from psychology to con

It is a DIFFERENT FIELD now - not much related to AI !!!

-----------------------------------------------------------------

C. Thinking rationally: machine must get to irrefutable arguments
(reasoning process) - machine must always do correct reasoning

    Logic needed

-----------------------------------------------------------------

D. Acting rationally: Acting to achieve a goal given one's belief
    - agent: perceives and acts
    - correct reasoning is part of acting rationally
    - cognitive skills are needed: represent knowledge, reason with it
    (to arrive at a good decision)

    + More general than the law of thought approach
    + More amenable to the scientific development

-----------------------------------------------------------------
4. Foundations of AI

    - Philosophy
    - Mathematics
    - Psychology
    - Computer engineering
    - Linguistic
-----------------------------------------------------------------
5. A little history of AI

    - the first work on AI (Warren McCulloch & Walter Pitts, 1943) - neural network
    - 1950 (Claude Shannon) and 1953 (Turing) - chess program for computer (von Neuman
    style)
    - 1951 first neural network computer (Marvin Minsky and Dean Edmonds)
      called SNARC with 3000 vacuum tubes & a surplus automatic pilot mechanism
      from a B-2 bomber to simulate a network of 40 neurons
    - 1956 McCarthy organized the Dartmouth workshop (Study of intelligence) -
        the birth of AI
        Allen Newell and Herbert Simon (CMU) - reasoning program (the Logic Theorist)
        that can prove logic theorems
    - 1952 - 1969: many programs
        + General Problem Solver (thinking humanly approach - imitate human thinking)
          Allen Newell and Herbert Simon
        + Geometry Theorem Prover
          Herbert Geernter (Nathaniel Rochester) --> search
        + Checker
           (from 1952) Arthur Samuel
        + LISP
          (John McCarthy, 1958)
        + Advice Taker (hypothetical system viewed as the first complete AI system) (*)
          (John McCarthy, 1958)
        + Resolution method for first order logic   (**)
          (J. A. Robinson, 1963)
        + planning system                           (**)
          (Cordell Green, 1969)
        + Minsky supervised many students try to solve problems that seem to require
          intelligence: intergration (first-year college calculus), geometric analogy
          to those appear in IG test, vision, query-answering in natural language,
          etc.
    - 1966 -1974  Down to earth!
        + no knowledge about the subject is represented
        + intracability
        + unreasonable assumptions
    - 1969 - 1979 Expert systems
        + MYCIN
    - 1980 - 1988 Industry
        + First commercial AI system (Configuring orders for new systems)
        + Japanese Fifth Generation project
        + MCC project (CYC)
    - 1986 - present
        + neural network returns
        + excitement about expert systems decreases
    - 1987 - present
        + progress in speech recognition
        + stress on theoretical foundations and implementations