ARTIFICIAL INTELLIGENCE

1.  What Is Artificial Intelligence?

a)    If the basic question of computer science is:

       *    What problems are computable?

      then the basic question of artificial intelligence (AI) is:

       *    Is "intelligence"/cognition/thinking computable?

(where by "intelligence" I am not necessarily referring to whatever it is that is alleged to be measured by IQ tests, but simply the kinds of mental abilities that cognitive psychologists study:  what they call "cognition", and what most people call "thinking").

b)    Given our definitions of "computable" and "algorithm", the basic question of AI becomes:

        *    Is there an algorithm (or a collection of them) that computes (human) cognitive processes?

c)    AI is the branch of computer science that investigates this question.

2.    Other Views on What A.I. Is

a)    According to Marvin Minsky, one of the founders of AI,

        "[AI is] the science of making machines do things that would require intelligence if done by humans."

    *    Note that this uses humans to tell us how to program computers.

b)    According to Margaret Boden, a psychologist,

        "[AI is] the use of computer programs and programming techniques to cast light on the principles of intelligence in general and human thought in particular."

    *    Note that this uses computers to tell us something about humans. (Just as, in other sciences, theories are expressed in the languages of English, or mathematics, or statistics, so, in AI, theories about cognition can be expressed in the languages of computer programs.)

c)    In fact, AI is both of the above, a 2-way street.
 

3.  A Side-Effect of (2b):

There is an interesting consequence of definition (2a):

If (human) cognitive processes can be expressed as algorithms,
  then they are capable of being implemented in (non-human) computers.

So:

Are computers executing such algorithms merely simulating cognitive processes, or are they actually exhibiting them?

One answer was given by Alan Turing's "Turing Test".

An objection to that answer was given by John Searle's "Chinese-Room Argument"
 

4.  The Turing Test

a)    For Turing's original paper, see:

Turing, Alan M. (1950), ``Computing Machinery and Intelligence'', Mind 59: 433-460.

b)    For a description of the Turing Test, together with a description of the Chinese-Room Argument, and my own views on them, see:

Rapaport, William J. (2000), "How to Pass a Turing Test: Syntactic Semantics, Natural-Language Understanding, and First-Person Cognition", Special Issue on Alan Turing and Artificial Intelligence, Journal of Logic, Language, and Information 9(4): 467-490.

c)    Briefly, the Turing Test considers an interrogator and either a human or a computer (the interrogator doesn't know which) in a room; the interrogator's job is to ask questions of whoever or whatever is in the room to see if it can be determined whether it's a human or a computer.

d)    Turing wrote (in 1950):

        "I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted"

        *    Note that it is now (Dec. 6, 2000, as I write) the "end of the century"!

e)    2 questions:

        *    How would you program a computer to pass the Turing test?

        *    What kinds of questions should the interrogator ask?

5.  An Artificial IQ Test

One way to begin to answer that last question is to consider the kinds of questions that are found on "IQ" or SAT-type tests.

I have prepared an AIQ test that you might find interesting.
 

6.  The Chinese-Room Argument

Searle's objection to the Turing Test is that it is possible to pass the TT, yet not (really) think.

Suppose that the interrogator is a native speaker of Chinese.

Suppose that there is a human in the other room who does not understand either spoken or written Chinese.

Suppose that the human in the room is equipped with a book that contains an algorithm, written in English (which the human does understand), that tells the human how to manipulate certain "squiggles" (actually, Chinese characters) in certain ways.

Suppose that the interrogator gives to the human in the room a story in Chinese, followed by a series of questions about the story, also in Chinese.

Suppose that the human takes this input, which, to him (I say "him", because in the original version of the argument, the person in the room is Searle) is nothing but meaningless squiggles, manipulates the squiggles according to the algorithm, and outputs more squiggles.

To the interrogator standing outside the room, the output consists of perfectly grammatical and correct answers in Chinese to the questions.

So, from the interrogator's point of view, the man in the room has passed a Turing test for understanding Chinese, but from the point of view of the man in the room, he does not understand Chinese.

Therefore, the Turing Test fails as a test of a computer's ability to "understand" or exhibit real cognition.

I will leave as a final exercise for the reader the answer to the question whether cognition is computable.

(For the record, I think it is.  For my reasons, see my paper on the Turing Test, cited above, and:

Rapaport, William J. (1998), "How Minds Can Be Computational Systems", Journal of Experimental and Theoretical Artificial Intelligence 10: 403-419

7.  For More Information on AI, See:

SOME STANDARD SOURCES OF INFORMATION ON A.I.


Copyright © 2001 by William J. Rapaport (rapaport@cse.buffalo.edu)
file: 572/S01/111.lecturenotes.12ja01.html