Re: Turing test

[Dan Pless <dpless@xxxxxxxxx>]

>> In the 1950s somebody (at IBM?) wrote a checkers program that was
>> self-learning.  He programmed the computer's moves only to follow the rules
>> of checkers--no intelligence in the computer program, except that when the
>> machine lost a game, it retained all the board positions for that game and
>> tried not to replicate any of them in subsequent games.  The author claimed
>> that after about 15 games the computer won one, lost the next two and was
>> unbeatable thereafter.

This understates how much checkers knowledege went into the program and
vastly overstates the strength of the program after training.  The program
(by Arthur Samuel of IBM) is still a very important early example of
machine learning.  It was ahead of its time in many ways.

>> I would like to read a primary source description of this early machine
>> intelligence experiment.  Can anybody cite the source material, or better
>> yet, send me a copy?

The reference you are looking for is:

Samuel, A.L.  (1959)  Some Studies in Machine Learning Using the Game of
Checkers, "IBM Journal on Research and Development", 3, 210-229.  Reprinted
in E. A. Feigenbaum & J. Feldman (Eds.), "Computers and Thought".  New
Your:  McGraw-Hill.

I believe it is also reprinted in:

Luger, G.F.(ed). "Computation and Intelligence: Collected Readings.",
Cambridge: AAAI/MIT Press (1995). 

but I am not sure about that one.

Dan Pless