RE: Chess and Go study

From: Gary Miller (garymiller@starband.net)
Date: Sun Dec 15 2002 - 07:46:24 MST


On Dec 15th

>> The rule is, that "it is almost always bad to lose a queen". On the
other
>> hand, a lot of calculation will prove you, that checkmate is more
probable, if
>> you've already lost your queen.

While it is true that chess programs and players who have formalized
these rules as heuristics in their determination of the next best move,
play better than programs or players who do 100% brute force look-aheads
that is not where the complexity lies. All is not really that cut and
dry!

The complexity lies in the heuristic discovery and the application of
multiple conflicting heuristics in order to determine the best move.

For instance if we followed the rule blindly to preserve our queen and
not let her be taken. We would miss a some opportunities for checkmate
on the next move by sacrificing the queen. Also many times it is
advantageous to trade queens in order to have a better position on the
board.

It would be correct to say that the better the heuristics the better a
chess program will play.

But even though computers are now capable of memorizing all chess
opening and endgames, and doing far more lookaheads than their human
counterparts, human grandmasters are still much better in discovering
and knowing when to apply what heuristics. They are so good in fact
that when chess masters work with programmers to improve a program's
heuristics, the chess masters in many cases can not say why a move is
better or what heuristic they are using to know it is better.

Although Samuel's Checkers program and Tsauro's Backgammon program were
able to learn to play optimal games by experience. I have not read of a
chess program that was capable of truly learning the game and being any
good at it in this same manner.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT