Author: Dave Gomboc
Date: 19:22:31 09/29/99
Go up one level in this thread
On September 29, 1999 at 20:52:37, Michael Fuhrmann wrote: > From my limited understanding of computer chess, I gather that >programs choose moves based on the assumption that their opponents always make >the best possible moves. This may not happen -- especially in the case of a >human opponent. Which leads to the paradox that computers might win more games >if they sometimes played weaker moves. > The reason for this is that humans know that a computer always makes >the best move according to its own analysis. This predictability can only >benefit the human. > Take this example: a computer sees that, given best play by it and >its opponent, it is about to lose the equivalent of a piece. > A GM (let's say white) in the same position might decide that loss of >the piece will lead to a hopeless position, and try to distract his human >opponent with a counterattack. The counterattack is unsound, and the GM knows >it. With correct play by black, white will actually be worse off. But in a >complex position, the counterattack may look sufficiently threatening to >distract black. The result: black abandons his own attack and is eventually >"faked out" of his advantage. White steals the game. > In some positions, white has little or nothing to lose and >everything to gain by this ploy. > So (a) Is any computer today capable of choosing this kind of >technically unsound, counter-attacking move (in positions where it will probably >lose anyway given best play) instead of the "correct" move according to its >eval? > And (b) Does anyone else agree that if a computer could do this, it >would add a human-like, "psychological" dimension to its arsenal? It is rumoured that some of the commercial programs will do this if they are hitting "guaranteed loss" scores, though none of the authors has stood up and admitted it (here, anyway). Dave
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.