Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Assumptions of fallibility/infallibility

Author: Bernhard Bauer

Date: 06:37:55 04/03/00

Go up one level in this thread


On April 03, 2000 at 07:52:35, Jeremiah Penery wrote:

>On April 03, 2000 at 07:35:44, guy haworth wrote:
>
>>
>>An interesting situation!
>>
>>On the one hand, if you assume that your opponent is infallible (e.g. HAS the
>>EGT), you will want to avoid the EGT situation.
>
>At least in my example case, the opponent wins easily if you avoid the TB
>position. :(  Even in _less_ moves than the TB would have required, because he
>now has QNN vs. NP instead of only NN vs. P.
>
>>On the other hand, if you assume that your opponent might be fallible (i.e. does
>>not have the EGT) you could go - as you did with the hard-to-win 'EGT' situation
>>in preference to the other.
>>
>>If an 'infallible opponent' would also win the non-EGT situation, maybe it is a
>>'no brainer' situation but certainly one where you have to override the
>>computer's automatic avoidance of lost EGT-positions.
>
>It seems to be a problem of computers in general - they assume their opponent is
>always 'optimal' according to their own standards, and so they can get into
>trouble because of this.  If the program searches 50 ply in the middlegame (a
>ridiculous example, I know, but it makes the point) and sees a spectacular
>winning combination for the opponent, it will sacrifice a piece immediately with
>a better score, but will go on to lose easily.  It's possible that the opponent
>didn't even see the combination, and the piece loss wasn't necessary.  A very
>interesting problem...
>
>Jeremiah

Yes, a very interesting problem. Don't know what to do in such a situation.
The 100:1 time odds game between Crafty and Rebel comes to mind.
Crafty avoided lines that Rebel had no idea of - and lost. May be Crafty could
avoid the loss by playing non optimal lines.
Sometimes it's better to know less :-)
Kind regards
Bernhard



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.