Computer Chess Club Archives


Search

Terms

Messages

Subject: Yes, agreed

Author: GuyHaworth

Date: 03:04:32 10/17/02

Go up one level in this thread



Ingo puts the case better than I did.

Let's suppose you had QP-Q as might have happened in game 5.  The EGT exists so
DF is showing an unarguable =0.00 but it has chances to win - Kramnik might make
a mistake.  The % of wins is significant.

DF may be able to choose between draws, maxxing VK's probability of error:  the
issue of how to do this has been aired more than once in ICGA_J.

The 50-move rule is a compromise between assuming that VK will play infallibly
(and declaring a draw off the EGT's declaration) and requiring VK to play
infallibly for ever.  However, if VK holds the draw for 10 moves, do the rules
then say that it is a draw.  This would be replacing a 50m rule with a 10m rule.

'Progress' is not reflected in the '0.00' but in the fact that VK is being
tested, and may yet fail as the situation is arguably getting harder to defend.

If the situation was QPP-QP, the same argument holds, except that the EGT is
being consulted in the search, and the =0.00 is a backed-up score.


I agree with those who say that, at the very least, VK resigned too early.  Why
assume that the computer is infallible:  its programmed by humans.

In the recent Checkers contest, one program lost a theoretical endgame-draw
because its EGT access code had a bug and selected a position that looked drawn
but was in fact lost:  q.v. reports by Martin Fierz.  Computers have inverted
their eval_functions, created phantom pieces etc before now.

As a retired woodpusher, I'm all for seeing the game played out to mate.

g




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.