Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Measuring playing strength

Author: Dann Corbit

Date: 19:58:30 12/30/99

Go up one level in this thread


On December 30, 1999 at 22:27:43, hgkjhg wrote:
[snip]
>
>may be programs shouldn't rely on their books so much.  They should use only
>solid lines and if they're truly the best, the results will show that.  By the
>way, fritz is the best against humans, definitely.

You are really onto something here [I think].  Unfortunately, the devil is in
the details.  I think the two biggest problems with chess programs is:
1.  Bad opening lines.
2.  Early endgame.

Solve those two and you will have a world-beater.  The problem is that with
millions of potential opening positions, how do I know if one is good or bad?

Also, there are two kinds of bugs in opening books.  The first bug is a true
bug.  It is an undiscovered tactical shot.  There are a frightening number of
these, even in carefully analyzed books.  Humans are constanly debugging the
lines of play and discovering problems in them and improvements, etc.  That's
how we got all these wonderful opening strategies.  However, just because a GM
played it ten times and it wins 60% of the time does not mean it won't have
serious tactical flaws.  This sort of flaw can be overcome (eventually) by a
project like C.A.P.

The second bug is due purely to the fact that we are talking about computers.
Suppose that there is a brilliant sacrifice or positional move that leads to
wins.  Now, we can write that move in an opening book and *know* that it wins
and for good reason.  *HOWEVER* the computer does not know the _plan_ behind the
move.  In order to capitalize on that move, the computer must follow the
sequence of moves that leads to a benefit.  But since it does not know *why* the
move was made or the plan behind it, the following moves are aimless and the
sacrifice or positional move is wasted.  The only way to fix this is to write
better chess algorithms or possibly to deepen the lines all the way to the
benefit (a lot easier said than done).



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.