Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF and the programmers............

Author: Bruce Moreland

Date: 10:35:19 03/17/98

Go up one level in this thread



On March 17, 1998 at 12:08:00, Ed Schröder wrote:

>Blame me for my part in it. With Rebel9 I joined the club. Now I
>step out. It was a mistake. I will not support this silly cooking
>race any longer. Back to the roots which is the chess engine.

If your program played on ICC you would get a different perspective on
the same issue.

I had a special version of my program, running on a seperate account,
that would play 5-minute blitz chess at maybe 0.2 seconds per move, so
it was moving more or less instantaneously.

This account was fun for humans to play against, it was pretty strong,
they didn't have to sit around waiting for it to make its move, and they
could beat it sometimes.  The games were always over very quickly, so
someone could play dozens of games in a row.

I often watched this account zoom up into the 2400-2500 rating region on
ICC.  But then I would go to sleep, and when I checked on it the next
morning it would be at 2100-2200, maybe lower.

When I checked my mail it would take forever to download, because there
would be like 200 games in the mail.

Eventually I would discover that someone had found a way to win, then
repeated it a whole bunch of times, gaining points each time.

Initially I got mad and refused to play these people anymore.  But the
more I thought about it, the more I realized that I was wrong to be mad.
 The fault was mine, not my opponent's.  My opponent was just making use
of their own highly developed adaptive facilities.  The opponent would
get beat repeatedly, then learn something and beat my program, and since
my program didn't learn anything, it would lose.

This is a constant problem for any computer on ICC, and it is bad for
everyone.  The program gets a low rating.  The opponent gets a high
rating that doesn't translate against other humans and against other
machines that are more adaptive.  And the opponent either stops playing
against you, or becomes unwilling to innovate, since they know that if
they deviate from known lines, they will lose games.

It is extremely important to pick the right person to blame for this
problem.  The blame lies not with the opponent, the blame lies with the
computer.

The best, most practical solution is to add the best learning you can.
It's an honest thing to do, it benefits you, it benefits your opponents
(they get to play chess rather than do experiments on a static
opponent), it benefits the rest of the people in the rating pool, it's
more human-like, etc.

You can't isolate the engine anymore.  It used to be that you could,
since all of the other programmers thought the same way.  But now they
are trying to build more complete chess players, so you have to build
one too.  Perhaps their main motivation is to inflate their SSDF
results, but it must be realized that regardless of why the learners
were added, they do result in a better *chess player*.

bruce



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.