Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Issue regarding GM strength

Author: Christopher R. Dorr

Date: 06:33:40 10/29/01

Go up one level in this thread


On October 27, 2001 at 11:08:38, James T. Walker wrote:

>I'd like to interject another thought into this discussion.  I play 9-ball
>(Billiards) a lot and practice at home on my own table.  I often "Break & Run"
>the rack.  I have done it at home more than 5 times in succesion.  I have NEVER
>done it twice in a tournament ! There is something diffent about knowing that
>your opponent is just waiting to pounce on your smallest mistake.  Playing in
>the comfort of your home with nobody else around is a nice "advantage" for most
>people.  Especially when nothing is really on the line.  You have to be able to
>perform under real tournament conditions to claim the prize.
>Jim (Choke-up artist)

That is quite true. The flip side, however, that if I (USCF 2100) had t play a
computer (Probably 2600) in a tournament (unlikely I know), I would have *no*
pressure on me; nobody would have any expectations of victory. However, the
drawing methods I frequently use against computers (tossing a pawn to reach a B
of opposite color ending, giving it huge amounts of space in return for a
protected passer, etc) are very easy to implement in even tournament play.

Modern programs are *very* hard to beat if you are a mere NM or strong expert.
But they are *dramatically* easier to draw than a human GM is. I feel very
confident that I could play a very good program (Tiger, Fritz, Crafty, etc) on
good hardware, and have at least a 25% chance of getting a draw. Against an
equivalent human GM, I would estimate the likelihood to be no more than 10%
(Although I have drawn a good GM in a tournament before, I have lost to several
others).

This is the issue that I focus on in these 'Computer = GM' discussions. of
course they *can* play like GMs....when *playing* GMs....but they can also play
like strong NM/FMs when playing weaker players intent on drawing them. So I
really don't know how to estimate their 'true' strength. If I remember Elo's
book correctly, he designed his system so that if a player played a different
sub-group of the rating pool, that person's rating should stay pretty much the
same. With humans, this is usually the case (i.e., when I play in open events
against human masters, I usually end up with a performance rating of 2150 or so.
When I play in local tournaments against 1800s and 1900s, I usually end up with
performance ratings around 2100...not that big a difference.

But Computers don't seem to behave like this. If I put Tiger in a PIII/1Ghz in
the US Championship, it would likely perform at a level of 2550-2650 FIDE. But
if I put the same program on the same box into the Expert/NM section it most
likely would have enough draws against 2100s to drop it's performance rating to
2300-2450. So can we honestly say that a program is 'GM strength' if it plays
like a GM *against* GMs, but also only plays like a FM against NMs? I don't see
how we can.

Chris



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.