Author: Albert Silver
Date: 07:48:09 06/27/01
Go up one level in this thread
On June 26, 2001 at 18:57:44, Slater Wold wrote:
>Dear Chris,
>
>A week ago, I wouldn't have been able to to contribute anything or say anything.
> However, I think I can say one or two things now, having some light shined on
>me a while back, by the Usual Assholes. :)
>
>People WILL blast you here. They will blast you for stupid things. Recently, I
>wanted to know which computer program to use against a couple of GM's I will be
>playing soon. I thought about it, and came to the conclusion, "I will just play
>them all against each other, whoever wins is what I'll use." And never thought
>another thing of it. I posted it here, and not a SINGLE person posted anything
>"delighful" about it. They all blasted me, telling me that I was a moron for
>trying to pick the best canidate to play a human, in a C vs C tourney!
>
>I was kind of upset, because I for one know that C vs C is NOTHING like H vs C.
>And I was not trying to claim that "Program X" would be the best against humans.
> I was simply trying to get a basis on picking one of them. No one got this
>point, and probably still don't. Oh well.
Clearly you weren't around when all the discussions on the BGN qualifier broke
out here, so perhaps a little clarification is in order. Your proposed qualifier
is pretty much identical to what BGN proposed to do, and with the same purpose
(not to mention the unfortunate name 'qualifier'). The BGN qualifier left a lot
of people very angry. Without going into many details, known programmers of top
programs were overlooked and ignored, and a couple of CB programs somehow got to
battle it out by themselves. In fact BGN even tried to label the winner of their
event as some kind of World Champion, stomping all over the official ICCA's
title held by Shredder as well. They dropped this in the end, but as you can
imagine, the resentment of this situation was huge. Then there was the proposed
purpose of the 'qualifier' that had intrinsic problems. How can a match between
two programs show which program will play best against a grandmaster? It is well
documented that in comp-comp matches a single extra ply will have a decisive
effect on the outcome, and Fritz is really the king of plies, and all programs
work that way: 5 plies - 0 sec.; 7 plies - 3 sec.; 10 plies - 54 sec. etc.
Against a grandmaster, this extra ply will certainly have _some_ effect, but
overall it will weight far less. The reason is that grandmasters fight with far
different weapons such as plans, maneuvering, and the almighty endgame
transition (steering a game into a favourable endgame). Put Fritz on a machine
where it is chugging out 10 plies in the middlegame, and then on another where
it gets 11 and against other programs the difference will be heartfelt, but
against a GM? How is 11 plies instead of 10 going to teach it that the endgame
transition is unfavourable in a specific position? That's why the extra ply
makes so much of a difference to them, but not necessarily to a GM. That's ALSO
why it really takes a GM to change this, because a 2000 player really won't play
like that. They may try, but first their judgement will be way off, and second
they won't have the technique to exploit that small edge in the endgame. So what
happens is that a GM isn't just a 2000 with a lot more knowledge and seeing
deeper, they are also fighting with very different weapons. Now, some programs
have been _especially_ developped to fight with greater understanding as opposed
to greater depth, and those are the ones I believe would have the better
fighting chance against a GM. That's why I said Shredder _might_ easily do
better against a GM because of its better endgame play and knowledge, yet still
get clobbered by Fritz because of those plies. I'm not saying Fritz is weak or
is a dummy, but its main strengths are clear. The only real way to know what
programs will do best against a GM is by playing them against GMs. If that data
is unavailable, your judgement is probably much more reliable than comp-comp
matches, hence the posts on flipping coins, etc. Anyhow, I hope this helps. :-)
Albert
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.