Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Can a PC programme beat a top GM in a match?

Author: Vasik Rajlich

Date: 15:02:37 01/02/05

Go up one level in this thread


On January 02, 2005 at 16:11:01, Anthony Cozzie wrote:

>On January 02, 2005 at 13:43:16, Vasik Rajlich wrote:
>
>>On January 02, 2005 at 10:00:43, Anthony Cozzie wrote:
>>
>>>On January 02, 2005 at 05:16:02, Vasik Rajlich wrote:
>>>
>>>>On January 02, 2005 at 04:34:27, Vincent Diepeveen wrote:
>>>>
>>>>>On January 01, 2005 at 08:52:13, Clive Munro wrote:
>>>>>
>>>>>>Forgive me if this question has already been discussed but I havn't read all the
>>>>>>threads on this site.
>>>>>>Has the time been reached that a commercial PC programme can beat Kasparov and
>>>>>>co over 40moves in 2 hours control? For instance if we had a 10 game match over
>>>>>>say two weeks could a commercial programme running on the latest retail hardware
>>>>>>(not 200 pcs linked together etc) beat the top GMs?
>>>>>>If not how close is it?
>>>>>>
>>>>>>Best
>>>>>>
>>>>>>Clive
>>>>>
>>>>>It depends upon payment. if you pay the GM regardless of result, what is
>>>>>happening always as the computer game company is way too much involved in the
>>>>>marketing it generates, then the top GM will of course not care and play 4-4 or
>>>>>something or lose.
>>>>>
>>>>>On the other hand if you only pay him when he wins, he will beat the hell out of
>>>>>you.
>>>>>
>>>>>Yet top GM's are too demanding. We know kasparov wants 1 or 2 million 'match
>>>>>fee' paid. You have no option if you want to play kasparov. he will demand
>>>>>payment in advance with a bank garantuee.
>>>>>
>>>>>Kasparov is simply the special case here. he draws so much publicity that you
>>>>>should play him if you can afford it. Yet playing him each few years would be
>>>>>too expensive for the sales in return :)
>>>>>
>>>>>The problem of other GM's is that you get near to zero publicity except within
>>>>>the chess world itself. You can play an (ex-)FIDE world champ for just a couple
>>>>>of thousands. No problem.
>>>>>
>>>>>Yet he'll demand also payment in advance: "to show up".
>>>>>
>>>>>He can then give a show without using any of his careful prepared openings, as
>>>>>those openings are used against humans only. If a GM has a novelty he'll sure
>>>>>won't play it against a computer. Shame.
>>>>>
>>>>>What we DO know is that the programs have increased in playing strength REALLY a
>>>>>lot last few years.
>>>>>
>>>>>Way more than i had expected myself to be honest.
>>>>>
>>>>>So a few years ago there was just one time someone who offered to GM's matches
>>>>>in the next form. If they would lose, they got nothing. If they drew then 250
>>>>>dollar, if they won then they got $500.
>>>>>
>>>>>Many very weak GM's took up the challenge and played Rebel. Rebel sure is a good
>>>>>program against humans, no question about it. Those real weak GM's 24xx rated
>>>>>and 25xx rated easily drew rebel and some actually won.
>>>>>
>>>>>If you organize again such a match i would expect you will see more of a
>>>>>difference. Certain 'profitting' type GM's who managed to kick Rebel by for
>>>>>example a sudden attack, they will more and more lose.
>>>>>
>>>>>However you still can't help certain players who play always the same opening
>>>>>and also use it against the computer.
>>>>>
>>>>>Offer IM Ziatdinov a match against a computer. Or offer GM Boris Kreiman a match
>>>>>against a computer. Especially the latter will just destroy it, no matter how
>>>>>many processors you use.
>>>>>
>>>>>He'll play a good opening and destroy it.
>>>>>
>>>>>Want to find out?
>>>>>
>>>>>Just pay him $500 a game, for each game he beats a machine of your choosing in
>>>>>40 in 2.
>>>>>
>>>>>Don't even offer money when he draws i would say.
>>>>>
>>>>>What will the result of a 8 game match be?
>>>>>
>>>>>Well that depends heavily upon what type of reward you give.
>>>>>
>>>>>If you offer $500 only for wins and nothing for draws, expect 3-5 wins from the
>>>>>GM. If you offer $4000 for winning a 8 game match, he'll beat you with 5 draws
>>>>>and 2 white wins and 1 loss from GM side. Just enough to cash in the money.
>>>>>
>>>>>I specifically mention Kreiman here, because he has a good opening and has
>>>>>experience playing software.
>>>>>
>>>>>I know so many GM's who will perhaps even lose a match from me if i prepare
>>>>>well, as their openings suck ass, and they would not prepare a match against me
>>>>>nor against the computer, and they have zero chance against any serious
>>>>>preparement. All software programs are pretty well prepared because of the
>>>>>openings book, but very little are really in depth prepared.
>>>>>
>>>>>Just mention the GM name, i'll lookup the openings the dude plays, and i can
>>>>>already give you a pre-prediction.
>>>>>
>>>>>Sutovksy? no, not a chance, he'll lose from Nimzo1998.
>>>>>IM Ziatdinov? yes, makes a good chance against the software.
>>>>>GM Ikonnikov? yes he'll even destroy software long after world champs have won
>>>>>from software. Ikonnikov knows he is tactical weak and plays every day in ultra
>>>>>safe anti-tactics mode and does do so by playing closed positions preferably.
>>>>>Even against 1.e4 !!!! He'll destroy anything.
>>>>>Offer him $100 for a draw, $250 for a win, and promise 20 games.
>>>>>This will be disastreous for your software.
>>>>>
>>>>>Rating of those guys doesn't really matter anymore when playing the computer.
>>>>>Personal style and motivation and 'bugfree' play are more important. I feel
>>>>>that's the difference now against todays hyperagressive software.
>>>>
>>>>I half-agree here.
>>>>
>>>>It's true that all engines still have massive problems. Someone with the
>>>>positional judgement & opening repertoire of a top player and enough tactical
>>>>accuracy could crush them.
>>>>
>>>>I'm not sure though that any human could pull it off. Kramnik had a big money
>>>>incentive to win, was well-prepared, has a clean sound style - and still
>>>>couldn't get it done. Chess just has too much tactics.
>>>>
>>>>Vas
>>>
>>>You have to admit Kramnik's match against Fritz was incredibly suspicious.
>>>First, Kramnik *embarrases* Fritz in their first 4 games, then he makes A) a 1
>>>ply blunder and B) a ridiculous sacrifice to even the score.
>>>
>>>anthony
>>
>>Not any more suspicious than the other man-machine matches. It's the normal
>>story - when the human wins he completely outplays the machine, when the human
>>loses he either overlooks something or tries something crazy.
>>
>>Vas
>
>That is why I am suspicious of ALL the man-machine matches :) :)
>
>anthony

I think it's pretty normal. When you play somebody who's really good at
something, he'll make you look really bad at it. In comp-comp the computer
positional weaknesses get masked ... ditto for tactical weaknesses in
human-human ...

Vas



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.