Author: Jonas Cohonas
Date: 09:24:17 12/04/01
Go up one level in this thread
On December 04, 2001 at 09:46:22, Christopher R. Dorr wrote: >Again, this brings us to the issue of 'true' computer strength. I Have the >next-to-latest versions of Tiger, Rebel, and Fritz. All of these could likely >'hang' with a 2550-2600 GM. But is it fair to say that is their 'true' strength? >I'm only a USCF 2150 or so, but I have been able to draw all of these programs >*much* more often that I 'should' be able to against a 2600 player. > >I've gotten very proficient at playing for a draw against programs. I have blitz >draws against Crafty online. When you play for a draw at the outset, these >programs do *not* win nearly as often as would an equivalent human GM. They >often fall for things like pawn sacs to get to an opposite-color bishop ending, >or pawn sacs to reach dead drawn R+P endings, or even pawn sacs to get to drawn >K+P endings (he has 2 connected passers, but I have one protected one that he >can't move away from). > >In a thread on this earlier, a couple other expert/NM level players expressed >the same thing. I have a very good friend who is a USCF 2100 who was able to >draw Fritz 5 almost at will. Over several dozen games with muliple 'GM' >programs, I was able to get enough draws to extrapolate a USCF 2400-2450 >performance rating at G/15 and G/30 on a Celeron 800. > >So which is it? Are these programs the 2600s that they look like when they plan >Anand, or are they the USCF 2450s that they look like when they play me? One's a >strong GM, the other a weak IM. > >I am certainly *not* saying these programs are weak. But I am saying that they >do not conform to the typical performance pattern of a human, hence saying >*either* that they are strong GMs or weak IMs is misleading. Eventually, this >artifact will disappear as they become capable of playing GMs at the 3000 level, >but for now, it really isn't fair to say that Program X is a strong GM any more >than it is to say the same program is only a weak IM. > >Chris > The problem here is that GM's playing against programs will not try to draw, they will go for the win from the word go. So making any conclusions about playing strenght of a given prog, should be based on that, and not that it is easy for a relatively "weak" player to draw the progs, since they cannot beat them at will, in the same ratio. It has been my argument before that it is comparativly easy for a 2000+ player to draw even the best programs at almost any time control, if that is all he tries to do, but it becomes so much harder when he tries to win from the word go and there lies the true test of strenght. If you played 100 games against say Fritz 7a at any time control you wish and you went for the win from move one 1: how many wins would you get? 2: how many draws would you get? now that would be interesting to know. When i play another human, not absurdly over my own rating, i try to go for the win, by playing my best game, not to try and play on his weaknesses, well of course if he makes a weak move i will try and take advantage of it, but i will play the board not the man. If everyone were to play the board, not the program so to speak, i am almost certain that the score would favour the program at any level. Q: how many times do you lose in% when you try to draw a prog? and how many times does the program lose when you try to draw? Regards Jonas
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.