Author: Robert Hyatt
Date: 07:11:11 07/07/01
Go up one level in this thread
On July 07, 2001 at 03:28:14, Uri Blass wrote: >On July 06, 2001 at 23:26:42, Robert Hyatt wrote: > >>On July 06, 2001 at 20:37:53, odell hall wrote: >> >>>On July 06, 2001 at 08:28:05, Robert Hyatt wrote: >>> >>>>On July 06, 2001 at 05:01:13, odell hall wrote: >>>> >>>>>On July 06, 2001 at 04:54:02, Mogens Larsen wrote: >>>>> >>>>>>On July 06, 2001 at 00:18:54, odell hall wrote: >>>>>> >>>>>>>If i am not mistaken didn't roman just lose a two game match to shredder at the >>>>>>>time control of 30 5, what evidence can you produce which says if humans have >>>>>>>more time they win??? All the 40/2 games we have seen in the last threee years, >>>>>>>does not prove that point. IN fact we have seen that even on hardware that is >>>>>>>barely decent tiger has performed on the level of 2700 elo. Can you show me one >>>>>>>bad result of a computer at standard time controls??? If you cannot then all you >>>>>>>have is conjecture vs our facts and hard data. Even century 1 performed at 2552 >>>>>>>over a period of many games, show me the results where grandmasters have gotten >>>>>>>the best of the computers, do you have even one result???? >>>>>> >>>>>>Bob already explained why timecontrols alone don't tell the complete story. The >>>>>>other aspect is the incentives to try and harness computer programs. I, for one, >>>>>>can't see what those would be. The only major carrot is money as far as I can >>>>>>tell and since it's not the primary source of income the effort limited IMO. In >>>>>>general the games against computer programs are few and far between. Why devote >>>>>>a lot of time to that? >>>>>> >>>>>>The GM strength discussion is a little strange in the sense that some in the >>>>>>computer community thinks of it as a competition, ie. beating GMs regularly >>>>>>proving strength. Unfortunately, the competitor (your average run of the mill >>>>>>GM) hasn't got a clue about the "contest", so he/she generally ignores them >>>>>>altogether. And since one is standing virtually still and all the programs >>>>>>moving forward, there comes a point of catching up. However, this fact will not >>>>>>prove anything about strength IMO. It's like running against Maurice Greene when >>>>>>he's tying his shoes with his back to the track. >>>>>> >>>>>>Mogens. >>>>> >>>>> >>>>>you and bob are both full of shit, bacause Ed Shoeder did offer a money >>>>>incentive during the grandmaster challenge, go to the rebel home page >>>> >>>> >>>>Two things: >>>> >>>>1. keep up this kind of posting and you will cease to exist here. >>>> >>>>2. look at the amount of money Ed offered. Compare that to the amount of money >>>>available at the typical GM tournament. Then figure out which you would spend >>>>more time trying to win >>> >>> >>> I don't know about you Bob, but I am not Rich, for me 500 dollars is enough >>>incentive to play my best at something that I would do for Free, To Be a >>>grandmaster you have to live chess right? Explain to me why 500 dollars is not >>>enough incentive for a Grandmaster to play his Best? Nowadays your average >>>Grandmaster is poor, you can read about that all over the net, this amount is >>>not pennies, it's not like they make the same money that other professional >>>athletes make. Michael jordan would not lose to you in a one on one, even if the >>>incentive was 0. >> >> >>Simple answer. You are a GM. You have two choices: (1) study like hell to >>do well at the next big tournament coming up in 6 weeks. First prize is $10,000 >>(or more). (2) study like hell to win 500 bucks by playing a computer, and >>throw away any chance to out-prepare your opponents for the big tournament. >> >>Win 500 bucks, miss out on 10,000 or more. >> >>Which would you _really_ do??? > >The fact is that there are other GM's who compete for the 10,000$ so practically >the average price that the average GM expect from the tournament is clearly less >than 10,000$ I don't think that matters. Everyone that buys a lottery ticket _really_ believes they have a chance to win the entire prize. Even though the odds are impossibly tiny. > >Most GM's are going to be very happy to earn 500$ per game so practically they >cared more about the result of the game against Rebel then about the result of >the game against one human. I only asked one about this on ICC (not Roman). He said "the comps have some clearly noticable strengths and weaknesses. Most comps have a recognizable set of weaknesses after you play them a lot of games. The problem is that most GMs are not willing to invest the time since it is so unlikely they will ever play a comp in a real tournament. A few of us study them on ICC. But maybe only 10 out of nearly a thousand. And of that 10, almost all try to play their normal game for practice, rather than trying to play the comp's weaknesses which doesn't help us against our usual opponents in (FIDE) tournaments." That is a pretty reasonable statement IMHO. > >If in this conditions Rebel could get performance of more than 2500 then it is >clear to me that if Fide let rebel to play in Fide tournament then it can do >better(The humans will have less motivation to learn it because the practical >money they can get from one game against it in a tournament is going to be less >than 500$ and the hardware today is better than the hardware of Rebel in the GM >challange) It depends. Let's see how the Kramnik match goes where he is going to prepare carefully due to the amount of money at stake. My analysis of the outcome: If Kramnik wins, even by 1/2 point, all we will know is that the computer is no better than him. Because in a match, if I won the first game, I would be playing for draws in the rest of the games to get it over, safely. If I found a case where the comp had no idea what it was doing, I might go for the second point for safety, but I would _not_ risk _anything_ after the first win. If Kramnik loses, then we clearly have at least one electronic GM player in his opponent. If the match is drawn, who knows? That might be my strategy were I in his place. Play for a draw in each game, and hope the comp makes that one critical misjudgement in an ending to give me the point and the lead. But drawing would give me enough money to make me _quite_ happy for a long while. It would be more interesting if the rules were "you get $100,000 for every win, nothing for a draw or loss." They he would be playing for a win in every game, and we would see what he really could (or could not) do. > >If Rebel or another program can get the GM norm's under these conditions(I >believe that they can) than it proves that computers are GM strength and the >fact that maybe humans can play better by anti computer tactics is not relevant >because humans do not use anti-computer tactics. But we know that they do. Roman is quite a problem for programs, for example. There are others. Mecking can do a reasonable job of anti-computer when he wants to. Most of the time Crafty (scrappy) eats him alive. But on occasion, he reverts to "all I want to do is draw the thing" and he will draw maybe 1/2 of the games and lose the rest. At blitz time controls this is not a bad result. > >I suspect that most of the GM's with rating of less than 2650 could never become >to be GM's if the opponents were motivated to win or draw against them by very >big prizes. > >It does not mean that they are not GM strength. > >Uri Of course it doesn't mean that. But humans are all pretty similar in what they can and can't do. Comps are very different. I would not go to a grass tennis court, and play against a "lobber", to prepare for a serious match on a clay court with a "smasher". I used to train against the same sort of opponent I would play against. Whether it be basketball, tennis, karate, chess, you-name-it. Until humans decide that it is worthwhile to "play the opponent and not the board" (they do this to a limited extent against humans.. ie avoid their pet openings and the attendant analysis) humans and comps are a poor mix if the purpose is to compare them to each other to figure out which is stronger.
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.