Author: Robert Hyatt
Date: 08:24:30 08/12/01
Go up one level in this thread
On August 11, 2001 at 04:00:03, Uri Blass wrote: >On August 11, 2001 at 00:30:06, Robert Hyatt wrote: > >>On August 09, 2001 at 16:13:32, K. Burcham wrote: >> >>> >>> >>>some here say that most GMs are not super GMs. >>>some here say that there are very few super GMs. >>>some here say that the super GM plays on a different level than a GM. >>>so i guess that i can safely say that this is a match between a program >>> and a super GM, kramnik. >>> >>> >>>i have heard all of the opinions here on programs "are GM level", >>> "are not GM level". >>> >>>so what will everyone say if fritz draws one game against this >>> "super GM"? >>> >>>what if fritz draws two games against this super GM? >>> >>>what if fritz draws three games against this super GM? >>> >>>what if kramnik barely wins the match? >>> >>>what if fritz wins one game? >>> >>>what if fritz wins two games? >>> >>>what if fritz wins three games? >>> >>>what if kramnik has a difficult time with this program? >>> >>>what if you are analyizing the games as they are played, and you discover >>> that the program is capable of holding the score close to zero >>> during most of the games against this super GM? >>> >>>what will you say if you experience any of these conditions? >>> >>>hey but maybe kramnik will make it simple for some of us and just dominate >>> the match, easily outscoring the program in every game. >>> >>>kburcham >> >> >>I've said this before. In my opinion: >> >>(1) if fritz wins the match, I will certainly have to re-think my position about >>whether computers are GM or not. >> >>(2) if Kramnik wins the match, we won't know a thing. He could win it 8-0 >>or he could win it 4.5-3.5, neither of which says a thing, really. He might >>take every obvious draw waiting on that one mistake that allows an easy win. >>Then he would coast the rest of the way with draws and we would not know much >>other than he is better. A lot better or a little better would be impossible >>to judge (IMHO). >> >>(3) if the match is drawn, it will be hard to draw a conclusion. I haven't >>seen the match specifics about the prize fund, but if I had to choose between >>the following three options: >> >>(1) lose the match and win 400K; >> >>(2) draw the match easily and win 700K; >> >>(3) try to win the 1M prize but take chances that might lead to case (1) >> >>Then I might go for (2). Or I might say "hell with it" and just take the 400K >>and run like the wind. That is a _lot_ of money. To go for case (3) is going >>to take a lot of courage. > >A lot of courage only if kramnik believes that Deep Fritz is a super GM. Not at all. You and I are going to have a gunfight at 500 yards. I know you haven't done much shooting with a 16x scope and the 300 Weatherby we will both have. In this match, I get one million if I hit you, I get 800K if we both survive after 10 shots each, and I get 600K if I get hit by you. I don't think that in a head-to-head shoot-out at 500 yards you have a chance against me as I have shot at that range for thousands of rounds. But I also know that one lucky shot by you will put me out of the match. What do I do? I play it cautiously and don't give you a good target, let you shoot your ten rounds and then I take the 800K and run. I don't have to believe you are good at all. I just have to know you can be lucky. And luck is all it takes... > >If kramnik plays a match of 8 games against a GM with rating of 2550 or even >2600 or 2650 he is not going to be afraid to go for case 3. The computer is a different animal. He _knows_ it is not going to overlook any tactical mistakes within its search horizon. The level of concentration is much higher as a result to double and triple-check the tactics before walking into a complex position. Kasparov showed this. 6 rounds wore him totally out. This is eight. > >He can be almost sure of winning the match. He also knows that if he misses a tactic, it is very likely his opponent will miss the same thing. But not the computer. It is a different type of opponent completely. Not nearly as positionally smart as a human, but _very_ tactically aware of what is going on at any instant. > >I see it differently. >I expect kramnik to do the best that he can. >I believe that he believes that he can defeat the computer and get better result >than 4.5-3.5 or 5-3 and I expect him to try to do his best even after winning. > I hope you are right. But I'd bet you are not. >Every result of Deep Fritz that is at least 25% is going to be a good result for >Deep Fritz and another evidence for me that Deep Fritz is a strong GM. > >25% against kramnik is a performance of almost 2600 and if Deep fritz can get >this performance under these conditions it means that under normal conditions >it can do better. > >Kramnik can get printout of Deep Fritz7's games in order to learn the program. >I am not sure if he can get the program >He is not going to get the opening book but he is going to get a lot of >information. I don't think the book is critical. A GM can steer the game toward the program's weaknesses, _if_ he knows about them. > >I expect Deep Fritz's performance against kramnik to be between 2500 and 2600 >because of this reason. > >I do not believe that kramnik is going to be happy with draws after winning one >game because I believe that winning the match with less games is better for >kramnik. > >If kramnik win the first game and draw 7 games then he has to work hard (8 >games) in order to win the match. > >If kramnik wins games 1,3,4 and draw games 2,5,6 then he wins the match with 6 >games so he can get the same money and invest less time. Back to the shooting match. Every time you pop your head up, you are going to have a .300 Weatherby on the way to you. The longer you pop up to aim to try to hit me, the greater my chances of hitting you are. Remember, I am a better marksman in this particular case. > >winning games 1,3,4,5 and drawing game 2 is winning the match with only 5 games >that is even better for kramnik. > >I am going to count the games only until kramnik gets at least the 4.5 that he >needs > >I believe that every result that force kramnik to play at least 6 games is not a >bad result for Deep fritz. > >5-1 or 4.5-1.5 are the results that I expect >only 4.5-.5 or 5-0 for kramnik are going to be a bad result for Deep Fritz. I don't think about "bad results". I think about "good". And a drawn match or a win for fritz is a good result. Anything else is bad because there is no telling how bad it could be. > >I believe that kasparv had chances to get 4.5-.5 or 5-0 against deeper blue if >he could get the program except opening book or even only a printout of deeper >blue's games with the evaluation of it and a commercial version of deeper blue >that is almost at the same level before the match. > >Uri That is possible. It just wasn't the question being asked...
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.