so what is a fair GM handicap vs Rybka?

[Dave McClain]

In NFL football, games are made theoretically even by give one team points. Example, a bad team may be given a 21 point handicap to make the contest even.
.
with similar time controls to the current match, what handicap would the GM need so the match is considered even?

Your analogy to football doesnt hold up. Its a bad one. The handicap is only for betting purposes. The actual underdog must STILL win the game straight up- without a handicap. And personally for me, i feel the same way about chess. Tho it may make it interesting for some spectators- not for me. Whether you are playing a computer or a person- you either win straight up- or you lose straight up. Your 3rd and only other option should be pass and not play at all.

George,

I agree. Clearly Ehlvest is past his prime and could probably be graded as a "B" Grandmaster now at best. They had to do something to make the match look like he had a chance. Taking pieces off the board as a handicap is ridiculous.

[ologist]

First of all I am not a 1400 player. I am unrated, but people I have had had people say I am between 1800-2000 USCF = FIDE - 150 (at this rating range) , more or less. Anyway, I would rather have the extra rook, sure, but if I make a not so good choice in the opening and Rybka gets a serious attack, I would be doomed. On the other hand, if Rybka is missing a minor piece I probably could easily make it to an endgame where I would have a very high chance of winning with the extra minor piece. This would be a hard decision for me to make if I were betting on this, but I most likely would end up taking the rook odds, although at my level it could prove more risky than taking the minor piece odds.

[NKOTB]

if you are 1800, rook odds is about an even match at 5 minutes per game. if you are 2000, knight or bishop sounds about right. You can not the smallest tactical mistake against engines rated 3000 and higher. One really needs to understand what 3000 means before club players try there hand. Lucky for us (I left chess at 1795 USCF), some of these 3000 type engines are free. Grab one, play at odds and report back the results.

[Graham Banks]

Last year, I played a 4 game match at 40/40 repeating between Rybka 2.1 and Mephisto Vancouver with Rybka giving up its KN and QN respectively playing both as White and Black.

From memory, the score was +2=1-1 in favour of Rybka.

[Uri Blass]

if you are 1800, rook odds is about an even match at 5 minutes per game. if you are 2000, knight or bishop sounds about right. You can not the smallest tactical mistake against engines rated 3000 and higher. One really needs to understand what 3000 means before club players try there hand. Lucky for us (I left chess at 1795 USCF), some of these 3000 type engines are free. Grab one, play at odds and report back the results.

I think that you are wrong here and at 5 minutes per game I expect rybka to beat most 2000 players with knight odd at blitz.

You probably confuse between rybka and free engines that are usually significantly weaker when most of them do not know not to trade pieces when they are at disadvantage.

My fide rating is 1975 and I lost 1.5:0.5 against rybka2.3.1 with rook odd
at 5+1(one game rybka was without rook h1 and one game rybka without rook a1)
I am not a good blitz player and I also think that I did not play well in the relevant games but it is certainly easier to trade pieces in order to win against free engines so you can learn nothing from games against free engines.

Here is the game that I lost.

practically I tried to play 45...Kc7 but rybka told me that it is a blunder and I saw winning advantage for rybka but at move 45 I already lose and 29...Rd8 is a significant mistake.

[Event "?"]
[Site "?"]
[Date "2007.03.08"]
[Round "?"]
[White "New game"]
[Black "Rybka 2.3.1 32-bit"]
[Result "1-0"]
[Annotator ",agur"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1"]
[PlyCount "91"]
[EventDate "2007.??.??"]

{512MB, Fritz8.ctg, URI-AMD} 1. Nf3 {-2.89/14 16} d5 {2} 2. g3 {-2.94/13 10}
Nf6 {2} 3. Bg2 {-3.03/13 10} c5 {1} 4. O-O {-2.89/12 21} Nc6 {1} 5. c4 {
-2.89/12 9} e6 {6} 6. cxd5 {-2.94/12 24} exd5 {2} 7. d4 {-3.06/12 10} Nxd4 {3}
8. Nxd4 {-2.90/10 2} cxd4 {1} 9. Qxd4 {-3.20/12 6} Be7 {10} 10. Nc3 {-3.22/11 6
} Be6 {9} 11. Be3 {-3.35/11 8} O-O {2} 12. Qd3 {-3.42/11 15} Qc7 {19} 13. Bd4 {
-3.40/10 9} Qc4 {10} 14. Qd2 {-3.51/12 7} Ne4 {12} 15. Nxe4 {-3.54/10 2} dxe4 {
1} 16. Bxe4 {-3.69/11 2} Rfd8 {15} 17. e3 {-4.21/13 5} Bf6 {24} 18. b3 {
-3.70/11 5} Qc7 {9} 19. Rd1 {-3.62/11 5} Bxd4 {3} 20. exd4 {-3.77/9 2} Bd5 {4}
21. Bf5 {-4.13/11 7} Be6 {6} 22. Be4 {0.00/15 2} Qb6 {6} 23. d5 {-3.56/11 7}
Rac8 {9} 24. Qe2 {-3.67/11 2} Qd6 {17} 25. Qe3 {-3.53/10 5} Bd7 {11} 26. Qxa7 {
-3.75/12 6} Qa6 {9} 27. Qxa6 {-3.38/11 5} bxa6 {0} 28. f3 {-3.52/12 4} Be6 {10}
29. Kf1 {-4.53/13 1} Rd6 {5} 30. Ke2 {-4.61/13 5} Rcd8 {4} 31. dxe6 {-1.21/15 4
} Rxd1 {16} 32. e7 {-1.29/16 1} Re8 {23} 33. Kxd1 {-1.16/11 1} Rxe7 {1} 34. b4
{-1.09/14 4} Kf8 {9} 35. Kd2 {-1.00/12 3} h6 {1} 36. a4 {-0.92/12 3} Re5 {1}
37. Bd3 {-0.88/14 3} Re6 {3} 38. Bc4 {-0.87/15 2} Rb6 {3} 39. Kc3 {-0.77/15 3}
Ke7 {5} 40. a5 {-0.81/17 10} Rf6 {1} 41. Kd4 {-0.57/17 4} Kd6 {4} 42. Bxa6 {
-0.40/12 1} Rxf3 {5} 43. Bc4 {-0.43/15 0} Rf2 {16} 44. b5 {0.00/18 4} f5 {4}
45. a6 {0.00/11 4} Rd2+ {14} 46. Kc3 {2.60/13 6} 1-0

[Uri Blass]

if you are 1800, rook odds is about an even match at 5 minutes per game. if you are 2000, knight or bishop sounds about right. You can not the smallest tactical mistake against engines rated 3000 and higher. One really needs to understand what 3000 means before club players try there hand. Lucky for us (I left chess at 1795 USCF), some of these 3000 type engines are free. Grab one, play at odds and report back the results.

Here is the opinion of larry kaufman from his experience about knight odd matchs against rybka on a fast quad machine(assuming the machine has some small opening book or the human is not prepared and the knight is knight b1)

if you are 2400 player knight b1 odd is right in 5+0 game
If the time limit is longer 5+5 then 2300 level is right
If the time control is 40 moves/2 hours then 2000 level is right.

I believe that his opinion is more realistic than your opinion.
link for his opinion

http://rybkaforum.net/cgi-bin/rybkaforu ... st#pid4732

see the post from
Date 2007-03-08 01:59

Uri

[JonP01]

Last year, I played a 4 game match at 40/40 repeating between Rybka 2.1 and Mephisto Vancouver with Rybka giving up its KN and QN respectively playing both as White and Black.

From memory, the score was +2=1-1 in favour of Rybka.

Very interesting results Graham. I might try similar handicap games once I get hold of Rybka.

[ologist]

Well, I played Rybka at queen odds last night, and I must say that the game is not even worth posting to waste bandwidth. It was simply confused and did not know what to do. It moved its knights out the first two moves, and and moved its knights two or three more moves in a row just to put its QN back on its original square! It was funny actually, it was like it was toying with me, lol. Any decent player could never be beat at queen odds by a machine or GM unless they blunder. No matter how much programming you do this will never be achieved, just the same as you could never get strong enough to lift an elephant over your head with one arm. In theory there is no limit to the expansion of your myocytes, but no matter how much weight training occurs a man will never be able to lift a grown elephant over his head.

[mschribr]

Maybe none of the above. Give the engine a time handicap and then we get a real game of Chess.

I agree. The best handicap is to cut the computer’s time in 1/2 or maybe 1/4.

I disagree.

I am even not sure if this is a handicap for the computer because the human can think less on the computer time.

rybka with 1/4 of the time beat easily chess programs that performed at 2700 level or better in tournaments.

I remember that old tiger and old shredder already got performance of more than 2700 some years ago.

Uri

Which tournament did rybka win with 1/4 of the time?
Computer vs computer is not the same as man vs computer.
The old tiger and old shredder performance of more than 2700 were not at handicap time.
Has there been man vs computer with time handicap?
Before we write off time handicap, lets have a man vs computer at ¼ the time and see what happens.
Mark

These tournaments were at relatively slower hardware than the hardware that rybka is using(tiger used only one processor) and I have no reason to think that computer vs computer is different than computer vs human.

If rybka with 1/4 of the time can beat tiger when both use one processor then I have no reason to think that the situation is going to be different against humans and that tiger is going to perform better against them.

Uri

Can you tell me where I can find more information about these time handicap tournaments?
When and where did they take place?
How many games were played?
What was the score?
Mark

[mschribr]

Maybe none of the above. Give the engine a time handicap and then we get a real game of Chess.

Hi Harvey,

This is one of the very few occasions when I strongly disagree with the contents of one of your posts: I think material handicap matches between humans and computers are vastly more interesting than time handicap matches.

We pretty much know that the strongest chess programs running on modern hardware are stronger than the best humans. The only thing we can hope to learn from time handicap matches between GMs and computers is how much a reduction in thinking time hurts the strength of a computer program. We already know the approximate answer to this question for games between two computers: A doubling in CPU speed gives somewhere around 60 Elo points. Maybe the constant is somewhat different for human vs computer games, but who cares?

Material handicap matches are much more interesting, because we can hope to learn something new about chess from them. How much is one pawn, two pawns or a knight worth in terms of Elo ratings? How big a material advantage would a top GM need in order to win against a perfect player? Nobody knows the answers to such questions, but experiments like the Rybka-Ehlvest match at least allow us to begin to speculate. That anyone can consider such a match to be less interesting than a time handicap match boggles my mind.

Tord

Hi Tord.
If you want to know how much a pawn or knight is worth then play a 100 games with 2 Rybkas, 1 Rybka with a pawn or knight less than the other Rybka. Then calculate the rating based on the 100 games. A computer vs computer match would give a more accurate value of the pieces because a computer would use its pieces more effectively than a grandmaster.
Mark