question to Prof Hyatt regarding milov vs rybka

[bob]

if you know some ones elo score you can predict percentage score against different opponents.

based on tests he did with rybka he found the exchange handicap worth about 400 elo points. milov elo is about 3700, so if the match would have been a 4 4 tie then its elo level would have been 3100 as it was 4.5 3.5 then he estimated 3050.

so let's say that 3050 elo means you can score 85% against kramink/anand then assuming this match is a representative sample of rybks's playing strength, and assuming the exchange handicap worth about 400 elo points (at the 2700 level [as he said it is worth more the higher the elo]) he would expect rybka to score 85% against kramink/anand .


that is what he seemed to be saying rather than he was in a position to judge chess skill.

I would like to hear your comment on this and what uri said.

My comment is this: what says that the Elo scale is linear at the upper or lower fringes of the range? What says that a pawn is worth 400? This is a _lot_ of speculation. I'd bet that between 1000-level players, a pawn means very little, while at the super-GM level, it is almost winning.

I'm not into this speculation stuff, and I don't consider it particularly interesting to play these handicapped matches. If people want to play 'em, and watch 'em, more power to 'em. I'm not among 'em however.

What a crap!

In Golf or Go handicaps are part of the game and in chess it's a known tradition. Already Morphy played it and sometimes he then lost such a game.

Grow up. In golf you do _not_ handicap players by say having one play two holes less. Players are rated, just like in USCF chess matches. And those are used to "equalize" the players by adjusting their scores by some delta value. It is not the same as removing a pawn or piece. And when you are competing in major tournaments, the handicap is _never_ used for anything.

Ok, so what is with Go? Is that also with a Delta?
(Charlie Gibson to loudmouth running mate)

Do you know what the handicapping systems do? They try to differentiate between player skill levels. They are _not_ used in any serious event. So what _are_ you talking about???

Please tone down or cool down as you might say down there. I'm specifically talking about such experimental events like this handicap etc match between Rybka 3 and the GM. Pllease try to concentrate on the topic and dont mention the same arguments that were already refutated. The IBM/DB guys are guilty, not Kasparov, all we should do now is exactly defining where and how they cheated. Cheated in my version not yours that implies something with justice issues. This is only for the sake of science methods, just like the political craziness that the USA is not man enough to accept The Hague for itself too but it wants to bring everybody else there. But such topics are for CTF. Here we talk about the science cheat.

And now Bob, no more blinking!

And no more wasting time, either. I'm done here...

[duncan]

you said before that 14 games by hydra against humans is not enough to establish a 3000 elo rating.

If Mr kaufman wanted to prove his rybka was better than the best human how many pawn handicap games would he need to draw against 2700 elp players before it could be reasonably assumed that it was.

(did deep thought get 26 consecutive wins against gm's before it was considered a grandmaster)

to challenge kramink/anand direct is not so feasible because of the huge funding costs.

I am not the duncan roberts you are referring to. I am a chess patzer.

[bob]

you said before that 14 games by hydra against humans is not enough to establish a 3000 elo rating.

If Mr kaufman wanted to prove his rybka was better than the best human how many pawn handicap games would he need to draw against 2700 elp players before it could be reasonably assumed that it was.

I don't know how to answer that question. Part of the equation depends on which pawn is removed. Another part needs to be answered by a _lot_ of games to see if a missing pawn turns the game more volatile, or more simple...

(did deep thought get 26 consecutive wins against gm's before it was considered a grandmaster)

The Fredkin rules required that a program produce a 2600+ rating over 24 consecutive games against FIDE GM opponents. It actually produced a 2650+ rating...

to challenge kramink/anand direct is not so feasible because of the huge funding costs.

I am not the duncan roberts you are referring to. I am a chess patzer.

[M ANSARI]

I think any handicap that alters the pieces or moves on the board is not really chess. It totally alters the balance of the game and has to be played in a totally different manner than classic chess ... and thus is not a valid interpretation of chess strength. But I do find some of the games interesting as the engine can be forced into some situations which could arise in classical chess and can show weaknesses and strengths. Of particular insterest was Milov's tremendous show of human strength when he found the correct plan to win the opposite bishop endgame where all engines thought it was drawing.

I do think that there is a place for handicap matches against engines ... but I think they should be by either using a time handicap or a White play only handicap ... or even no book or reduced book handicap. This way the pieces will not be altered from the original position and the centuries of chess theory and painstaking optimizations of engines will not be degraded.

[duncan]

for arguments sake, let us say that removing the f7 pawn made the game more volatile.

but since removing the f7 pawn at super gm level is almost inevitably a loss for the handicapped side then if a computer can consistently draw such a position against 2700 elo players while we would have no way of knowing its elo score, why cannot you conclude that the computer can play better and has a higher elo than any human ?

[Ovyron]

What do you think about handicaps in where Rybka plays unusual openings like a6 and h6 against anything?

[Uri Blass]

<snipped>

particular insterest was Milov's tremendous show of human strength when he found the correct plan to win the opposite bishop endgame where all engines thought it was drawing.

It is not correct that all engines thought it was drawing.
Rybka had always positive evaluation for black in analysis.

I do not know about a single engine that shows 0.00 evaluation.

Uri

[bob]

for arguments sake, let us say that removing the f7 pawn made the game more volatile.

but since removing the f7 pawn at super gm level is almost inevitably a loss for the handicapped side then if a computer can consistently draw such a position against 2700 elo players while we would have no way of knowing its elo score, why cannot you conclude that the computer can play better and has a higher elo than any human ?

I have not given this any great amount of thought, but my sort of glossed-over thinking says that removing the f7 pawn changes the character of the game significantly, perhaps better fitting the computer than the human. The human is continually trying to find ways to win the pawn back, when down a pawn, while the computer treats the position differently and such positions might better fit a computer than a human for that reason. For example, if you can do something to make the game more tactical, then the computer will benefit. Say a game where at one point, a player can make two moves in a row, but only once in a game. This would likely alter the balance in favor of the computer is a significant way since it would be a foreign idea to the human. So I am not sure what removing a pawn does, other than significantly altering the initial neutral balance of things...

[duncan]

I would be interested on your intial thoughts on the computer giving up a castle for a bishop.

[bob]

I would be interested on your intial thoughts on the computer giving up a castle for a bishop.

It is the same idea. playing down an exchange is different from a neutral start. Means you have to be careful when opening files since your opponent can dominate them with the extra rook. A program could certainly be tuned to play better in such handicapped games, but it would take some tuning to play optimally, which would be counter-productive when playing normal games.