Why handicapping a machine might fail to succeed

[Rolf]

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.

This is base on a false concept of balance of classical chess. Also there you have all kind of square situations. Nobody would care if it's intentiously played by a master or resulting by chance. But if it's set in the beginning it shouldnt be chess? Apparently this a false conclusion. If made by weaker amateurs it's even more illogical from the spirit of the game.

Chess is always about problem-solving in a concrete position. With a concrete distribution of material.

Of course the long-term consequences of imbalanced material favors the human player at first, whether he has material advantages or disadvantages, but also a program can be tuned for such chess variations. It's even a matter of fact that a machine can play much better material imbalances in the endgame. Not many human players can win with Q vs R against the R of a machine. GM Nunn has shown lengthly endgames with N whose machine-wise roblem-solving cannot be understood by us humans. Understanding in that sense that we could learn how to play it. We dont.

In that context we could even ask if a seeming handicapping of a machine does really disadvantage it the whole game through. In the endgame a machine has a technical "book" advantage. And of calculations too.

To end this with an extreme. Even a handicapping by allowing the human player a one-time correcting a disadvantaging move say maximally for a 10-move intervall, to equalise calculation somehow, wouldnt change the character of the classical problem-solving situation of chess.

If we dont want to accept that, and there are also reasons for that decision, we could forbid games between machines and human players as such, because, this was also the key reason for the clash in 1997, it's always a hoax, unless the machine isnt organising its own play on its own the whole event through.

If human operators are interfering it's always the constellation of a smart human problem-solver, a naive calculating monster machine and some vicious hiuman operators who can try to disrupt the normal flow of the game with all kind of (known and often unidentifiable) cheats. The easiest is to disturb the concentration of the human player, which we called psyching-out. The highest form of evil is reached if the operators announce that from now on they would use everything in connect with dirty ressources. Of course then it all doesnt make sense anylonger, except taking the money...because no normal human chessplayer is made to apply his thought process in such an imballanced and of course highly unfair situation.

The psychological situation is actually here in such a forum already that the machine operators tend to argue that if there were no space to gamble and at best to literally cheat, that THEN a competition between a holy machine and a smart human being doesnt make sense. The operators, favorably with the second best human player, do only hold the event open against a vicious human champion who is only the best (allegedly - in the bad fantasies of the chess lay operators) because he also cheats all the time. He uses his eidetic memory. He hides his preparation and he gets unallowed help from God. But first of all he doesnt tell what he's thinking. So, therefore everything is allowed against such human falsehood...

I would enjoy feedback that this isnt yet the case except in the "match" in 1997. <cough>

[M ANSARI]

I am sure chess engines "could" be tuned to play much better in some of the handicap games ... simple rules like castling long or even castling at all might have to be dramatically changed. Also more emphasis could be made on keeping exchanges to a minimum and not allowing a blocked position even if it means material concessions. That engine would play much better against a human, but would play much weaker chess in general. The handicap where a human is allowed to take back a move or even 3 moves back when he misses a tactical shot is reasonable ... anything that would keep all the pieces on the board before the start of the game would be acceptable. I also think that removing the engine book entirely or maybe keep it to only a few moves is also an interesting alternative. I wouldn't want to remove all book moves because then the GM could play many games at home and he would have an entire map of moves the engine would play. What is sure is that humans, even at the highest levels, cannot dream of competing mano a mano against even a dual core Rybka 3.

[Uri Blass]

<snipped>

I am sure chess engines "could" be tuned to play much better in some of the handicap games ... simple rules like castling long or even castling at all might have to be dramatically changed. Also more emphasis could be made on keeping exchanges to a minimum and not allowing a blocked position even if it means material concessions. That engine would play much better against a human, but would play much weaker chess in general.

I think that you underestimated top human players if you think that chess engines can perform better against them by playing much weaker chess in general.

You may be right if the target is to score 100% and not 99% against players like pablo but I believe that you are not right if the target is to beat top GM's.

Uri

[BubbaTough]

I think that you underestimated top human players if you think that chess engines can perform better against them by playing much weaker chess in general.

You may be right if the target is to score 100% and not 99% against players like pablo but I believe that you are not right if the target is to beat top GM's.

Uri

It is not about humans vs. computers, it is about stronger vs. weaker. It all just amounts to working on contempt. And the better Rybka gets compared to its opposition (whether they are human or computer) the more important an effective contempt value is. Rybka has a better version of contempt than it did (Vas chatted with me about his contempt strategy for Rybka 3 a few months ago which apparently is working better than I thought it would) but there is still a lot of progress to be made in this area.

-Sam

[Rolf]

<snipped>

I am sure chess engines "could" be tuned to play much better in some of the handicap games ... simple rules like castling long or even castling at all might have to be dramatically changed. Also more emphasis could be made on keeping exchanges to a minimum and not allowing a blocked position even if it means material concessions. That engine would play much better against a human, but would play much weaker chess in general.

I think that you underestimated top human players if you think that chess engines can perform better against them by playing much weaker chess in general.

You may be right if the target is to score 100% and not 99% against players like pablo but I believe that you are not right if the target is to beat top GM's.

Uri

Please could you tell me what you meant with the by me underligned paragraph, Uri? I dont get the sense. Foreign language.

And then I want to state that I see both MAnsari and Bob in a wrong if they see a handicapped machine leading to a change

-- in balance (MAnsari) and
-- in character (Bob)

of the game of chess.

Could you explain this?

I tried to elaborate that the core of chess doesnt change at all, as a problem-solving in concrete positions. So character doesnt change at all, and as far as balance is concerned, it will be a matter of experience to define what difference in material or position could "heal" what difference in strength.

Of course it should be seen that human GM are not prepared, neither willing, to damage their own chess abilities by suddenly playing such odd variations of chess. And it's also negative to prepare on such chess for too long.

GM Milov was really nice in playing Rybka 3 but I doubt that he's a true super top GM because I never saw him play in one of the few top tournaments. The other way round one could say that he might be as strong as the top players, even better adapted to computerchess than bigger names. So, I only want to mention that the question of the true strength, apart from the factual definitions by super tournament results, must be considered. However, another aspect, the big money flows only for the big names no matter if they really are interested in computerchess games.

[Ovyron]

One step backwards was mixing open/closed positions with contempt, as it was apparent, there are occasions in where Rybka would wish to open the position and like to draw or close the position and want to win, but this is currently not possible.

[Uri Blass]

<snipped>

I think that you underestimated top human players if you think that chess engines can perform better against them by playing much weaker chess in general.


Please could you tell me what you meant with the by me underligned paragraph, Uri? I dont get the sense. Foreign language.

I meant that top humans are strong enough not to be fooled by significantly weaker anti-human moves.

My opinion is that programs can play slightly weaker in general(20 elo weaker against chess programs) and play relatively better against top humans thanks to playing style but they cannot play significantly weaker and still be better against humans.

If they play significantly weaker then the best chess players can learn to take advantage of their weaknesses.

Uri

[Rolf]

<snipped>

I think that you underestimated top human players if you think that chess engines can perform better against them by playing much weaker chess in general.


Please could you tell me what you meant with the by me underligned paragraph, Uri? I dont get the sense. Foreign language.

I meant that top humans are strong enough not to be fooled by significantly weaker anti-human moves.

My opinion is that programs can play slightly weaker in general(20 elo weaker against chess programs) and play relatively better against top humans thanks to playing style but they cannot play significantly weaker and still be better against humans.

If they play significantly weaker then the best chess players can learn to take advantage of their weaknesses.

Uri

Now this makes sense to me, I mean logically. How far this is exactly true, that is something I cant judge. Also I dont understand your different take on either machine opponents and human ones. But perhaps the question comes again in a better context. Thanks so far, Uri.

[Dirt]

Inspired by the Bilbao scoring, here is a handicap that would give a GM a fair chance but would result in real chess games. Just use the following point table:

Code: Select all

           Rybka      GM
Lose         0         0
Draw         1         3
Win          2         4

If the human never wins this is equivalent to draw odds, yet there is still an incentive to try for the win.

[swami]

Inspired by the Bilbao scoring, here is a handicap that would give a GM a fair chance but would result in real chess games. Just use the following point table:

Code: Select all

           Rybka      GM
Lose         0         0
Draw         1         3
Win          2         4

If the human never wins this is equivalent to draw odds, yet there is still an incentive to try for the win.

If Rybka gets a draw, human deserves 3 points? But if Rybka wins, only 2 points will be rewarded? Don't you think that's slightly unfair.

I would slightly modify your version and propose this scoring instead which would make sense mathematically:

Code: Select all

           Rybka      GM
Lose         0         0
Draw         1         2
Win          2         4