Anti-Human feature

[reflectionofpower]

True, because a position with many pieces and less pawns is complex for
humans. But the engine might sac all of its minor pieces for a bunch of
pawns.


Probably including a counter for recaptures (or exchanges) on each variation
would do, plus a bonus for stopping the human for castling. Would be slower for engines to count their recaptures and evaluating the "sharpness" of the position.

Anti-GM from the Rebel Page. Would be effective against Grand-Father
http://www.rebel.nl/antigmex.htm#POS1

http://www.rebel.nl/antigmex.htm#POS1

[MikeGL]

True, because a position with many pieces and less pawns is complex for
humans. But the engine might sac all of its minor pieces for a bunch of
pawns.


Probably including a counter for recaptures (or exchanges) on each variation
would do, plus a bonus for stopping the human for castling. Would be slower for engines to count their recaptures and evaluating the "sharpness" of the position.

Anti-GM from the Rebel Page. Would be effective against Grand-Father
http://www.rebel.nl/antigmex.htm#POS1

http://www.rebel.nl/antigmex.htm#POS1

Thanks, my extra space at the end of the html link prevented the link from being displayed properly.

With anti-human featgure and recapture (exchanges) counter for engines, the following difficult Nolot #3 position might be easily considered by engines.
[d]r2qk2r/ppp1b1pp/2n1p3/3pP1n1/3P2b1/2PB1NN1/PP4PP/R1BQK2R w KQkq - 0 1

[MikeGL]

Creating an anti-human eval should be pretty easy, just add a small incentive for removing the first 6 pawns from the game.

I'd say:
* asymmetric material evaluation, to penalize exchanges (reward them from the opponent's pov).
* asymmetric pawn evaluation: penalize pawn rams (reward them from the opponent's pov).
* obviously use contempt (penalize draw by chess rules, reward them for the opponent).

should be more than enough...

Nice to hear that summary especially from a programmer.
Also, simply: reducing pawn value 0.9 for engines and increased to 1.1 for humans will encourage pawn exchanges, while increasing all piece values like 5.3 for engines rook and 5.0 for humans will let engines avoid piece exchanges.

And majority of the game will end in a pawnless endgame with many pieces which is very difficult for humans.

[lucasart]

Contempt =-DrawValue is important, to avoid trivial repetitions. But it's not enough to avoid draws against a typical human coward, who will block all the pawns, creating a fence dividing the board in two (and shuffle until 50 move draw).

You also want the engine to actively prevent that from happening. Pawn rams are the basic pattern you want to avoid. But then the engine will lose elo in engine vs. engine rating lists… You can't have everything.

[Lyudmil Tsvetkov]

Good idea, but does not work - SF with an antihuman feature would be simply a predecessor version, weaker in strength.

Same about Komodo.

SF with very big contempt or something like that is simply Crafty or an engine of similar strength. The further you raise the contempt, or introduce funny antipositional features, the weaker the engine becomes.

Engine should not play like that, while losing strength, but simply fix their positional problems over time.

Besides, not all antihuman features work against every human: for example, they would not work against Tal, Shirov, etc.

Other, even more important thing to consider: a knowledgeable human player can of course adapt over time to an antihuman strategy, so that he capitalises on most engine mistakes.

Changing the engine while losing strength is not the way to go, but rather eliminating engine deficiencies.

[MikeGL]

But with the brutal accuracy of current top engines, after analyzing some games of Tal with his sacrifices, one can conclude that those are actually unsound sacrifices and can be defended with accurate play.

Actually it will make the engine weaker against engines, but some strange behaviour was discovered by Rebel programmer too.


http://www.top-5000.nl/int/aaron.htm

[reflectionofpower]

But with the brutal accuracy of current top engines, after analyzing some games of Tal with his sacrifices, one can conclude that those are actually unsound sacrifices and can be defended with accurate play.

Actually it will make the engine weaker against engines, but some strange behaviour was discovered by Rebel programmer too.


http://www.top-5000.nl/int/aaron.htm

http://www.top-5000.nl/int/aaron.htm

[reflectionofpower]

But with the brutal accuracy of current top engines, after analyzing some games of Tal with his sacrifices, one can conclude that those are actually unsound sacrifices and can be defended with accurate play.

Of course ... but we are talking about humans playing (C). This is where you have to open it up and play aggressively.

Actually it will make the engine weaker against engines, but some strange behaviour was discovered by Rebel programmer too.

I agree with you there too.


http://www.top-5000.nl/int/aaron.htm

[lucasart]

Good idea, but does not work - SF with an antihuman feature would be simply a predecessor version, weaker in strength.

Same about Komodo.

Got proof ?

[AdminX]

How about a "Let's Check" type algorithm that auto detects if a human is playing and would auto switch to anti- human mode on the fly.?