How to Make Computer Chess More Interesting

[Norm Pollock]

I have some suggestions to authors of GUIs in order to make computer chess games more interesting and possibly fairer.

What bores me the most about computer chess are the seemingly endless games that will obviously end in a draw provided no major blunder occurs. Human games do not have that issue because the players would agree to a draw.

One solution is to play faster time control, but then the moves come too fast and the quality of the moves is not as good they could be.

Here are a couple of options that I would like to see:

1. Instead of an adjudicated draw if both players have an evaluation of exactly 0.00 for a specified number of moves, give the user the option to use an interval. For example the user could choose an adjudicated draw if the both evaluations are from -0.05 to +0.05 for a specified number of moves.

2. Instead of letting games go on endlessly in a brutal wait for the "50 move" rule to apply, allow the user to specify a cut-off move number that will force a "replay". A "replay" is better than adjudicating a draw because sometimes the game will eventually result in a win if allowed to finish. For example, the user could specify that after "100" moves, replay the game. This option should be enhanced into allowing extra moves if either side has an evaluation greater than a specified value. For example, if after 100 moves, one side has an evaluation of +5.00, then play on for another 20 moves. This would apply to those long "syzygy" mates.

An existing option already out there is one that I have mixed feelings about. It is adjudicating a win when both evaluations are beyond a pre-selected value for a fixed number of moves. But as good as this option is, it really doesn't shorten the game much because a mate is on the horizon, and it deprives the spectator of seeing the mate. And rarely, but it does happen, a game is adjudicated as a win even though the 50-move rule is about to apply.

Another issue is making tournaments "fairer".

There are a couple of excellent options out there already, such as reversing colors on the opening. Maybe it's me, but in Arena I can only do this if I use an "epd" opening suite. If there is NOT already a way to do this with a "pgn" opening book, then that should be an option.

An option I would like to see is to use the same opening position for all games in the same Round of a Tournament. Then if an opening is extremely "drawish", then all players in the tournament have to deal with it that ROUND even though they have different opponents. Likewise if an opening is "open", all players play it that ROUND.

-Norm

[yurikvelo]

you want Arena to implement

Code: Select all

-draw <n> <score>

option of cli-cutechess in Tournament mode?

[carldaman]

I have some suggestions to authors of GUIs in order to make computer chess games more interesting and possibly fairer.

What bores me the most about computer chess are the seemingly endless games that will obviously end in a draw provided no major blunder occurs. Human games do not have that issue because the players would agree to a draw.

One solution is to play faster time control, but then the moves come too fast and the quality of the moves is not as good they could be.

Here are a couple of options that I would like to see:

1. Instead of an adjudicated draw if both players have an evaluation of exactly 0.00 for a specified number of moves, give the user the option to use an interval. For example the user could choose an adjudicated draw if the both evaluations are from -0.05 to +0.05 for a specified number of moves.

2. Instead of letting games go on endlessly in a brutal wait for the "50 move" rule to apply, allow the user to specify a cut-off move number that will force a "replay". A "replay" is better than adjudicating a draw because sometimes the game will eventually result in a win if allowed to finish. For example, the user could specify that after "100" moves, replay the game. This option should be enhanced into allowing extra moves if either side has an evaluation greater than a specified value. For example, if after 100 moves, one side has an evaluation of +5.00, then play on for another 20 moves. This would apply to those long "syzygy" mates.

An existing option already out there is one that I have mixed feelings about. It is adjudicating a win when both evaluations are beyond a pre-selected value for a fixed number of moves. But as good as this option is, it really doesn't shorten the game much because a mate is on the horizon, and it deprives the spectator of seeing the mate. And rarely, but it does happen, a game is adjudicated as a win even though the 50-move rule is about to apply.

Another issue is making tournaments "fairer".

There are a couple of excellent options out there already, such as reversing colors on the opening. Maybe it's me, but in Arena I can only do this if I use an "epd" opening suite. If there is NOT already a way to do this with a "pgn" opening book, then that should be an option.

An option I would like to see is to use the same opening position for all games in the same Round of a Tournament. Then if an opening is extremely "drawish", then all players in the tournament have to deal with it that ROUND even though they have different opponents. Likewise if an opening is "open", all players play it that ROUND.

-Norm

[Arena does allow you to load a pgn openings suite, so the same opening will be used by both engines with white and black. I use this feature all the time.]

Good suggestions, Norm. Adjudication rules are primitive in most GUIs, unfortunately. Also, the engine protocols themselves could be greatly improved. If you could make the engines aware of the opponent's relative strength, then contempt could be managed on the fly, with great results - translating into enhanced quality of play and reduced draws. At the moment, pretty much every top engine uses UCI, but this is a rudimentary protocol that prevents such automated tweaking.

I'm a big fan of match-ups between top engines and lesser engines, as they can often be spectacular. Contempt here comes in very handy, for the engines that have the functionality. It can be handled manually for any single match, so I like to do matches rather than tournaments.

Other tweakable parameters can be very interesting, IF the engine provides them. Engine personalities can be created that score much better vs lower rated engines that the default settings, winning almost every game against engines rated 300-400 below them, and thus performing better than the Elo difference would suggest.

Computer chess does not have to be boring.

Cheers,
CL

[hgm]

There are a couple of excellent options out there already, such as reversing colors on the opening. Maybe it's me, but in Arena I can only do this if I use an "epd" opening suite. If there is NOT already a way to do this with a "pgn" opening book, then that should be an option.

In WinBoard you can do this even when using the GUI book. (And of course also when you play from FEN or PGN files.) Even when you use concurrency, as all games will share the same random seed, even when playing in different agents.

An option I would like to see is to use the same opening position for all games in the same Round of a Tournament. Then if an opening is extremely "drawish", then all players in the tournament have to deal with it that ROUND even though they have different opponents. Likewise if an opening is "open", all players play it that ROUND.

In WinBoard, when you play from FEN or PGN, you would get the same opening in the entire CYCLE of a round robin. Each pairing numbers the games in it, and all first games of any pairing would use the same line, etc.

I am not sure how what you describe could be fair. If I play a stronger opponent in a round with a drawish opening, and my competitor playes a weaker one, we both draw. If in the next round the pairings are reversed, and we get an 'open' opening, I win, and my competitor loses.

If you could make the engines aware of the opponent's relative strength, then contempt could be managed on the fly, with great results - translating into enhanced quality of play and reduced draws. At the moment, pretty much every top engine uses UCI, but this is a rudimentary protocol that prevents such automated tweaking.

Alsthough I agree with your characterisation of UCI as a 'rudimentary protocol', fairness demands that I point out that there actually is a UCI standard option that does exactly this: UCI_Opponent. When you run UCI engines under WinBoard + Polyglot, engines supporting this option will be informed of the name of their opponent, and whether this is a human or a computer.