An idea to kill draws in computer chess by different rules

[Uri Blass]

The idea is the following.

The game has 2 stages
stage 2 is a normal game when one of the sides need to win and the other need to draw.

stage 1 is to decide the position you start at stage 2.
In every move of stage 1 both side send their evaluation of the expected result of the game after all the legal moves to the interface.
If both sides send the same information the interface send the side to move choose a move and the game continue in stage 1.
If both sides do not send the same information then the side to move has to protect the evaluation in one of the moves when there is a disagreement
when the game continue in stage 2.

The side to move has a free choice of what moves to choose.

Here is an example

For example suppose white send 1.e4 wins, 1.g4 and 1.h4 lose when other moves draws and
black send 1.e4 and 1.g4 lose and 1.f3 lose and other moves draw.

The interface tell white the choices of black and
In this case white has the following options to continue at stage 2 that is a normal chess game

1)play white with 1.e4 when white need to win(every other result is a loss)
2)play black against 1.h4 with the need to win(every other result is a loss)
3)play white with 1.f3 with the need to get at least a draw(draw is considered a win for white in this case)

If the players always agree about the theoretical result and the game is finished in stage 1 then black wins
Programmers will need to modify their engines to allow them to play this type of game.

The idea can be used not only for chess but for every game that is not theoretically solved but practically solved in the meaning that the result of almost every game between top engines is a draw.

[mvanthoor]

To be honest, I think it is somewhat convoluted. It seems the only way to make chess harder for computers is to make it also harder for humans.

I've always wondered at the simplicty of Go as compared to chess:

The rules (except for some very special situations, I think these are all of them...)
- You can move anywhere you want
- As long as you don't kill your own stones and don't move on top of another stone
- Repeating captures (ko) are not allowed. One has to make a different move to change the board first.
- Stones never move when placed (except when captured)

Avoiding draws: After Komi was introduced, a game can never be a draw.

White starts with 6 points ahead, because black's first move advantage (which is 5-7 points worth; opinions differ).
Because black has sente (initiative; can dictate how the game starts), white starts out in a defensive position. White is compensated for this with another half point in Komi: If the game actually ends up a draw with both sides having an equal amount of territory, white wins.

Even though I like chess better (in the end) because of the different pieces, I think chess will never be as elegant with regard to rules and avoiding draws as Go is.

[Ras]

I don't even like the underlying idea. In chess, one player can only win if the other player makes a mistake. In absence of mistakes, i.e. with perfect play, draw is the natural result - and not a problem to be solved.

[Ovyron]

I think chess will never be as elegant with regard to rules and avoiding draws as Go is.

I think that this area has been largely unexplored, as there hasn't been a good incentive to create a good draw-less variant of chess.

Here's another idea:

If the game ends in a draw, white is considered the winner if the draw is achieved before move N, and black is considered the winner if the draw is achieved at move N or after. <- Not very convoluted, is it?

I hold that there's some N that would produce fair games where both sides have equal winning chances from the start, and where a skilled player could beat the other no matter what side they start with, by forcing a draw before N or extending the game beyond N.

There's probably hundreds of ideas like this that people haven't yet thought about, but if they did we could find one more elegant than Go's this very year.

[MonteCarlo]

I've put way too much time into chess and enjoy it too much to spend a lot of time in other games, but in terms (complexity of play)/(simplicity of rules), I've always been even more partial to hex than go.

It has a much smaller community than either chess or go, though.

[lkaufman]

I don't even like the underlying idea. In chess, one player can only win if the other player makes a mistake. In absence of mistakes, i.e. with perfect play, draw is the natural result - and not a problem to be solved.

I agree that a perfectly played game should be a draw. The problem is that with current rules, one player must make either several mistakes or one very big mistake to lose. Ideally any mistake should lose, a draw should result only from ideal play which no one should ever be able to achieve. Or at least draws should result from only nearly perfect play. That's what this is all about.

[lkaufman]

I think chess will never be as elegant with regard to rules and avoiding draws as Go is.

I think that this area has been largely unexplored, as there hasn't been a good incentive to create a good draw-less variant of chess.

Here's another idea:

If the game ends in a draw, white is considered the winner if the draw is achieved before move N, and black is considered the winner if the draw is achieved at move N or after. <- Not very convoluted, is it?

I hold that there's some N that would produce fair games where both sides have equal winning chances from the start, and where a skilled player could beat the other no matter what side they start with, by forcing a draw before N or extending the game beyond N.

There's probably hundreds of ideas like this that people haven't yet thought about, but if they did we could find one more elegant than Go's this very year.

More elegant than an artificial move number is some point-count to break ties, as is already done in both shogi and janggi. It could be just a count of material, in which case the main issue is whether to treat bishops as more than knights, or only if you have two bishops. Or it could also consider the rank and/or file of pawns, since these only change with capture or pawn move. There should be some point count rule, with komi for Black, that is as close to fair as the komi rule is for Go. Possibly just summing the ranks of the pawns, with appropriate point values for the pieces, with Black winning ties would be good enough, or perhaps Black needs 1.5 Komi to be fair.

[Ras]

The problem is that with current rules, one player must make either several mistakes or one very big mistake to lose.

Yes, and a lot of endgame knowledge is based on that. Also, I'm just an hobby player, but I've heard that top masters don't just exploit mistakes of opponents. Instead, they make their opponents make mistakes. This seems to be the secret.

Some deep games of LC0 vs. SF had SF losing despite large search depth, and the kicker is that SF thought all was fine for a long time. So it is indeed possible.

Also, what's even the problem with computer draws? Human's can't do hundreds of games in a match, computers can. So what if 480 out of 500 games are draws? Discard them and done.

[Ovyron]

It could be just a count of material, in which case the main issue is whether to treat bishops as more than knights, or only if you have two bishops

No, the main issue here is "why not the opposite?"

The most beautiful games I've seen in chess have included giving away a lot of material to mate the king. If the rule was that the side with least material got the win in a draw, it'd produce more beautiful games, as there are already beautiful games in chess where a side with a disadvantage goes out with a bang, sacrificing material to force a perpetual.

The chess engine Rybka 1.0 Beta was commended for her playing style where she would treat her pawns like rubbish and give them away for positional factors.

The most examined Alpha Zero games have been the ones where the AI just threw away its material for seemingly no reason and held the game or gained strategical advantages.

It seems we'd want a game where we have more frequent gambits and sacrifices that are spectacular and give the win to the side that plays them and holds the draw, instead of giving wins to solid and materialistic players that take no risk but manage to snatch a pawn.

[Alayan]

I agree that a perfectly played game should be a draw. The problem is that with current rules, one player must make either several mistakes or one very big mistake to lose. Ideally any mistake should lose, a draw should result only from ideal play which no one should ever be able to achieve. Or at least draws should result from only nearly perfect play. That's what this is all about.

In other words, the draw margin in chess is too wide, at least for engines and top players.

It seems we'd want a game where we have more frequent gambits and sacrifices that are spectacular and give the win to the side that plays them and holds the draw, instead of giving wins to solid and materialistic players that take no risk but manage to snatch a pawn.

Indeed.