Page 1 of 11

AlphaZero No Castling Chess

Posted: Tue Dec 03, 2019 4:30 pm
by Javier Ros
It seems that AlphaZero has been trained to play "no-castling chess", see the interesting article

Kramnik And AlphaZero: How To Rethink Chess‎

at

https://www.chess.com/article/view/no-c ... -alphazero

Re: AlphaZero No Castling Chess

Posted: Tue Dec 03, 2019 5:21 pm
by dragontamer5788
It seems odd that they didn't discuss the draw-rate of this variant. Or did I miss it? The two example games listed there are draws.

I think Chess should have sharper games overall with a lower draw rate. Armageddon chess is a good step forward, but seems to favor Black. There have been other discussions about other rulesets similar to Armageddon (complicating the rules: 50-move == White Win, Stalemate == Black Win, etc. etc.)... but finding a set of Chess rules with fewer draws and more balanced gameplay should be the goal.

----------

The idea of using AlphaZero as a "easily trained expert" for new chess variants seems like a good methodology. Perhaps we could encode this as a search problem: how to change the rules of chess to increase win/loss rate of experts and decrease the draw rate.

AlphaZero Returns With "No Castling Chess"!

Posted: Tue Dec 03, 2019 9:29 pm
by supersharp77
Javier Ros wrote:
Tue Dec 03, 2019 4:30 pm
It seems that AlphaZero has been trained to play "no-castling chess", see the interesting article

Kramnik And AlphaZero: How To Rethink Chess‎

at

https://www.chess.com/article/view/no-c ... -alphazero
The Return Of Alpha Zero With No Castling Chess! When Will The LC0 vs Alpha Zero Chess Championship match be Played?
The Chess World Awaits This Great Match :D :wink:
Alpha Zero (A) vs Alpha Zero (B)

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 2:48 am
by Ovyron
dragontamer5788 wrote:
Tue Dec 03, 2019 5:21 pm
I think Chess should have sharper games overall with a lower draw rate.
I just thought about this variant the other day:

If the game ends in draw, the first player that checked the other during the game is awarded a win.

There we go, no more draws and no more hitting our heads against the wall figuring out draw odds. Just, a check and defend to avoid losing and you got the game.

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 6:50 am
by lkaufman
Ovyron wrote:
Wed Dec 04, 2019 2:48 am
dragontamer5788 wrote:
Tue Dec 03, 2019 5:21 pm
I think Chess should have sharper games overall with a lower draw rate.
I just thought about this variant the other day:

If the game ends in draw, the first player that checked the other during the game is awarded a win.

There we go, no more draws and no more hitting our heads against the wall figuring out draw odds. Just, a check and defend to avoid losing and you got the game.
The last player to check rather than the first makes more sense to me as a tiebreaker, with Black winning in case no one ever gave check. But I think there are much better ways to tiebreak that are more in the spirit of chess. Ideally making repetition illegal, stalemate a win for the stalemating side, and bare king a loss, with Black winning fifty move rule draws, could work, but I think it would still favor Black too much. It could be tested with engines, and if Black does win too much, further tiebreak rules could be added until it becomes about fair.

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 9:51 am
by Ovyron
lkaufman wrote:
Wed Dec 04, 2019 6:50 am
But I think there are much better ways to tiebreak that are more in the spirit of chess. Ideally making repetition illegal, stalemate a win for the stalemating side, and bare king a loss, with Black winning fifty move rule draws, could work, but I think it would still favor Black too much. It could be tested with engines, and if Black does win too much, further tiebreak rules could be added until it becomes about fair.
I'm looking for a simple solution, one you can explain to a child that just learned to play chess. All those things are very complicated and arbitrary (a position could have been won for the other side, for no reason!)

I just played 3-check chess the other day and it can get very intense, but a problem it has is that once you can give a check and increase the chances that you'll make two more, players start making very unlike-chess moves. But what if there was an optimal amount of checks to give that would solve the problem?

I'm not talking about no-draws here, and I'm not talking about using this for tie breaking, I'm saying that this subset of chess makes sense, because if your advantage is strong enough to win the game, then surely it's strong enough to deliver the amount of checks needed also.

Players would start with a "Life Bar", say, 10 Hit Points for their king. They're playing a normal chess game, but every time their king is checked, they lose a point, and once it's 0 the other player wins. If the game ends in draw the player with most Hit Points wins. If you are up on Hit Points and the opponent can't deliver a check anymore they can resign already.

Such a thing is easy to explain and just have normal chess with a few words added to the win condition:

Checkmate the opponent king, give it 10 checks or be the one that gave the most checks if the game ends in draw.

Hmm, turns out even simpler would be to count how many checks happened in the game and award a win to the player that gave the most ones. Would this change so much the nature of the game?

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 10:27 am
by Raphexon
Hard to see how this would increase the amount of decisive matches.

No castling cuts down the search tree quite hard so it should be much easier for an engine to see the forced win or forced draw.

Maybe for human chess it can make the game a lot more decisive because positions become sharper, but it gets a lot easier for Stockfish and the gang.

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 3:27 pm
by the_real_greco
There have been a lot of crazy rule changes to chess during the last thousand years... but they occurred because people started playing the new variant, not discussing it.

Did Kramnik actually play a no-castling game? I see a bunch of A0-A0 games but nothing with him. Which is not exactly a ringing endorsement of no-castling.

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 6:20 pm
by BeyondCritics
Chessbase India has produced an interview, where Kramnik share his thoughts about Nocastling chess and demonstrates a beautiful Alpha Zero self play game.

Re: AlphaZero No Castling Chess

Posted: Wed Dec 04, 2019 9:15 pm
by carldaman
lkaufman wrote:
Wed Dec 04, 2019 6:50 am
Ovyron wrote:
Wed Dec 04, 2019 2:48 am
dragontamer5788 wrote:
Tue Dec 03, 2019 5:21 pm
I think Chess should have sharper games overall with a lower draw rate.
I just thought about this variant the other day:

If the game ends in draw, the first player that checked the other during the game is awarded a win.

There we go, no more draws and no more hitting our heads against the wall figuring out draw odds. Just, a check and defend to avoid losing and you got the game.
The last player to check rather than the first makes more sense to me as a tiebreaker, with Black winning in case no one ever gave check. But I think there are much better ways to tiebreak that are more in the spirit of chess. Ideally making repetition illegal, stalemate a win for the stalemating side, and bare king a loss, with Black winning fifty move rule draws, could work, but I think it would still favor Black too much. It could be tested with engines, and if Black does win too much, further tiebreak rules could be added until it becomes about fair.
What would be more drastic - changing the rules whereby how the game is won, or altering the starting position, or just the castling rules?

It looks to me that the first option (changing the scoring rules) is way too radical and would mess with the fabric of the game. I know there was a time when bare king or stalemate were counted as a win, but why were those rules changed? It had to be because the current rules work better than the old ones.

For instance, making stalemate a loss would directly affect all (piece &) pawn endgames, but by extension the whole game. People would hesitate to take any chances because the loss of one pawn would doom the player to an almost certain loss in the endgame. Play would become more sterile and materialistic, basically killing the game.