Another possible toy for AI programmers: 4-player chess

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Tony P.
Posts: 81
Joined: Sun Jan 22, 2017 7:30 pm
Location: Russia

Another possible toy for AI programmers: 4-player chess

Post by Tony P. » Thu Sep 21, 2017 3:17 pm

So far, it doesn't look like the chess.com version of four player chess has gained much interest, but at least there's an influx of newbies into the variant (a substantial portion of the players in the pool at any given moment are rated 1200, i.e. playing their first ever game in the variant), and I hope that the company succeeds at the promotion of this variant.

Of course 4-player chess in general isn't new at all (chaturanga itself was 4-player), but what's new is that a big company seems determined to inject some marketing budget into it and improve the rules, maybe even make it playable for real money on a good day 8-) (this looks doable because it's possible to modify the rules often to disrupt engines).

What's attractive in the variant from a chess dev's viewpoint is that it's multiplayer, thus there are many different equilibrium concepts instead of the well-understood Nash equilibrium of 2-player zero-sum games. And it's played with familiar chess pieces.

Most of the algorithms designed for 2-player chess won't work in 4-player one. This means that those who've lost the hope to beat the top 3 engines in 2-player chess may try to conquer 4-player one by writing a program from the ground up, in the field of multiplayer games that has been explored outside chess but, so far, not much within the chess community.

Any takers? :wink:

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Another possible toy for AI programmers: 4-player chess

Post by Lyudmil Tsvetkov » Thu Sep 21, 2017 4:48 pm

Tony P. wrote:So far, it doesn't look like the chess.com version of four player chess has gained much interest, but at least there's an influx of newbies into the variant (a substantial portion of the players in the pool at any given moment are rated 1200, i.e. playing their first ever game in the variant), and I hope that the company succeeds at the promotion of this variant.

Of course 4-player chess in general isn't new at all (chaturanga itself was 4-player), but what's new is that a big company seems determined to inject some marketing budget into it and improve the rules, maybe even make it playable for real money on a good day 8-) (this looks doable because it's possible to modify the rules often to disrupt engines).

What's attractive in the variant from a chess dev's viewpoint is that it's multiplayer, thus there are many different equilibrium concepts instead of the well-understood Nash equilibrium of 2-player zero-sum games. And it's played with familiar chess pieces.

Most of the algorithms designed for 2-player chess won't work in 4-player one. This means that those who've lost the hope to beat the top 3 engines in 2-player chess may try to conquer 4-player one by writing a program from the ground up, in the field of multiplayer games that has been explored outside chess but, so far, not much within the chess community.

Any takers? :wink:
once standard chess is solved, I am taking up 4-men chess.

Leo
Posts: 839
Joined: Fri Sep 16, 2016 4:55 pm
Location: USA/Minnesota
Full name: Leo

Re: Another possible toy for AI programmers: 4-player chess

Post by Leo » Thu Sep 21, 2017 5:28 pm

Lyudmil Tsvetkov wrote:
Tony P. wrote:So far, it doesn't look like the chess.com version of four player chess has gained much interest, but at least there's an influx of newbies into the variant (a substantial portion of the players in the pool at any given moment are rated 1200, i.e. playing their first ever game in the variant), and I hope that the company succeeds at the promotion of this variant.

Of course 4-player chess in general isn't new at all (chaturanga itself was 4-player), but what's new is that a big company seems determined to inject some marketing budget into it and improve the rules, maybe even make it playable for real money on a good day 8-) (this looks doable because it's possible to modify the rules often to disrupt engines).

What's attractive in the variant from a chess dev's viewpoint is that it's multiplayer, thus there are many different equilibrium concepts instead of the well-understood Nash equilibrium of 2-player zero-sum games. And it's played with familiar chess pieces.

Most of the algorithms designed for 2-player chess won't work in 4-player one. This means that those who've lost the hope to beat the top 3 engines in 2-player chess may try to conquer 4-player one by writing a program from the ground up, in the field of multiplayer games that has been explored outside chess but, so far, not much within the chess community.

Any takers? :wink:
once standard chess is solved, I am taking up 4-men chess.
You wont be alive for that event.
Advanced Micro Devices fan.

CheckersGuy
Posts: 273
Joined: Wed Aug 24, 2016 7:49 pm

Re: Another possible toy for AI programmers: 4-player chess

Post by CheckersGuy » Thu Sep 21, 2017 10:23 pm

Leo wrote:
Lyudmil Tsvetkov wrote:
Tony P. wrote:So far, it doesn't look like the chess.com version of four player chess has gained much interest, but at least there's an influx of newbies into the variant (a substantial portion of the players in the pool at any given moment are rated 1200, i.e. playing their first ever game in the variant), and I hope that the company succeeds at the promotion of this variant.

Of course 4-player chess in general isn't new at all (chaturanga itself was 4-player), but what's new is that a big company seems determined to inject some marketing budget into it and improve the rules, maybe even make it playable for real money on a good day 8-) (this looks doable because it's possible to modify the rules often to disrupt engines).

What's attractive in the variant from a chess dev's viewpoint is that it's multiplayer, thus there are many different equilibrium concepts instead of the well-understood Nash equilibrium of 2-player zero-sum games. And it's played with familiar chess pieces.

Most of the algorithms designed for 2-player chess won't work in 4-player one. This means that those who've lost the hope to beat the top 3 engines in 2-player chess may try to conquer 4-player one by writing a program from the ground up, in the field of multiplayer games that has been explored outside chess but, so far, not much within the chess community.

Any takers? :wink:
once standard chess is solved, I am taking up 4-men chess.
You wont be alive for that event.
I think he is implying that he is never going to work on 4 man chess :lol:

Daniel Shawul
Posts: 3758
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Another possible toy for AI programmers: 4-player chess

Post by Daniel Shawul » Fri Sep 22, 2017 12:53 am

Nebiyu alien can play 4-player chess and other multi-player games. I spent quite a lot of time on this a few years ago studying algorithms, developing a GUI and CLI game managers, game servers etc which was fun. There is also an android app : https://sites.google.com/site/dshawul/home which I am not sure works now due to java issues but you can try it. A triple-chess in action using javaboard.

Image


Daniel

CheckersGuy
Posts: 273
Joined: Wed Aug 24, 2016 7:49 pm

Re: Another possible toy for AI programmers: 4-player chess

Post by CheckersGuy » Fri Sep 22, 2017 1:10 am

Daniel Shawul wrote:Nebiyu alien can play 4-player chess and other multi-player games. I spent quite a lot of time on this a few years ago studying algorithms, developing a GUI and CLI game managers, game servers etc which was fun. There is also an android app : https://sites.google.com/site/dshawul/home which I am not sure works now due to java issues but you can try it. A triple-chess in action using javaboard.

Image


Daniel
What algorithm did you end up using ? I would probably use MCTS and be satifsfied that it would probably beat an average player :P

Daniel Shawul
Posts: 3758
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: Another possible toy for AI programmers: 4-player chess

Post by Daniel Shawul » Fri Sep 22, 2017 1:33 am

I use both MCTS and a variant of min-max for multi-player games. Don't know which one did better but it can play reasonably with both given the inherent issue with collusion. The algorithm called "paranoid" that assumes every one else is an opponent did the worst, but easy to code since you can use standard 2-player algorithms for that.
My memory is not very good on this since it has been few years since i did this.
Daniel

Tony P.
Posts: 81
Joined: Sun Jan 22, 2017 7:30 pm
Location: Russia

Re: Another possible toy for AI programmers: 4-player chess

Post by Tony P. » Fri Sep 22, 2017 3:38 am

Daniel Shawul wrote:Don't know which one did better but it can play reasonably with both given the inherent issue with collusion.
The only really dangerous form of collusion is a pre-agreed coalition of 3 players against 1, which chess.com will likely prohibit explicitly and also try to prevent by rule changes and pseudorandom opponent selection.

If 2 players, A and B, form a coalition vs C, it's better (in terms of expected score) for D to ally with C and attack A or B (whichever of them is more exposed) than to join the attack on C, basically have a 1/3 probability of being the one who checkmates C and gets 20 points, and, after C is defeated, face the joint attack by A and B with no hope for C 'opening a second front' vs A or B.

That's why it's important that there are 4 players, not 3 - this allows the game to spontaneously turn into a battle of 2 equally strong alliances even if 2 players are colluding.

User avatar
hgm
Posts: 23725
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Another possible toy for AI programmers: 4-player chess

Post by hgm » Fri Sep 22, 2017 6:19 am

The problem with these kind of games is always: "why do anything at all?". If A and B team up against C, the D can just lea back and wait until they exhausted themselves in destroying D. In 4-player Japanese Chess (Yonin Shogi) this problem doesn't exist, as you actually gain strength there by being successful in battle, because you get the captured pieces in hand.

Tony P.
Posts: 81
Joined: Sun Jan 22, 2017 7:30 pm
Location: Russia

Re: Another possible toy for AI programmers: 4-player chess

Post by Tony P. » Fri Sep 22, 2017 3:32 pm

This version has substantial bonuses for piece captures, so 'leaning back' is still less lucrative than attacking - in the process of 'exhausting each other', armies will earn points for captures and double / triple checks that might turn out vital in the final tally.

For future reference, here are the current rules (they can be changed during the beta test and, later on, be different at different chess sites):

Image

The order of turns is Red, Blue, Yellow, Green.
Pieces move and capture like always (though there is no en passant at this time)

Points are usually earned by capturing live (colored) pieces:

Pawns: +1
Knights: +3
Bishops and Rooks: +5
Queens: +9
Kings are +20

Checkmate eliminates a player and earns +20 points

Your pawns promote to Queens on your 8th rank (at the center of the board)

Promoted Queens are worth only 1 point

Dead (grey) pieces in the armies of eliminated players do not provide points

Delivering a double-check is worth +5, and triple-checks are worth +15

When a player resigns or times out, leaving two or more active players on the board, their King remains live and is still worth +20 to whoever can capture it; but if there are only two players remaining, the remaining player will be given +20 points

The player with the most points at the end, wins!

Stalemate will result in +20 for whoever caused it (including the stalemated player)

The current time control is 1 minute with a 15 second delay. That means each player has 15 seconds to complete their move and after 15 seconds their total game time will start ticking down. Players are eliminated when their time runs out!

Post Reply