Computer v Computer Bridge

[EricK]

There are a lot of chess playing programmes out there, and there are also people who spend much of their time playing one programme against another to work out which is the best and by how much etc. (see eg http://www.husvankempen.de/nunn/)

 

But apart from the annual world championship there seems to be almost no bridge programme v. bridge programme matches. Why is that?

[P_Marlowe]

No money involved.

 

The first chess programm were developed out of fun,

but in the past, you could earn some money, selling a

chess program, that has changed, the market for bridge

programms is a lot smaller.

 

With kind regards

Marlowe

[EricK]

No money involved.

 

The first chess programm were developed out of fun,

but in the past, you could earn some money, selling a

chess program, that has changed, the market for bridge

programms is a lot smaller.

 

With kind regards

Marlowe

But the people who are running the chess engine v chess engine tournaments aren't doing it for money. They are strcitly hobbyists.

 

Why aren't there any people running Jack v GIB v Bridge Baron v WBridge etc tournaments? Surely bridge freaks are just as bad as chess freaks!

[cherdano]

This has little to do with money, as there is no more money in go programs either, but there is an active go server only for computer go programs: http://cgos.boardspace.net/9x9/index.html.

 

I think it's the opposite, there are a lot of hobbyists trying to build computer go and chess programs, but hardly any for bridge programs.

[EricK]

This has little to do with money, as there is no more money in go programs either, but there is an active go server only for computer go programs: http://cgos.boardspace.net/9x9/index.html.

 

I think it's the opposite, there are a lot of hobbyists trying to build computer go and chess programs, but hardly any for bridge programs.

But this still doesn't explain why nobody appears to be running computer v computer bridge tournaments with the programmes which are already written. is it that nobody is really interested in the results?

[Guest Jlall]

Maybe its because the computers suck in bridge currently.

[EricK]

Maybe its because the computers suck in bridge currently.

I think the chess computer v computer matches took place even when chess computers sucked too. And wouldn't this sort of competition encourage improvements in the software?

[hrothgar]

Maybe its because the computers suck in bridge currently.

I think the chess computer v computer matches took place even when chess computers sucked too. And wouldn't this sort of competition encourage improvements in the software?

Here's some idle speculation

 

Standardizing an interface that permits a pair of computers to compete at Chess or Go is extremely simple. The only information that needs to be exchanged is encapsulated in the move. I expect that designing a bridge interface to handle declarer play or defense at bridge would also be quite easy.

 

Bidding is another story. The bidding interface needs to describe both the level / denomination of a bid and the definition of that bid. This is a really ugly problem, especially if different programs don't share any kind of common vocabulary. From my perspective, the most promising approach would combine a rules based approach with a statistically significant sample of hands that are consistent with the bidding. However, anything like this would require a significant amount of work.

 

As I recall, Matt Ginsburg stopped entering GIB into the Computer Championships because no one could agree on bidding system regulations or disclosure methods.

[cherdano]

Maybe its because the computers suck in bridge currently.

They are A LOT better than the go programs.

[Aberlour10]

Another question. Due to complexity of the bridge game, is there a real chance in the future to develope software which would be able to compete as a "team" vs world champs squad in a long match?

 

Robert

[cherdano]

Another question. Due to complexity of the bridge game, is there a real chance in the future to develope software which would be able to compete as a "team" vs world champs squad in a long match?

 

Robert

Yes.

[hrothgar]

Another question. Due to complexity of the bridge game, is there a real chance in the future to develope software which would be able to compete as a "team" vs world champs squad in a long match?

 

Robert

Back in 1998, GIB competed in a par contest at the World Championships in Lille. The program was competing against the best players in the World. GIB placed 12th out of 34. Even if the various programmers haven't made any improvement to their code base over the last decade, faster hardware would still be expected to significantly improve the performance of the programs. If we look at declarer play and defense, I suspect that the computers may already have an edge.

 

As I mentioned earlier, bidding is another question. Programming a computer to bid well strikes me as a more difficult problem. You'll also run into the same disclosure issues that I mentioned in an earlier posting. (I suspect that disclosure between humans and computers is likely to be much more complicated than disclosure between a pair of computer. For example, the sampling methods that I suggested wouldn't work nearly as well)

[Guest Jlall]

Another question. Due to complexity of the bridge game, is there a real chance in the future to develope software which would be able to compete as a "team" vs world champs squad in a long match?

 

Robert

Back in 1998, GIB competed in a par contest at the World Championships in Lille. The program was competing against the best players in the World. GIB placed 12th out of 34. Even if the various programmers haven't made any improvement to their code base over the last decade, faster hardware would still be expected to significantly improve the performance of the programs. If we look at declarer play and defense, I suspect that the computers may already have an edge.

 

As I mentioned earlier, bidding is another question. Programming a computer to bid well strikes me as a more difficult problem. You'll also run into the same disclosure issues that I mentioned in an earlier posting. (I suspect that disclosure between humans and computers is likely to be much more complicated than disclosure between a pair of computer. For example, the sampling methods that I suggested wouldn't work nearly as well)

It was leading for a while too until it got a couple of 0's because it didn't take a key inference from the bidding which was supposed to guide the play.