Online Chess & Bridge Analyse softwares

[H_KARLUK]

I also play online chess.

 

Above lines calculated and when required by human players analyse deep by some intelligently designed software. At least cuts "irrelevant gossipy chat". That software only focus and compare which playing line is superior to the other.

 

Silent and nice.

 

Depth Score Time Nodes Principal Variation

32. ... Rg4 33. Nc3 Bxg2+ 34. Kxg2 Qg7 35. Nb5 c5 36. bxc5 dxc5

32. ... Qh5 33. Qe2 Qh7 34. Nc3 Bxg2+ 35. Kxg2 Rh8 36. Kg1

32. ... Qh3 33. Qf3 c6 34. Qe2 Qh7 35. Nc3 Bxg2+ 36. Kxg2 Rh8 37. Rh1 Qh3+ 38. Kg1 Qe6 39. a3 d5 40. Kg2

 

When player right clicks able to execute move, add to move a comment, open variation in new board.

 

I remember Fisher, Karpov, Kasparov (many famous great WGMs) played vs very strong softwares.

 

I see commercial and development. I know at this age one day some software team will produce similar for "online bridgegame". That chess analyser works instantly online.

 

Plus I liked "boss" button.

 

Simple and sweet.

[Rossoneri]

Hi Shubi.

[TylerE]

Chess is a much, much, much easier game to solve than Bridge, because the entire state of the game is known to both the players and neutral observers. You can basically throw CPU at the problem and it solves itself. That doesn't work for bridge.

[Lobowolf]

Chess is a much, much, much easier game to solve than Bridge, because the entire state of the game is known to both the players and neutral observers. You can basically throw CPU at the problem and it solves itself. That doesn't work for bridge.

While I agree with the gist of this in principle, and while chess is obviously a finite system and a game of complete information, it's still nowhere near being solved. I think all 7-piece or possibly 8-piece endings have been solved, and each additional piece is a substantial order of magnitude; extrapolate to the fact that chess is, in essence, a 32-piece ending from the outset, and it's clear that it won't be "solved" any time soon.

[TylerE]

Well, I guess I was a bit loose with the word solved. What I meant in context is that is possible, without 'cheating' to determine the exact state of match.

[Lobowolf]

Well, I guess I was a bit loose with the word solved. What I meant in context is that is possible, without 'cheating' to determine the exact state of match.

I just thought you were loose with the term "easy(ier)". B)

[jikl]

At the risk of putting myself up for abuse or ridicule I will link in an article I wrote a while back on just this subject. B)

 

ABF Newsletter article about Computers and Bridge

 

The article is on page 8.

 

Sean

[H_KARLUK]

It's like Robot rewards tours. Difference Robot rewards does not analyse variants. In that chess website you play vs human, you save deal as *.pbn. then that software analyses variants deeply.

 

I wonder WGMs chance vs very intelligent software on "individual" format basis. This is done in past on Chess. Why not same in Bridge?

 

I strongly offer bridgesoftware programmers to visit such Chess Websites.

[mike777]

Chess is a much, much, much easier game to solve than Bridge, because the entire state of the game is known to both the players and neutral observers. You can basically throw CPU at the problem and it solves itself. That doesn't work for bridge.

While I agree with the gist of this in principle, and while chess is obviously a finite system and a game of complete information, it's still nowhere near being solved. I think all 7-piece or possibly 8-piece endings have been solved, and each additional piece is a substantial order of magnitude; extrapolate to the fact that chess is, in essence, a 32-piece ending from the outset, and it's clear that it won't be "solved" any time soon.

How many possible moves are there in chess....if computers are operating at 100 teraflops per second or "if soon" a billion teraflops I would guess we are getting close..soon assuming standard chess time limits.

 

I would think as we approach or exceed a billion teraflops per second even many bridge players may loose to a computer.

[H_KARLUK]

There may be Ayes and Nays on this platform.

 

I can only speak for myself :

 

It would be nice to use such a very strong software to analyse each deal after finished with " superior and impartial " dbase.

 

No matter friendly or not, at least it's silent. What I am sure "it sells very good". Bcoz tests th real power !

 

Just a challenge for brainstorms :rolleyes: It's legal.

[awm]

If you're looking for double-dummy analysis (i.e. what is the best card play) then GIB or Deep Finesse will do this for you.

 

If you're looking for the "right play at the table" this is much more difficult. It depends on what information is available at the time. For example, I once played a hand where the same contract was declared at both tables from the same side, on the same lead. No defensive or declarer play mistakes were made at either table. Yet one table made the contract and one didn't! How is this possible? Because different inferences were available from the (different) auctions at the two tables. Computers are pretty far from working this type of thing out.

[H_KARLUK]

I agree with your words except those :

 

 

awm Posted on Nov 11 2008, 07:41 AM

<..right play at the table...Computers are pretty far from working this type of thing out.>

 

Does that mean computer softwares are unable to "compare one-to-one and one-to-many linked datas" ?

 

Sorry, I disagree. I see everyday with my eyes. They calculate possibilities, shortly variants, aberrations etc etc. They point me why one line is superior to th other/s with diagrams :rolleyes:

 

I just right click and execute offered variant. Then I see it in a new window. Simple. Then I jump to another variant. Easy to see winning line. Really.

 

Chess players respect statistics science like many serious bridge players.

 

P.S. I am also a soccer fan. I saw software processing action movie format with arrows. It analyses the game within preset widely respected classic and new found but mostly safe manouvres by authorities.

[xcurt]

Chess is a much, much, much easier game to solve than Bridge, because the entire state of the game is known to both the players and neutral observers. You can basically throw CPU at the problem and it solves itself. That doesn't work for bridge.

While I agree with the gist of this in principle, and while chess is obviously a finite system and a game of complete information, it's still nowhere near being solved. I think all 7-piece or possibly 8-piece endings have been solved, and each additional piece is a substantial order of magnitude; extrapolate to the fact that chess is, in essence, a 32-piece ending from the outset, and it's clear that it won't be "solved" any time soon.

How many possible moves are there in chess....if computers are operating at 100 teraflops per second or "if soon" a billion teraflops I would guess we are getting close..soon assuming standard chess time limits.

 

I would think as we approach or exceed a billion teraflops per second even many bridge players may loose to a computer.

Nope. The estimated game-tree complexity of chess exceeds the number of atoms in the universe.

 

On the other hand I'm much less pessimistic than some about the possibilities for computer bridge. Partly I think that the bidding databases are still very weak (which also impacts play of the hand because the inferences are poor). Partly I think that even the very best players make a lot of sub-optimal plays, so there is more opportunity for the computer to pick up points.

[cherdano]

If you're looking for double-dummy analysis (i.e. what is the best card play) then GIB or Deep Finesse will do this for you.

 

If you're looking for the "right play at the table" this is much more difficult. It depends on what information is available at the time. For example, I once played a hand where the same contract was declared at both tables from the same side, on the same lead. No defensive or declarer play mistakes were made at either table. Yet one table made the contract and one didn't! How is this possible? Because different inferences were available from the (different) auctions at the two tables. Computers are pretty far from working this type of thing out.

I think this statement is a little unfair. The problem is just to teach a computer how his opponents bid. And the problem with that is that his human opponents are typically unable to present the information about their agreements in a way the computer can understand.

 

It's a bit like saying that I am unable to work out inferences from the bidding when I am playing in Hungary when I have little clue what system or style my opponents are playing and none of them speaks English.

 

In restricted settings (a long match against a single pair where the programmer is allowed to enter the opponents agreements in advance, and maybe ask questions during the auction and pass on the information to the program) this is more easily done. I understand that Jack did pretty well against top Dutch players in such a setting.

 

I am confident that if the same effort that has been put into computer chess had been put into computer bridge, then we would have recognized world class programs by now.

[H_KARLUK]

Wow

 

Tyvm "xcurt". I liked ""It is not enough to be a good player. You must also play well." -- Tarrasch"

 

In loving memory to a huge brain :

http://en.wikiquote.org/wiki/Siegbert_Tarrasch

 

I think some of his ideas works for bridge game also.

[awm]

There are a lot of deep inferences available in high-level bridge that computers are just not equipped to make as things stand now. For instance:

 

(1) Often there are negative inferences based on the lead. However, in order to evaluate these, we need to know how "appealing" different leads will look to the player making the lead. We need to be able to reason something like: "He lead from xxx, does he have king in another suit?" and determine that either "with the king in the other suit, he would likely have preferred to lead that rather than xxx" or perhaps vice versa "if he had the king in the other suit, he would never underlead it." How do we know what lead a person likes? Perhaps we can determine what we would lead given the hand and the same information from the auction, but this person may not lead like we lead etc.

 

(2) Similarly we can make inferences about people's hands based on their tempo or choice of line of play or defense. For example, say declarer is in a spade contract and dummy has a singleton heart. But declarer just starts drawing trump right away. One might infer that there is not really a problem in hearts (i.e. declarer has no heart losers). But this only makes sense if you assume declarer's line of play is reasonable. Randomly generating hands for declarer (as for example GIB does) will ignore this type of data (i.e. declarer cannot have that hand, because even though it is consistent with the bidding, he would not be taking this line of play...)

 

(3) GIB is also very bad at figuring out when (and when not) to falsecard or signal. The fact is that often the best "double dummy" defensive play (i.e. least likely to cost a trick assuming optimum play and defense) is not actually the best play, because an alternative action would make declarer less likely to find the double dummy play (or make partner more likely to defend optimally). An easy example is that leading from small cards tends to look better than it is in practice, because anything you "give" to declarer is something he could've done anyway playing double dummy (i.e. if you pick up partner's Jxxx, so what, declarer could've done that) and if you haven't blown the needed tempo partner will automatically find the proper double dummy switch when he gets in.

 

(4) Finally, to go back to bidding, it's not just what bids the person made. It's what bids they didn't make. Obviously if people play a "really unusual system" then human experts can have trouble with these inferences too. But most of the time we can work things out logically even if the actual decision made by opponents was based on feel/experience and not formal rules. Computers really have no capability to do this.

 

Games of "hidden information" are fundamentally more difficult than games like chess. While it's not obvious that computers will always be bad at bridge, we are very very far from the kind of expertise that chess computers already display. Faster processors will not be enough to make up this difference, nor will a few years of concentrated research at IBM or Microsoft (if that were to occur).

[cherdano]

All true for existing computer programs (or at least for GIB), but I don't really think these are fundamental problems that computers are by principle unable to solve.

 

If you had a large dataset of played hands, together with information

- about agreements on the auction,

- about the players,

I don't think it would be all that hard to develop good models predicting what an opponents would lead from a particular holding. But given the messy interface between human agreements and computer programs you will never have such a database.

 

I don't think bridge is fundamentally more difficult to program than go.

[mike777]

There are a lot of deep inferences available in high-level bridge that computers are just not equipped to make as things stand now. For instance:

 

(1) Often there are negative inferences based on the lead. However, in order to evaluate these, we need to know how "appealing" different leads will look to the player making the lead. We need to be able to reason something like: "He lead from xxx, does he have king in another suit?" and determine that either "with the king in the other suit, he would likely have preferred to lead that rather than xxx" or perhaps vice versa "if he had the king in the other suit, he would never underlead it." How do we know what lead a person likes? Perhaps we can determine what we would lead given the hand and the same information from the auction, but this person may not lead like we lead etc.

 

(2) Similarly we can make inferences about people's hands based on their tempo or choice of line of play or defense. For example, say declarer is in a spade contract and dummy has a singleton heart. But declarer just starts drawing trump right away. One might infer that there is not really a problem in hearts (i.e. declarer has no heart losers). But this only makes sense if you assume declarer's line of play is reasonable. Randomly generating hands for declarer (as for example GIB does) will ignore this type of data (i.e. declarer cannot have that hand, because even though it is consistent with the bidding, he would not be taking this line of play...)

 

(3) GIB is also very bad at figuring out when (and when not) to falsecard or signal. The fact is that often the best "double dummy" defensive play (i.e. least likely to cost a trick assuming optimum play and defense) is not actually the best play, because an alternative action would make declarer less likely to find the double dummy play (or make partner more likely to defend optimally). An easy example is that leading from small cards tends to look better than it is in practice, because anything you "give" to declarer is something he could've done anyway playing double dummy (i.e. if you pick up partner's Jxxx, so what, declarer could've done that) and if you haven't blown the needed tempo partner will automatically find the proper double dummy switch when he gets in.

 

(4) Finally, to go back to bidding, it's not just what bids the person made. It's what bids they didn't make. Obviously if people play a "really unusual system" then human experts can have trouble with these inferences too. But most of the time we can work things out logically even if the actual decision made by opponents was based on feel/experience and not formal rules. Computers really have no capability to do this.

 

Games of "hidden information" are fundamentally more difficult than games like chess. While it's not obvious that computers will always be bad at bridge, we are very very far from the kind of expertise that chess computers already display. Faster processors will not be enough to make up this difference, nor will a few years of concentrated research at IBM or Microsoft (if that were to occur).

at one point will computer infer ok.........trillion billion teraflops or many more?

I do not mean perfect ...i just mean ok

[hotShot]

Given the right input and enough computer power GIB and other programs are able to find and make the move most likely to succeed, based on the statistical data they get from simulating similar deals.

 

If the input data are insufficient, the set of possible distributions GIB gets from applying the known facts as restrictions, is much bigger than it could be, using all available inferences. So the set often includes deals that are not similar to the deal being played, and statistical data taken from this set are misleading GIB.

Typical examples are GIB going wild during the auction and leads.

After the dummy is visible, GIBs knowledge of the deal gets better, leading to fewer glitches.

[y66]

If Nate Silver played bridge, this problem would probably be solved by now.

[H_KARLUK]

On 3rd May 1997 , New York (If I m not wrong)Deep Blue Supercomputer played a fascinating match versus reigning World Chess Champion, Garry Kasparov. It was a very tuff one.

A dramatic victory occurred in 6th game; software won.

 

The system derived its playing strength mainly out of brute force computing power. It was a massively parallel, RS/6000 SP Thin P2SC-based system with 30-nodes, for a total of 30 120 MHz P2SC microprocessors (one microprocessor per node), enhanced with 480 special purpose VLSI chess chips. Its chess playing program was written in C and ran under the AIX operating system. It was capable of evaluating 200 million positions per second, twice as fast as the 1996 version. In June 1997, Deep Blue was the 259th most powerful supercomputer, capable of calculating 11.38 gigaflops.

 

Kasparov vs. Deep Blue (The Rematch) was one of the most popular live events ever staged on the Internet.

 

The web site received more than 74 million hits representing more than 4 million user visits from 106 countries during the 9-day event.

 

Anyway Soviet superstar later on beated silicon :( Then "deepblue" name changed to "deepthought". I forgot date.

 

I wish same nice history happens for Bridge Fans.

 

Regards.

Hamdi

[Lobowolf]

Wow

 

Tyvm "xcurt". I liked ""It is not enough to be a good player. You must also play well." -- Tarrasch"

 

In loving memory to a huge brain :

http://en.wikiquote.org/wiki/Siegbert_Tarrasch

 

I think some of his ideas works for bridge game also.

Tarrasch was ahead of his time, and one of the better players not to become world champion.

 

 

How many atoms are in the universe, anyway?

[awm]

If Nate Silver played bridge, this problem would probably be solved by now.

Nate Silver is a smart guy for sure.

 

But so are Matt Ginsberg and Chip Martel, both of whom are computer scientists who do play bridge. Neither of them has solved the problem (and Ginsberg put in a whole lot of time).

 

The point is that chess is in principle solvable if you throw enough computing power at it. All the information is always there, only a finite number of possible game positions exist. Obviously we are nowhere near having the computing power to "solve" chess completely (it may be impossible to have such computing power given the size/lifetime of the universe) but in principle it could be done. So the trick is just to find "shortcuts" to speed up the computation without losing too much accuracy. This is the kind of thing that computer scientists know how to do.

 

Go is also in principle solvable. It does have a (much) larger set of possible game positions. And not as much time has been put into finding the proper "shortcuts" to speed up the computation without losing accuracy. But a serious, well-funded effort to build a computer program that can beat the best human Go players would probably succeed after a few years.

 

Games of incomplete information are a totally different matter. This includes most card games (bridge, spades, poker, etc). Not all the information is there to simply compute the right action because we cannot see the opponents cards (yes, we could in principle enumerate all possible sets of hands for the opponents, but this is not the same as seeing the opponents cards). So there is a fundamental problem here -- how do we compute the right play without the necessary information? We can use statistical methods, but these depend on getting our probabilities right. We need to know conditionals like "what is the probability that LHO would've made that lead if he had this hand." Note that it is not enough to know the conditional "what is the probability that I would've made that lead if I had this hand" because I am not LHO. There are also situations where it is necessary to play randomly, or to make a play that simplifies the position for partner, even though on a double dummy basis neither of these actions is relevant.

[cherdano]

Go is also in principle solvable. It does have a (much) larger set of possible game positions. And not as much time has been put into finding the proper "shortcuts" to speed up the computation without losing accuracy. But a serious, well-funded effort to build a computer program that can beat the best human Go players would probably succeed after a few years.

This is absolutely wrong.

 

[Of course this is just my opinion but I can assure you that everyone else who has ever been involved in computer go would also disagree with your statement. Also, while I am not sure what you mean by serious and well-funded, the market for go programs in Asia is big enough that there are several commercial go programs; I know of at least one with 3-4 full time programmers, which might be more than the top chess programs.]

[awm]

Go is also in principle solvable. It does have a (much) larger set of possible game positions. And not as much time has been put into finding the proper "shortcuts" to speed up the computation without losing accuracy. But a serious, well-funded effort to build a computer program that can beat the best human Go players would probably succeed after a few years.

This is absolutely wrong.

 

[Of course this is just my opinion but I can assure you that everyone else who has ever been involved in computer go would also disagree with your statement. Also, while I am not sure what you mean by serious and well-funded, the market for go programs in Asia is big enough that there are several commercial go programs; I know of at least one with 3-4 full time programmers, which might be more than the top chess programs.]

It was IBM that built the computer which beat Kasparov.

 

IBM is a huge multinational company which threw very large amounts of money at building a machine to play chess. In fact they designed custom hardware -- basically constructed a super-computer optimized just to play chess.

 

This is orders of magnitude more resources than any of the commercial Go programs (with 4-5 full time programmers) are applying to the problem.

 

Why did IBM do this? Mostly for public relations/advertising. They don't expect to recoup their investment by selling chess programs.

 

And if you think I'm wrong about there being a finite number of positions on a Go board, well, I'd like to hear your reasoning for that one!