Is computer chess "solved"?

[Uri Blass]

Chess is certainly not solved. But computers now play at a level way beyond normal human capabilities (and significantly beyond Grandmaster level). In this sense strength increase are of little to no significance for humans. But they are of significant to other programs.

Steve

I think that the best corr. players will still beat any program, any computer and will do so for a long time to come.

I think that you are wrong and inspite of the fact that the best correspondence players use chess programs to help them I do not believe that they are more than 50 elo better than the best computer(software+hardware) and the 50 elo advantage that they maybe have is easily closed by using more time(or having hardware that is 4 times faster when I say 4 and not 2 because I believe in diminishing returns).

[Terry McCracken]

Chess is certainly not solved. But computers now play at a level way beyond normal human capabilities (and significantly beyond Grandmaster level). In this sense strength increase are of little to no significance for humans. But they are of significant to other programs.

Steve

I think that the best corr. players will still beat any program, any computer and will do so for a long time to come.

I think that you are wrong and inspite of the fact that the best correspondence players use chess programs to help them I do not believe that they are more than 50 elo better than the best computer(software+hardware) and the 50 elo advantage that they maybe have is easily closed by using more time(or having hardware that is 4 times faster when I say 4 and not 2 because I believe in diminishing returns).

It's easy to make such claims based on arbitrary numbers and opinion but without empirical data your dismissal of my opinion falls flat.

Can a machine plan? Can it reflect or think about the game or positions that arise objectively? Can is understand or learn where it went wrong 50 moves later? Can it play intenionally from the opening with or without books how the endgame will be played out? Can it anticipate the most likely positions that will arise during the early middlegame and see what type of basic ending will be played out with correct play?

Until it can it is below man.

[Dr.Wael Deeb]

Chess is certainly not solved. But computers now play at a level way beyond normal human capabilities (and significantly beyond Grandmaster level). In this sense strength increase are of little to no significance for humans. But they are of significant to other programs.

Steve

I think that the best corr. players will still beat any program, any computer and will do so for a long time to come.

I think that you are wrong and inspite of the fact that the best correspondence players use chess programs to help them I do not believe that they are more than 50 elo better than the best computer(software+hardware) and the 50 elo advantage that they maybe have is easily closed by using more time(or having hardware that is 4 times faster when I say 4 and not 2 because I believe in diminishing returns).

It's easy to make such claims based on arbitrary numbers and opinion but without empirical data your dismissal of my opinion falls flat.

Can a machine plan? Can it reflect or think about the game or positions that arise objectively? Can is understand or learn where it went wrong 50 moves later? Can it play intenionally from the opening with or without books how the endgame will be played out? Can it anticipate the most likely positions that will arise during the early middlegame and see what type of basic ending will be played out with correct play?

Until it can it is below man.

I totally agree with you Terry,it can't,regarding these pure human abilities....but it can do other things I wish I can do

[Terry McCracken]

Chess is certainly not solved. But computers now play at a level way beyond normal human capabilities (and significantly beyond Grandmaster level). In this sense strength increase are of little to no significance for humans. But they are of significant to other programs.

Steve

I think that the best corr. players will still beat any program, any computer and will do so for a long time to come.

I think that you are wrong and inspite of the fact that the best correspondence players use chess programs to help them I do not believe that they are more than 50 elo better than the best computer(software+hardware) and the 50 elo advantage that they maybe have is easily closed by using more time(or having hardware that is 4 times faster when I say 4 and not 2 because I believe in diminishing returns).

It's easy to make such claims based on arbitrary numbers and opinion but without empirical data your dismissal of my opinion falls flat.

Can a machine plan? Can it reflect or think about the game or positions that arise objectively? Can is understand or learn where it went wrong 50 moves later? Can it play intenionally from the opening with or without books how the endgame will be played out? Can it anticipate the most likely positions that will arise during the early middlegame and see what type of basic ending will be played out with correct play?

Until it can it is below man.

I totally agree with you Terry,it can't,regarding these pure human abilities....but it can do other things I wish I can do

Yes, me as well, they are fantasic calculators and don't miss a tactical combination within their horizon, it never tires or becomes distracted and this ensures it will usually win the day over the board in a match at 40 in 2 or faster time controls. Corr. is our last stand. Soon, we'll have to be cyborgs to have a chance.

[Don]

It seems that Anthony Cozzie has solved the chess game and haven't told us how.
Clearly, almost all chess engines around are using essentially the same search techniques and algorithms: ID + PVS + Null Move + LMR and so on... Some do this differently, with some new tweaks, more knowledge, less bugs, etc. But, essentially, it is all the same.

The current approach we are using to make computers play chess is very efficient and successful, although it appears that after the advent of the null move (Late move reduction, maybe?), we are, after all, only fine tuning and tweaking to improve a chess engine.

Do you think computer chess is mostly "solved" (as stated by Anthony Cozzie) and why?

My opition is that we are not nearly close to solve it. We are all struggling to take more from the classic approaches, and the fact we are making it in a very slow pace, doesn't mean computer chess is "solved".

Another thing I guess I didn't get straight: Is Cozzie's assertion "the computer chess is mostly solved", the same as "the game of chess is solved"? If yes, I presume he thinks that the current classic approach is the best one can get...

Regards,

No, it's not even close to being solved. If someone can take Ivanhoe, work on it for a few months and achieve a 50 ELO rating improvement it's not even close to being solved.

I think we can say it's close to being solved when you can play a 1000 game match between the top 2 programs and only a dozen or so are not draws.

Don

[Don]

.....Do you think computer chess is mostly "solved" (as stated by Anthony Cozzie) and why?

My opition is that we are not nearly close to solve it. We are all struggling to take more from the classic approaches, and the fact we are making it in a very slow pace, doesn't mean computer chess is "solved".......

How can it be solved when programmers keep coming up with new innovations to raise the bar higher than it was previously?

When endgame tablebases meet in the middle with opening theory then it will be solved.

When is that?

Not is our lifetimes....nor the next nor the next nor the next nor the next nor the next nor the next nor the next nor the next nor the next......ad infinitum......

If they do solve it in 4765, it will not matter to me. I will have enjoyed the insolubility of it all.

This is not tic tac toe..... it's much more than that.



Chris

PS It must be profound night.....

I think the questions was more a practical one that a theoretical one because he asked about when it was "close" to being solved.

I would prefer to ask it a different way however - how far away are the top computers from perfect play? Then we can at least attempt to assign a numeric value to it.

Because what we are going to see happen and already have seen is that computers draw more and more games - an indication that they are playing better and better. Until they play perfectly, they will lose a game on occasion.

There will come some point when it becomes like checkers is now. Checkers programs are still not quite perfect, but close. Although chinook has "solved" the game, programs in practical time controls are still not quite there. But checkers can be pretty boring because in order to determine if one PC checkers program is better than another, you have to play a lot of games, the vast majority of which are draws. You might play 100 games and the score is 3 to 2 for example with 95 draws. (This is just an example, I don't know the exact percentage of draws modern checkers programs attain.)

If a program draws 98% of it's games against an equal opponent, it may still be a significant distance away from perfect play. It could be that if you draw 98% against an equal and great playing opponent you would still only draw 60% against a perfect player. I'm guessing of course.

[bhlangonijr]

I think the questions was more a practical one that a theoretical one because he asked about when it was "close" to being solved.

I would prefer to ask it a different way however - how far away are the top computers from perfect play? Then we can at least attempt to assign a numeric value to it.

Well, I meant to ask that too.
what is your take on that? What would you use as reference?

As we know Elo means nothing in this regard. Draw/Loss percentage? I am not sure I'd go in that direction, because the draw percentage only shows that both top engines playing are very close to each other in terms of strength. The perfect play would require an oracle provided with a kind of 32-man endgame table-bases. How to know if the engine is playing a perfect game?
Even if you, hypothetically, match up top perfect-play chess engines, it would be possible that a lot of start positions lead to a win for white or black, so that the outcome of the games wouldn't show by itself if the engines are playing perfectly, without knowing before-hand what should be all the best played moves in all games. Confusing...

Because what we are going to see happen and already have seen is that computers draw more and more games - an indication that they are playing better and better. Until they play perfectly, they will lose a game on occasion.

There will come some point when it becomes like checkers is now. Checkers programs are still not quite perfect, but close. Although chinook has "solved" the game, programs in practical time controls are still not quite there. But checkers can be pretty boring because in order to determine if one PC checkers program is better than another, you have to play a lot of games, the vast majority of which are draws. You might play 100 games and the score is 3 to 2 for example with 95 draws. (This is just an example, I don't know the exact percentage of draws modern checkers programs attain.)

If a program draws 98% of it's games against an equal opponent, it may still be a significant distance away from perfect play. It could be that if you draw 98% against an equal and great playing opponent you would still only draw 60% against a perfect player. I'm guessing of course.

[Albert Silver]

I think the questions was more a practical one that a theoretical one because he asked about when it was "close" to being solved.

I would prefer to ask it a different way however - how far away are the top computers from perfect play? Then we can at least attempt to assign a numeric value to it.

Well, I meant to ask that too.
what is your take on that? What would you use as reference?

As we know Elo means nothing in this regard. Draw/Loss percentage? I am not sure I'd go in that direction, because the draw percentage only shows that both top engines playing are very close to each other in terms of strength. The perfect play would require an oracle provided with a kind of 32-man endgame table-bases. How to know if the engine is playing a perfect game?
Even if you, hypothetically, match up top perfect-play chess engines, it would be possible that a lot of start positions lead to a win for white or black, so that the outcome of the games wouldn't show by itself if the engines are playing perfectly, without knowing before-hand what should be all the best played moves in all games. Confusing...

Because what we are going to see happen and already have seen is that computers draw more and more games - an indication that they are playing better and better. Until they play perfectly, they will lose a game on occasion.

There will come some point when it becomes like checkers is now. Checkers programs are still not quite perfect, but close. Although chinook has "solved" the game, programs in practical time controls are still not quite there. But checkers can be pretty boring because in order to determine if one PC checkers program is better than another, you have to play a lot of games, the vast majority of which are draws. You might play 100 games and the score is 3 to 2 for example with 95 draws. (This is just an example, I don't know the exact percentage of draws modern checkers programs attain.)

If a program draws 98% of it's games against an equal opponent, it may still be a significant distance away from perfect play. It could be that if you draw 98% against an equal and great playing opponent you would still only draw 60% against a perfect player. I'm guessing of course.

My personal take is that ultimately, chess played perfectly from the get-go is a draw. That said, eventually, there will be a line, much as exists today, where the players (computers) will have to determine how far they are willing to go to increase their winning chances.

This may seem like an odd comment as I just declared I think it is a draw, but imagine this:

in the position, the best five moves lead to a draw, however, 2 of the moves create a greater number of chances down the road for the opponent to lose, whereas the other three lead to fewer chances. Then there is one move which may lead to dangers of losing (which I will not fall for as I am perfect) but far more chances to lose for the opponent, and then there is the sixth move which is objectively losing, but leads to the greatest number of winning chances. This last move might be desirable if the opponent is very unlikely (with X%) to find the correct defense, much less the winning defensive sequence. A calculated gamble.

[Don]

I think the questions was more a practical one that a theoretical one because he asked about when it was "close" to being solved.

I would prefer to ask it a different way however - how far away are the top computers from perfect play? Then we can at least attempt to assign a numeric value to it.

Well, I meant to ask that too.
what is your take on that? What would you use as reference?

It's still about your odds of winning and I think ELO is still relevant. I could be wrong about this, but I think we can get a fairly good measure of how close we are to perfect play by how often the top programs draw. And I think it's almost certain that the game is a draw because as programs get stronger (and at the top human levels) we see more and more draws. We can take a cue from checkers, where the top programs draw almost every game against each other, and we know they are close to solving the game because Chinook DID solve the game, and the other programs are not far behind.

If two of the top chess programs of the future play a 200 game match and every game is a draw, I think we can confidently speculate that we are within a few ELO of perfect play.

I would predict that an omniscient player would beat the best current modern programs almost every game. We are probably 500-1000 ELO from perfect play. You are probably going to hear that computers have practically solved chess, but nothing could be farther from the truth. We just think that because we are so weak at chess and are impressed with their playing ability, not realizing the very best humans suck pretty badly.

As we know Elo means nothing in this regard. Draw/Loss percentage? I am not sure I'd go in that direction, because the draw percentage only shows that both top engines playing are very close to each other in terms of strength. The perfect play would require an oracle provided with a kind of 32-man endgame table-bases. How to know if the engine is playing a perfect game?
Even if you, hypothetically, match up top perfect-play chess engines, it would be possible that a lot of start positions lead to a win for white or black, so that the outcome of the games wouldn't show by itself if the engines are playing perfectly, without knowing before-hand what should be all the best played moves in all games. Confusing...

Because what we are going to see happen and already have seen is that computers draw more and more games - an indication that they are playing better and better. Until they play perfectly, they will lose a game on occasion.

There will come some point when it becomes like checkers is now. Checkers programs are still not quite perfect, but close. Although chinook has "solved" the game, programs in practical time controls are still not quite there. But checkers can be pretty boring because in order to determine if one PC checkers program is better than another, you have to play a lot of games, the vast majority of which are draws. You might play 100 games and the score is 3 to 2 for example with 95 draws. (This is just an example, I don't know the exact percentage of draws modern checkers programs attain.)

If a program draws 98% of it's games against an equal opponent, it may still be a significant distance away from perfect play. It could be that if you draw 98% against an equal and great playing opponent you would still only draw 60% against a perfect player. I'm guessing of course.

[frcha]

in the position, the best five moves lead to a draw, however, 2 of the moves create a greater number of chances down the road for the opponent to lose, whereas the other three lead to fewer chances. Then there is one move which may lead to dangers of losing (which I will not fall for as I am perfect) but far more chances to lose for the opponent, and then there is the sixth move which is objectively losing, but leads to the greatest number of winning chances. This last move might be desirable if the opponent is very unlikely (with X%) to find the correct defense, much less the winning defensive sequence. A calculated gamble.

I play like this all the time ...and it happens a lot in online blitz ..There are many "odd opening lines" that are really bad at least according to computers, but humans have problems with them.


Even if chess is solved -- chess960 even if solved would be exciting for humans to play. -- and i suspect other variants would be too.