How long do you think chess (8x8 board) will be solved by engines?

[抽刀断水水更流]

Despite the fact that human have been exploring the world of chess for many centuries,now many superhuman engines are pushing the progress of solving chess much faster than ever before.
Many engines are superhuman,how long does it take to conquer the ultimate goal-----solve 8x8 chess.

I would like to hear your thoughts

[Guenther]

You first have to define what you would call 'solved', otherwise all of this is meaningless.

[抽刀断水水更流]

I mean solved
Solved means 8x8 chess has the same pieces of material,so
to solve chess,it must be perfect play,and perfect play means every single game from engines will be a draw,whether you have opening book or not,at the end it will be a draw.
That what I define solved chess
All Games are draws

[Guenther]

I mean solved
Solved means 8x8 chess has the same pieces of material,so
to solve chess,it must be perfect play,and perfect play means every single game from engines will be a draw,whether you have opening book or not,at the end it will be a draw.
That what I define solved chess
All Games are draws

Perfect play is much more. It even knows about what positions produce more traps / possible wrong answers / complications for each and every
ply for the whole search space, to be exploited against non perfect players, which still would be far above current entities.

Two 'perfect players' game example (w/o further heuristics)

Code: Select all

1. a3 a6 2. Nf3 Nf6 3. Ng1 Ng8 4. Nf3 Nf6 5. Ng1 Ng8 1/2-1/2

What you define might not even be called 'weakly' solved though by some, just 'ultra-weakly'.
forum3/viewtopic.php?f=7&t=77509&hilit=weakly+solved

[towforce]

My view:

Win rate 0.001 (1 game in a thousand): 15 years

Win rate undetectable: 30 years

Chess fully resolved: 60 years

[Lazy_Frank]

I think this is right time to post one of TCEC quotes here:

Terminology differs per author.

[Uri Blass]

I mean solved
Solved means 8x8 chess has the same pieces of material,so
to solve chess,it must be perfect play,and perfect play means every single game from engines will be a draw,whether you have opening book or not,at the end it will be a draw.
That what I define solved chess
All Games are draws

Perfect play is much more. It even knows about what positions produce more traps / possible wrong answers / complications for each and every
ply for the whole search space, to be exploited against non perfect players, which still would be far above current entities.

Two 'perfect players' game example (w/o further heuristics)

Code: Select all

1. a3 a6 2. Nf3 Nf6 3. Ng1 Ng8 4. Nf3 Nf6 5. Ng1 Ng8 1/2-1/2

What you define might not even be called 'weakly' solved though by some, just 'ultra-weakly'.
forum3/viewtopic.php?f=7&t=77509&hilit=weakly+solved

Note that his definition is that chess engines are unbeatable.

Playing perfect games with no mistakes(something similiar to your game) is not enough if there is a way to beat the engine by some opening book.

Of course it is possible that we do not know that chess engines are theoretically unbeatable but only that they are practically unbeatable that does not prove that they are theoretically unbeatable.

[lkaufman]

If you only require that the engine (with a book of openings it approved with deep analysis or even no book) just never lose, then I think we are already there. Probably Stockfish or Dragon could play 10000 games vs any opponents at correspondence time controls on 64 cores and never lose now. Not practical to prove I know. But solving chess should at least mean that the engine will never miss a forced win if the opponent allows one, and for this we may need more than fifty years.

[JVMerlino]

For me, "solving" chess would require proof that White did not have a forced win from the starting position.

[Uri Blass]

If you only require that the engine (with a book of openings it approved with deep analysis or even no book) just never lose, then I think we are already there. Probably Stockfish or Dragon could play 10000 games vs any opponents at correspondence time controls on 64 cores and never lose now. Not practical to prove I know. But solving chess should at least mean that the engine will never miss a forced win if the opponent allows one, and for this we may need more than fifty years.

Maybe and Maybe not.

I think that 64 cores and correspondence time control is too long to get 10000 games.

If somebody suggest million dollars for the first human who beat stockfish or dragon in the next 3 years at 1,000,000,000 nodes per move with deterministic setting(for example single core and 4096 mbytes hash and using the 5 piece tablebases) and nobody is going to win the prize then it is going to be a convincing evidence.

Unfortunately I do not know about a tool to try to beat a deterministic opponent in chess so we can test things.

If you ask me how to try to beat stockfish at 1,000,000,000 nodes per move then my idea is to start a game with 10,000,000,000 nodes per move against 1,000,000,000 nodes per move.

I guess it is going to be a draw so you have a candidate game that you try to improve.
Try to change a single move in the game and play stockfish with 10,000,000,000 nodes against stockfish with 1,000,000,000 nodes per move.

If the maximal evaluation does not go up try to change a different move until you get one of the following:
1)You tried every move and failed to improve the maximal evaluation(in this case go to a different opponent)
2)You got a game with a bigger maximal evaluation that is still a draw.
In this case try to improve this game by the same process.
3)You got a win against stockfish.

I do not know if by this process 10,000,000,000 nodes per move is going to win against 1,000,000,000 nodes per move and of course I am not going to try it manually.

We need some automatic tool for this idea and first try it with smaller number of nodes to see how much time you need to produce a win and maybe to get faster ways to get a win(you may try to play with contempt or maybe have a different idea not to try to change every move but to try to change only in positions that are close to the maximal evaluation function).

Edit:Note that I would like to have this tool also for handicap game.

You play stockfish against stockfish when Stockfish starts without knight b1.
I believe Stockfish is going to get a loss with white.
In this case improvement is losing with more moves or a draw or a win and the question is how much improvement you can get against fixed number of nodes and it may be interesting to see it against small number of nodes.