If chess is ever "solved" by computers, we will reintroduce the use of dice, as chess was played in the middle ages.
The future of computer Chess
Sort:
What would an unsolveable chess look like? Infinite number of squares? Infinite number of pieces? Random variables within our already known game?
Gonnosuke wrote:
Ricardo_Morro wrote:
The question for chess, a much more complex logical system, is this proposition: "Given any position, it is possible to know whether the outcome will be a win or draw." If this proposition is true, chess is solvable. If it is not true, chess is not solvable. But what if this proposition is undecidable?
The answer to that question is definitely "yes". We already know *how* to answer the question -- with tablebases -- we just don't have the technology (or will) to build them. The process for generating tablebases is well known and scientifically proven so the solution is theoretically within reach but it will be many, many years before tablebases will be a viable solution for solving chess.
It's a matter of scale.
Thank you. As was stated earlier in this thread, a result of solving chess would be a 32 piece table-base. We already have them for up to 6 pieces, so the problem is demonstrably decidable.
I'm under the impression that chess is indeed solveable, but that there will be a huge amount of "perfect" moves or paths to the outcome. I do not believe there is going to be a single perfect move for every single position.
Sjolden wrote:
I'm under the impression that chess is indeed solveable, but that there will be a huge amount of "perfect" moves or paths to the outcome. I do not believe there is going to be a single perfect move for every single position.
I think this is particularly likely if the result is that the game is drawn, which I suspect is the case.
Let's say this is going to be the case. Even when chess is solved, it does not mean that computers are going to win 100% of games against humans. Because if perfect play from both sides is a draw, and there are many paths to this outcome, humans need not worry to much. They will at least be able to play for draw, even if a win is impossible against these computers. And besides, even after chess is solved, its very nature is so complex that it won't change anything for Human vs Human games.
Agreed, although it would be a bit of a shame to find out it's a guaranteed win for White it still wouldn't change anything with respect to the way we play it.
The way computers currently play chess is to examine the branching lines for so many ply deep, then evaluate the positions according to some algorithm. But for chess to be solved, all the branching lines have to be pursued to the end, without any evaluating (Evaluations being inherently imperfect). The table of your tablebases must have a depth, however large. But the only way of knowing how large this will be is by actually doing it. It seems to me that in these conditions, the only way to know whether or not chess is solvable is by actually solving it.
May 27, 2009
0
#110
I think chess can be solved, but perhaps not in the next 50 years and perhaps not in the lifetime of anyone here.
The thing about chess though is the human touch....being able to know your opponent. This is hard for computers to do. It will always be hard for them. They aren't human, and that's most of the battle in 'solving' chess.
In terms of computers, they can get better and programs will go deeper in lines of chess thinking. 99.99% of all humans will always be soundly trounced by computers.
May 27, 2009
0
#111
I think chess can be solved, but perhaps not in the next 50 years and perhaps not in the lifetime of anyone here.
The thing about chess though is the human touch....being able to know your opponent. This is hard for computers to do. It will always be hard for them. They aren't human, and that's most of the battle in 'solving' chess.
In terms of computers, they can get better and programs will go deeper in lines of chess thinking. 99.99% of all humans will always be soundly trounced by computers.
Here's something to consider: you can have two identical positions in chess with two different "best moves." In the one case an "en passant" capture is possible, in the other it is not. It all depends on the move order leading up to the position. So position by itself is not always determinitive of best move. Nor is it enough to know the immediately preceding position. Take two positions, identical, one where castling is possible and one where it is not, because the king has moved and then moved back to its original square: we must know not only the position, but also relevant history, to evaluate it.
So I can feed your computers positions where the best move is unknowable, because their history is unknown. I think a fair application of the touch-move rule would be that if the computer even considers making that illegal king move, he will be deemed to have touched the king--
Haha, very interesting to think about. I'm going to wait and see what anyone else has to say about that. Food for thought.
Those are distinct positions -- en passant, castling rights, whose move it is and the set of previous positions (for repetition detection) all need to be taken into account when comparing positions.
(Number of moves since the last capture or pawn move can also be considered, but is optional if we're looking for forced draws as opposed to claimable ones since it's a subset of repetition, detection of which would be the appropriate measuring stick for the purpose of identifying drawn positions)
Is it impossible then to "work backwards" in solving chess? How would we know if castling and such was applicable? Until we rewound all the way back to move 4 or so?
May 27, 2009
0
#117
I believe solved means win or draw a match with a chess computer or a person. I believe the chess computer will be control by levels of difficulty and still be useful to people. My computer has 136 levels of difficulty. So far I've never won a game at higher than the fourth level. I play a lot of games at the first level and enjoy them.
An endgame tablebase has up to six pieces solved chess for every legal position. Fortunately the 50 move rule would allow the opponent to claim a draw, though of course no mere mortal could find a mate in 262.
EDIT: Interestingly you can claim a draw here at chess.com online chess for the 50 move rule even though I believe using an endgame telebase is permitted.
I think that chess will be solved by computers, but it won't ruin the game for humans because we don't know the solution. The only way the game will be ruined is if someone memorizes the solution, but it's impossible for a human to memorize all the variations possible.
May 27, 2009
0
#120
Chess is a finite game and therefore is solvable. Go is also a finite game and therefore is also solvable.
Written above is that it is "just" a matter of getting a 32 piece tablebase. This is not the case. Any potential solution to chess would not involve storing the correct move for every possible position, since there is not enough space in the universe to store this information. So that leaves us with some combination of what is storable (in the current tablebases) along with what is searchable.
My opinion, for what it's worth, is that it is unlikely to be solved in the pure sense of the word. Any potential solution would have to involve assumptions (for example, if you're up a "free" rook then the game is as good as won), but such notions cannot be part of a formal solution.
And what does a random generator do to the concept of "best play"? It seems that in some sense to solve the game is to destroy it, for once solved it is no longer a game.
To return for a moment to the example of tic-tac-toe, which by analogy was used to shoot down my application of Godel's theorem. Even though it is easy to list all possible games of tic-tac-toe (it only being a maximum of 9 moves long), there are still things within its system that cannot be known. For example, given a concluding position, is it possible to know what order the x's and o's were put in the boxes? No, the best we could do would be to say one of several possible ways. So if we made the statement, "the order or moves was such-and-such," we could not know whether it was true or false, and that would be an undecidable proposition.
So Godel's theorem applies even to tic-tac-toe. The question for chess, a much more complex logical system, is this proposition: "Given any position, it is possible to know whether the outcome will be a win or draw." If this proposition is true, chess is solvable. If it is not true, chess is not solvable. But what if this proposition is undecidable?
Given any position, or given EVERY position? Given any position, we know that some positions are possible to force mate. I could give you a puzzle now that has a forced mate in it if you like. So, in that sense, chess is solvable. I know I am just trying to be difficult, but word your statements carefully.