Will computers ever solve chess?

There is no short cut in chess (well, only possibly one, and that is not enough).

In checkers, the game was solved in a sense that allows optimal play (in the precise sense of game theory, i.e. recursively best against best opposition), by first of all generating a complete tablebase of endgames with 47 trillion positions, then deriving a strategy to reach such endgames to achieve an optimal result. This was only possible because moves that were highly likely not to be better were ignored (all opponent moves were still analysed, or the result would not be entirely reliable).

https://www.newscientist.com/article/dn12296-checkers-solved-after-years-of-number-crunching/

Seems like hardly anybody bothered to read what I said.

I said that I am using the engine on this website from Menu | Learn | Analysis

This is stockfish.JS (JavaScript) not stockfish.cpp (C++)

Using a tablebase the MAX number of moves required is 34 for B+N but these moves make no sense to a human.

A human will take more moves, say 43, to mate from that position because he is using the B+N technique to mate.

If you have a good computer with multiple cores and a good engine that uses bit boards (which JavaScript can not do) then it is possible for the computer to calculate the win providing that it can search to a deep depth, in this case 34 moves x 2 = 68 plys.

Computers however will always struggle with the horizon effect, if they can not see past the game horizon (max search depth) then they can not determine the result.

In this example add 2 white connected pawns on the 2nd rank and one black knight.

In a real game between humans blacks only chance is to set up a blockade or to sacrifice the knight for the 2 pawns and hope that white does not know the B+N technique.

For a computer how many plys (half moves) need to be played to avoid a blockade so that black has to sacrifice the knight for the 2 pawns, in other words how long before DTC (Depth To Conversion)

In this case to avoid problems the engine needs to see past DTC + Max B+N Plys

Computers always have a problem with the horizon effect and this is why they use tablebases, so they can in effect cheat and peek over the horizon.

How long is a piece of string?

 You should be reported for your username and the comments I have seen you post.

I read what you wrote. You made several caveats, pointed out you were using a weakened engine, then made the claim that engines have a very long way to go.

There's a very old computer programming term for this: GIGO. (Is it still in use?)

OK

Holding my breath until you solve chess.

Already turning blue, but I know you can do it.

The two are very alike, as I have myself pointed out here.

Trial division is a logical solution, but not reasible for a computer as slow as the presents ones. 

Same thing with space, which is more of a killer.

Not enough atoms in the universe to do it.

Elroch, 

this is the the sort of solution I'm talking about. Of course chess is far more complex than checkers, but the same principle of following best moves until they reach a database will work for chess without having to store every single possible position.

May be in the future computers can solve thus new problems.

By the time computers can play perfect chess everything will be able to be automated by computers.

I agree with you. My only comment is what is the basis to believe chess is a draw? I can't find any analytical or mathematical reason for that conclusion.

As for solving chess, there is some reason to believe it can happen within a decade or two. Here's a brief analysis:

Checkers has a game tree complexity of 10^31, and was solved in 2007. It was a draw, so every game had to be studied (make sure there was no way for white or black to win).

Chess is more complicated. The game tree complexity is 10^123. So chess is more complicated by a factor of about 10^92. (doesn't sound good so far).

Now, some factors that might simplify the calculation for chess:

1) If we can find a forced mate, not all positions need to be investigated.
2) If there is a forced mate, it would make sense to look for it by studying only well-played games (ignore games where one side purposely makes bad moves, like losing pieces for no reason).
3) Chess opening theory has already been highly studied.
4) All chess endgames with 7 pieces and less have already been solved.

Now to calculate how much each of these simplifies the calculation (note: one ply = 1/2 a move, such as a move by white)

Instead of a 10^123 game tree complexity, by checking only games with normal play (halting any calculation where one side makes a stupid move) the game tree complexity is reduced to 10^40. (see Wikipedia "Shannon number/Sensible games"). Now our problem is only 10^9 times (1,000,000,000) as complicated as checkers.

Now use existing chess theory to ignore all bad chess openings. Let's say we use opening theory for the first 3 plies. (3 plies x 20 choices = 20^3 = about 10,000). 1,000,000,000/10,000 = 100,000. Our problem is now about 100,000 times as complicated as checkers.

Now use an existing chess tablebase to play the final 6 plies of the game (6 plies x 5 move choices = 5^6 = about 10,000). 100,000/10,000 = 10. Our problem is now about 10 times as complicated as checkers.

Checkers took 18 years to solve using 200 computers at the peak of the calculation. Our calculation is 10x as difficult, so it will take 180 years to solve chess. How to do better: Use 10 times as many computers. So chess might be able to be solved in 18 years.

Now we're still at a disadvantage compared to checkers, because calculations may proceed more slowly. Each move needs to be checked if it was "stupid" or not (like losing a queen with no compensation). If stupid, stop the work and go to the next game.

On the other hand, we have one big advantage, If we find a forced mate, we're done. The rest of the game tree can be ignored.

So, with distributed computing, using about 2000 computers, and the assumptions above, we have a problem that can conceivably be solved in 15-20 years.

True

My question is why would you want to completely solve chess when it could take all the fun out of the game.  Chess is really best the way it is

I'm dying to find out if one side can force a win. But even if chess is solved, I won't be able to memorize all the correct moves. Playing OTB would still be fun. (But playing on-line I would worry about my opponent cheating using the perfect play database). But if chess is solved and I still want to play on-line by correspondence, there's plenty of other chess-type games that still won't be solved.

You are double and triple counting eliminations (I assume that you can see that your bullet points above are not mutually exclusive, but you are recounting positions that would have already been eliminated by previous and subsequent bullet points), and vastly overestimating.

You are also comparing evaluation of chess positions to checkers positions as apples to apples.  Not the case at all....just like Go, Checkers is simpler and much faster to evaluate a position for.  The reason Go is harder than Checkers was is because the tree is much larger than Chess.  Checkers' tree is both far smaller and much less processing power per evaluated position.

Even granting all your other overall generous estimates as-is otherwise (a very tall order), you are way off. 

Never say never. We have already established the premise that chess can be solved, yet it is just a matter of time. Even if it takes 1 million years, that is but a wink of the eye, when comparing it to the age of the universe. The one thing this thread has introduced, which I never considered before, was the idea that chess can be solved with a win. This would indeed kill the game for sure. Based on how many ways draws can be made, I doubt there will be a winning solve of chess though.

There's no way chess can be solved completely because there simply isn't enough space to store all the positions. Let's say you build a system that stores one position per atom, unfortunately the number of chess positions far exceeds the number of atoms in the entire universe (observable), as a result you find yourself running out of physical space to store them.

So, there's no way you can solve it completely, but one thing you can do is filter out positions that give one side a huge advantage because we already know who will win. So you start from the initial position and branch out, every time you get to a position that is +/- 3, you chuck that line. This way, the number of possible positions reduces drastically and there's some hope.