Will computers ever solve chess?

2.g4?

Nah. A human+engine might lead to a scenario like in Terminator Rise of the machines.

This made my day, I'm still laughing. Why is this thread still going.

You've also violated the 75 post rule: 75 posts without a new piece of evidence, topic of discussion, interpretation of evidence, or other novel contribution are an automatic draw by referee.

IT IS IMPOSSIBLE.

There are more possible chess games than electrons in the universe. 

Computers are made out of atoms. 

Computers will not solve chess.

The number of chess games is limited by the three move repetition rule and the 50 move rule. 

Unless quantum computing does something really funky, but I mean... there's A LOT of chess games. 

Definitely not in your lifetime, since technology tends to double in speed every 2 years (quite a consistent pattern during our tech revolution era), not nearly fast enough to compete with the sheer complexity of chess.

Still not gonna beat Chuck Norris:

http://www.funnyjunk.com/funny_pictures/2951233/Chuck#d4643a_2950789

But you don't have to look at every game to solve chess.

This is the type of wild speculation that keeps this thread popular.

Which is why solving it by trying out every possible game is probably not the way to do it. As many have said, not every possible game is necessary. Who knows, the answer might come in a dozen years from an average 13 year old player from Sri Lanka who is playing a supercomputer and stumbles on THE game that wins.

I apologize for my ignorance folks : ) 

Y'all make a good point ! Don't need to play every possible game of tic tac toe to master that game, after all!

s23bog, you are describing a vague sort of positional evaluation. Some sort of positional evaluation is needed, but deep, powerful search is far more important. The strongest chess programs use positional evaluation which is not entirely human designed, and this has been taken to a higher level in AlphaGo. Hard-coding of positional evaluation with input by strong players has increasingly limited value as programs surpass human players.

That being said, the notion of including space and activity in an evaluation is absolutely sound, and is one of the things that distinguishes a good player (I also see it as something on which I am a bit weak).

Humans wouldn't know it if they saw it.  A winning line could already have been played, but how would you prove it was winning?  By traversing all the games in the line from the beginning...like you have to do for all lines in order to find one that works this way.  There.  Are.  No.  Shortcuts.

such a mistake could be and would probably be in the opening-not middle game or endgame.

if you find the mistake in the opening--no need to look at middlegame or endgame.

Computers continue to solve chess.

Your "point" is frivolous. Illogical. Searching for mistakes from the end leads nowhere. There is no ground to proceed, nowhere to search for an answer. Irrational comes to mind. A mistake may be found, but possibly another mistake was made prior. Where does that leave us?

The nearest to a shortcut is that inferior moves tend to lead to branches that require less computation. In many positions, it is likely that a small fraction of the possible opponent moves (eg 1-3) would lead to the large majority of the computation needed for all 40 of them.

The reason is combinatorial. Suppose there are N moves at each stage. If a bad move makes the result 1 move nearer, that branch is a factor of N smaller than the toughest defense. Based on my experience of endgame tablebases, bad moves mostly bring a result many moves nearer, so unless N is very small, such a branch is of almost insignificant size (N^M smaller, where M is the number of moves quicker).

This is roughly speaking how computers achieve search to huge depth: side branches tend to mostly shrivel up as they become very highly probably inferior, so it possible to see, say, 20 moves ahead, while simple combinatorics would suggest 20 moves would be unmanageable.

But, even allowing for this, it is reasonable to think the effective branching factor (allowing for most branches being small or effectively small) is 2 or 3 for opponents moves in a typical position. Again, we can get some clue about this from endgame tablebases, although these have a tiny number of pieces, which leaves fewer legal candidates from the outset.

==============================================================

On an entirely separate topic, I wonder how strong a program would be running on a modern supercomputer. These are about a million times faster than a high end desktop computer.

If doubling speed gained 40 Elo points, such a computer would be about 800 points stronger.

I think it is safe to say no-one is ever going to justify the budget (never mind porting the code) to test this, but perhaps there is a potential 4000 Elo machine out there!

btickler wrote:

Humans wouldn't know it if they saw it.  A winning line could already have been played, but how would you prove it was winning?  By traversing all the games in the line from the beginning...like you have to do for all lines in order to find one that works this way.  There.  Are.  No.  Shortcuts.

This is disproved in several ways. But. it comes down to a "belief". Much like a religeon. One believes the only "proof" lies in testing all possibilities, or that human intelligence allows for shortcuts.  

@Elroch .. speed of calculation does not increase it's rating, except that which involves the chess clock. In shorter time controls, it's rating will increase vs humans. Give the best players days to decide on the best move and it would make no difference. The program will not find a better move because it can process quicker, once it has used it's time.

Superior hardware only means the silicon monster can find it's best move faster, not a better move.

I do not think this is correct. Sometimes a chess engine changes its mind.

Sometimes the best chess engines do not find the best move.

The Lomonosov supercomputers in Moscow were used for several months to develop the 8-piece endgame tablebases. But I wonder if a major Chinese or U.S. university would ever allocate their supercomputers for a chess study though.

On the other hand, solving chess would be the type of thing that would draw worldwide attention, and bring international acclaim to whoever does it first.