Will computers ever solve chess?

Obviously assessments don’t change as often as they once were, simply because the power of computation has increased. The machine simply ‘sees’ much more than a GM in the past would have. That doesn’t mean anything, as the present computers’powers of computation will be a laughing stock for future machines, and present assessments may change drastically.

  ‘Solving chess’ only would offer the definitive answer. Until then we are left with guesses, which despite looking stronger than those of the past, are still a far cry from the final verdict, if you look at the power of computation of present computers which, although impressive, are still able to look only at a small fraction of all the variants. And that's regarding computers! Imagine how small the humans’window of seeing the big picture is...

 I thank you for that interesting post.

I neglected earlier to respond to the last part of your post. Yes, white wins a higher percentage of games than black at the grandmaster level. That's understandable because moving first gives white a slight advantage. If one side has a small advantage at the beginning, it makes sense that over time that side will win more games. Small advantages may accumulate to become decisive. An inaccurate move by white may allow black to equalize - and the small advantage white started with is gone. But an inaccurate move by black will increase the advantage white began the game with. In other words, because white begins with a slight advantage, he can get away with a bit more.

But we're talking about a perfectly played game. And you made my argument for me above. As you said, "After all, nothing can beat perfect play." Nothing (including perfect play by white) can beat perfect play (by black). Even perfect play by white cannot convert his slight advantage into a decisive advantage - against perfect play by black. In order for a perfectly played game to result in a win for white, the game must be a theoretical win for white before the first move is played.

Over time it appears that white's winning percentage (including draws) at the highest level varies between 52% and 55%. That range seems about right to me if having the first move gives white a small advantage. But it seems way too low if white's first move advantage is decisive - if his game is theoretically won at the outset.

Think about this. Suppose white's first move advantage is decisive. How do we explain white's dismal results? White begins with a theoretically won game and manages a winning percentage of just 52% to 55%. And how do we explain black's extraordinary results? Black begins with a theoretically lost game and yet manages to attain a winning percentage of 45% to 48%. We're talking about the same players! If white's first move advantage is decisive, white should have a winning percentage of 100% with perfect play on both sides. Black should have a winning percentage of 0%. Human players are not capable of perfect play of course - with the white or black pieces.  But how do you explain why the same players, when playing black, play so much better than when playing white? You've got to explain that if you're going to argue that white's first move advantage is decisive.

I agree that White's initial first-move advantage is insufficient to win, given perfect play on both sides. There is such a thing as "having the better of the draw", and that is likely the case here.

It could be that Black has a forced win.  It is known that it is not always good to be the player on the move in a symmetrical position.  Perhaps the initial position is in a subtle equilibrium, and being forced to move is to be forced to lose, if only Black knew the proper responses at each and every move.  In other words, it can be speculated that the initial position of the game of chess is one grand Zugzwang that mere mortals have yet to comprehend.  In that case, the very best first move for White under the circumstances could be something like 1 a3, disturbing the equilibrium the least and  inflicting the least damage on one's self.

Of course, I don't really believe this for a moment.  Based on the development of opening theory over time, and the experience of many generations of masters, I believe that it is very likely that the game is a draw with best play on both sides.  (It is only an impression; obviously I can't prove it.)

There is such a thing as "evidence". The evidence is overwhelming that chess is a draw with optimum play for both sides.

 That’s what I meant when I suggested earlier a win for either White or Black: The intitial position  being one of  Zugzwang—who moves first loses. 

  It could go either way, and all we have right now are guesses.

In the 19th century, white's advantage was between 52% and 53%. By the 1930s it was up to 54%. In most modern databases it seems to be around 55% (chess.com's data gives about 55.5%).

In the 1200 games between AlphaZero and Stockfish using specified openings each of which had 100,000 recorded examples in a database (i.e. the most popular mainlines), white scored 58.7% (standard error 0.69% ) In the 100 game match between the same two engines with complete freedom in the opening, white scored 61% (standard error 2.4%).

With the trend observed as chess knowledge has advanced, plus the hint of further increases in white advantage as play becomes even stronger, it would be unwise to be too sure that the theoretical result of chess is a draw. Think of it this way, if white ever got to 75% in empirical results, the two hypotheses would have equal standing even without extrapolating further. With the results above around 60% we are already quite a long distance from 50% to that point, and heading strongly in one direction.

There are two factors at play in the empirical results. One is the more comfortable task that white has which leads to a computational advantage and an effective rating boost, and the other is the noise introduced by the errors of the two players. The fact that doubling computing power has less and less effect at higher levels might suggest that white would have less advantage, but it is clear that the reduction in noise from errors is the stronger effect and dominates this.

The empirical advantage of the first move is increasing, and we cannot be sure how far this will go: there is certainly no sign of the trend stopping yet.

OMG, it's an intelligent, factual post...in this thread.  Kudos .

You said mostly what I was already thinking. If white goes first, and plays perfectly, even perfect play by black is not good enough. If two runners, of equal ability start a race, one starts first by a tiny amount, then the two running "perfectly" at the same pace will always result in the person going first winning. But like I said, I dont even know what perfect play is since I have never seen it. It's all guesses and speculation. If both sides ever did play perfectly this conversation wouldn't even exist, chess would be solved.

What I didn't know was the exact percentage that white wins. I figured it would be about between 51 to 53%. Some are saying it's as high as 55% which surprises me. That says to me that whites advantage is bigger than I thought origninally.Given that human chess players are terrible at chess, that's an impressive number. And even among computers supposedly white wins more often. And since computers are in their infancy it makes sense to me that over time white will win more often as computer, and people, get better. A thousand years from now maybe chess wont be solved. But I'll guess the computers a thousand years from now will have white winning more often than black, by a bigger percentage than even 55%.

Your race analogy is flawed, though.  Black can by rule hold a draw with white "ahead" up to a full piece.  Black can also draw with greater material deficits, and can force perpetual check or stalemate.  If you could tie Usain Bolt in the 100yd dash by finishing a half second behind him, then your analogy would sort of work.

Yes. The key factor is that chess CAN be a draw. In a game where draws don't exist we can be sure that one player has a winning strategy and if we can somehow convince ourselves that white cannot be at a disadvantage, white must be able to win. There are games where you can prove this without brute force. One large class of these is where the first player can mimic the second player after he has played his first move and where the rules are such that the first move cannot be a disadvantage.

Noughts and crosses (tic-tac-toe) provides a refutation of the argument that if the first player has an advantage, he must have a win.

Until that happens, you will see 2 more million pages of guesses.

You didn't answer my question. If white begins the game with a theoretical win, that means black begins with a theoretical loss. However, white ends up with a winning percentage of just 52% to 55%. If you're right that white's first move advantage is decisive, white's winning percentage should be 100% if both sides play perfectly. So white's imperfect play results in a winning percentage far lower than it should be. But why does black's imperfect play result in a winning percentage far higher than it should be? The same players play white and black! If white's first move advantage is decisive, obviously imperfect play by him could blow that advantage - but isn't white's imperfect play neutralized by black's imperfect play? If white starts out with a decisive advantage (a theoretically won game) and white and black play equally imperfectly, white should have a very high winning percentage - far higher than 52% to 55%.

I think this is one of the most popular threads on these forums. The possibility of computers solving chess has not been ruled out.

It's one of those great mathematical questions, where a surprise can happen. It reminds me when some mathematicians said in 1968 that all shapes which tile the plane have been found and that "no more are possible".

But then a few years ago, a new shape was found, creating huge chatter in the math world. This isn't it, but it's one of the cool ones:

You're right. It's a poor analogy. A game of chess is nothing like a race. The runner ahead at the end of the race wins. One side in a chess game may have a material advantage at the end of the game - but not a winning advantage and the game is drawn. An obvious example is king and bishop vs. king. One side is up an entire piece but cannot win.  

Computers will solve chess because the computing power will and continue to increase in the future. It's just a matter of time. 

I agree. It won't be Alpha Zero though. Those folks have bigger fish to fry than solving chess. I'm wondering if (or rather when) they'll use that technology to try to solve mysteries like black holes, dark matter and dark energy.

I guess I didn't answer your question the way I should have.  If white begins with the chances of winning at 52 to 55% isn't that a theoretical win? If a casino had chances of winning at 55% and 45% of losing wouldn't that be considered a pretty huge advantage? I think the reason white wins more is because people are terrible chess players. That and also because chess isn't solved, or played perfectly. If it were played perfectly, then white would win 100% of the time. But even computers dont play perfectly so probably more computer games are won by white I would guess. Maybe it just comes down to our opinions on what an advantage is. I think 55% is a huge advantage, you think it should be higher. We both agree white has an advantage, so what do you think whites winning percentage should be even with our current terrible level of play? 60%? More? Do you or anyone else know what the percentage of games won is by the world chess champion? I would think if it's around 55% that should be enough to win most tournaments. That to me is a pretty big advantage and I think over time as computers get better that number will rise until it eventually gets to 100%.