Is Chess Something We Can Solve?

The starting position is either a win, draw or loss. All computer evaluations except mate and exactly 0.00 are an admission that the engine is not certain.

An oddity of Stockfish is that even its "mate in N" evaluations are sometimes refuted by deeper analysis (they are displayed without being rigorously checked). And while 0.00 is usually a position that is heading for a draw, it can also be a position which is completely unclear, but where the engine finds the chances to be balanced. Neural networks like Leela offer a better output as a probability for each of win, draw and loss. This distinguishes unclear equality from a dead draw.

Elroch - in the Stockfish evaluations in the over 50,000 tactics problems on the website - I've never seen '0.00' except when the engine is assigning a dead draw evaluation. That's in looking at variants to the solution moves too.
In other words it always assigns some kind of numerical advantage to one side - often something less than 0.5 but not zero.

Dead equal does not mean dead drawn for practical play. Hence the value of the trinomial probability output of neural networks. It's a simple fact that a few percent of positions with evaluations between say 0.2 and -0.2 have an evaluation of 0.00 regardless of whether the position is dead drawn or balanced. The only way for this to be avoided would be if the engine somehow "corrected" 0.00 evaluations if the position was unclear but with no side having any advantage (according to the engine calculation).

Tactics are a special case. They have to have an objective of achieving a clear result. This is a win or a clear draw, not an unclear, double-edged position!

Real chess does not.

If by assigning a dead draw evaluation you mean a stalemate, triple repetition, ply count 100 or over or insufficient material then try this (which is none of those).

If that's not what you mean, what do you mean? It will assign 0.00 in positions that are theoretical wins, I've seen that many times.

This assigns 0.00 to the top move for example if you click the magnifying glass (depending on how long the analysis runs). Does that mean the engine thinks it's a dead draw?

reminder to all viewers that by this logic a 50 error stalemate with martin "justifies" all black moves

Just gobsmacking...

One class of positions where Stockfish can initially mistake a win for dead drawn is when there is a brilliancy down the line to break open a blocked position.

If by dead drawn, you mean dead positions it has no algorithm to detect that. Nobody knows one. If not what's the difference between drawn and dead drawn?

chill

By dead draw I mean there's no way for either player to win except if somebody makes a mistake. And the 'dead' refers to it being obvious.
The Stockfish 0.00's I saw so very many of - we're obvious draws when giving them any attention. Also notable was very few pieces/pawns on the board.
They happen in two instances - when the solution is a win but the player blunders allowing the dead draw.
And - when the idea was to draw and that was the solution - and non-solutions lost.
Included some stalemates. Both ways.
A few perpetuals. Both ways.

Martin if you're talking about some of the spectacular blundered findings of Stockfish ... yes you're good at posting those.
I don't know if you get them from the tactics puzzle though.
I'm referring to every instance I've seen of 0.00 in the tactics puzzles.
They weren't 'unclear' or 'equal'.
They were draws. Dead draws.
Maybe 'dead draw' isn't the best terminology.
Two lone Kings. Definitely Dead. Maybe that should be the 'usage'.
In over the board tournaments - there used to be a rule that may still be there that if in the opinion of the tournament director a master could swindle a C player in the position concerned - then the director isn't supposed to grant an adjudicated draw when only one player claims a draw.

@160

"ICCF is not a team trying to solve chess" ++ No, the 17 ICCF WC candidates try to become the World Champion, but as a by-product are weakly solving chess, just like Sveshnikov predicted.

"human players pretend they are having an impact" ++ The humans are the decisive factor.
That is how they got through preliminaries, semifinals, candidates to the finals.

"selecting engine moves" ++ No.

"write an algorithm and some kind of wrapper application that would do essentially same thing...run half a dozen different engines and select a consensus engine move using a set of criteria and weightings. The difference is, such an app would be more consistent and score better" ++ If you think so, then you should do so, and then enter an ICCF preliminary tournament, and get kicked.

even with his carefully selected opposition tygxc cant help but fumble.

At this point, chess is not solvable. But it's also worth mentioning that as time goes by, human participation in solving chess becomes less and less important. Games where computers and grandmasters combined become flawed because grandmasters introduce a lot more mistakes than computers do.

I don't know where the exact point will come, and maybe it's already happened. But eventually the VERY best grandmasters contributions will only drag down a computers performance. Claims about computers and grandmasters combined showing some sort of result on solving chess are laughably inadequate.

Solving chess is a very complex problem. One that the human brain cannot solve. If it's ever going to be solved, it's going to require very powerful computers.

tyxgc, please leave the contributions of grandmasters out of the equation. Not only do they limit the performance of solving chess, but the limitation also grows by the day.

@183

"What is ICCF"
++ International Correspondence Chess Federation.
They play correspondence chess at 50 days per 10 moves, engines and data bases allowed.
The ongoing World Championship Final now has 112 draws out of 112 games.

WC33/final, World Championship 33 Final

This ultra-weakly solves Chess: it is a draw,
as at least 1 of the 112 games is a perfect game with optimal play from both sides.

This is also part of a weak solution of chess,
as the 112 draws redundantly show how black can draw against reasonable white tries to win.

@184

"the VERY best grandmasters contributions will only drag down a computers performance."
++ This is a common misconception among those ignorant of correspondence chess.
The humans take care of the opening with the help of data bases and by analysing with the help of engines all games ever played by their opponents.
The humans take care of long term strategy beyond the calculation horizon of their engines, even at 5 days average per move.

"If it's ever going to be solved, it's going to require very powerful computers."
++ Each of the 17 ICCF WC finalists uses 2 servers of each 90 million positions per second.
In the ongoing tournament they considered 10^17 positions and they spent 6120 CPU years, 61.2 times more than the 50 CPU * 2 years Schaeffer spent to weakly solve Checkers.

reminder to observers that this by definition is not a solution of any type as it is not a rigorous proof. literally trillions upon trillions of variations are completely ignored without justification (beyond vague engine/player evaluations).

reminder again to all observers that chess ≠ checkers. chess is literally over a quadrillion times more complex, and the solutions (of all types) are many million times more complex.