Do you think chess computers will eventually be able to do this?

Right now, chess computers give 2 types of evaluations: +/- X advantage, or Mate in X.
However, chess isn't really like this.
A position, in the end, is always either winning for black, a draw, or winning for white.
With the current rate of technological advancements, do you think eventually, stockfish (or another, newer chess engine) can look at a position, and say one of three things:
Draw
Mate for White in X moves (ex. If white is up a rook in the middlegame, it is probably white to win in ~50-100 moves with the best play)
Mate in Black in X moves

I know chess has many possible positions, I looked it up and there are about 10^40 legal positions, so this may be a bit far fetched.
Do you think this will ever happen? If so, when? (also if this did happen it would be really funny if computers figured out chess was actually White to win in 3,065 moves with perfect play)

Side Note: Do you ever think about how a side is never winning by +/- 3 or +/- 6, because in the end, it will result in white/black winning with perfect play?
That also means there is no such thing as an inaccuracy or a mistake. A move is either worsening (Winning -> Draw, Draw -> Losing, or Winning -> Losing) or not.

This is called a tablebase, where positions with x number of pieces are completely solved, so every outcome is known. So far positions with 7 pieces or fewer are solved. Tablebases are generated retroactively - reducing a 4-piece position to a known winning 3-piece position "solves" that position, and so on.

It gets exponentially harder to get each successive tablebase, so this is almost impossible to do for non-endgame positions anytime soon.

Won't happen, never ever. Tablebases for up to 5 pieces are 939 MB large. 8-piece tablebases are estimated to have a size of 8 PB (1 Petabyte = 1.000 Terabyte). Which is nice to know but to even get there is a life long work or so I guess.

By the way: Since SF 15 or 16, the +/- advantage doesn't display the material/positional advantage a side has. Instead it the shows the winning probability. For more: https://github.com/official-stockfish/Stockfish/wiki/Stockfish-FAQ#interpretation-of-the-stockfish-evaluation and https://github.com/official-stockfish/Stockfish/discussions/4754 . So there has been some kind of progress towards a more realistic evaluation method.

@ThatRandomPerson111 said in #1:

Right now, chess computers give 2 types of evaluations: +/- X advantage, or Mate in X.
However, chess isn't really like this.
A position, in the end, is always either winning for black, a draw, or winning for white.
With the current rate of technological advancements, do you think eventually, stockfish (or another, newer chess engine) can look at a position, and say one of three things:
Draw
Mate for White in X moves (ex. If white is up a rook in the middlegame, it is probably white to win in ~50-100 moves with the best play)
Mate in Black in X moves

This is just a wordy way of saying "Do you think chess will ever be strongly solved?"

I know chess has many possible positions, I looked it up and there are about 10^40 legal positions, so this may be a bit far fetched.

I thought it was actually far more than that. Like more positions than atoms in the universe (which is about 10^80 I believe).

Do you think this will ever happen? If so, when?

It doesn't seem plausible unless we make some monumental discovery about the universe that allows us to harness different timelines and merge them back together or something like that. Or maybe break out of the universe entirely.

(also if this did happen it would be really funny if computers figured out chess was actually White to win in 3,065 moves with perfect play)

Why? What is so special about that number? I remember someone calculating the longest possible chess game, but I thought it was something like 6000 moves. Maybe that was half moves though.

Side Note: Do you ever think about how a side is never winning by +/- 3 or +/- 6, because in the end, it will result in white/black winning with perfect play?
That also means there is no such thing as an inaccuracy or a mistake. A move is either worsening (Winning -> Draw, Draw -> Losing, or Winning -> Losing) or not.

Yes, indeed it is interesting. Although to be honest this fact will probably never be relevant for human or even engine play.

Another interesting thought experiment is this: Suppose chess were solved, so that in any position we could produce a list of moves that preserved the evaluation (i.e., win, loss, or draw). Then, create an engine that selected the worst-looking move from this list for some definition of "worst-looking" (such as given the worst evaluation by Stockfish 17) and played it. Would this engine be easy to draw against even though it always played perfectly and drawing against it would also require perfect play? I think it would, but there's no way to know for sure. Also, if current Stockfish 17 played against such an engine, what would its maximum evaluation for itself be, even though the actual evaluation never changed (assuming Stockfish does not make a mistake)? I imagine it would be maybe +3 or so, but maybe it would be even higher. A similar question is, what is the position for which Stockfish's evaluation most disagrees with the actual evaluation (which would only be win, loss, or draw). There's that famous puzzle with all the knight promotions that Stockfish will actually evaluate as winning for the losing side even if you let it think for a long time.

@AsDaGo said in #4:

I thought it was actually far more than that. Like more positions than atoms in the universe (which is about 10^80 I believe).
No. 10^80 can easily be refuted. Each of 64 squares can be in one of thirteen states. 13^64 equals 1.96 * 10^71. Even with a factor for castling and en-passant, you end up under 10^80.

Others have done better estimations, e.g.
https://www.chessprogramming.org/Shirish_Chinchalkar#Number_of_Chess_Positions

There are some interesting aspects about evaluations.

Sure, from a game theory standpoint only +1, 0, -1 make really sense. Other values are more representing winning chances.

At the same time, imagine the engine would tell you "draw" in the starting position. For human play, this would be a rater useless information. Even winning would need some more information to make progress (i.e. moves to mate, or similar).

I am not so sure that for the "average player" the latest ultra strong engines are the most useful ones if they want to understand the game better. Their evaluation is based on so deep and refined tactical calculation which often seems irrelevant for human play.

Sure, a good move is a good move, and it's hard to argue with that, but I really wish there would be more refined evaluation, like "positionally, White is really good, but Black has this tactical shot hence Black is better".

That's really important question: winning chances... for whom? A strong player looks at a position and can see that white is a piece down but having more activity and better development more than compensates for it so that white is clearly better. When I sit at the board, there is a good chance that ten moves later, the positional advantages will be gone but I'll still be a piece down.

And the problem with engines is that they are way stronger than even the strongest human player. Recently I was watching a streaming of a (classical) game where the commenter (GM titled) said on multiple occasions something like "Yeah, the computer says it's equal but that's because it's really good at defending. I definitely wouldn't want to play this as black."

So maybe some effort should be made to go in exactly in the opposite direction than classifying "white winning" / "equal" / "black winning": to teach the engines to provide also some human oriented additional score, e.g. to distinguish between "0.0" meaning "you are essentially safe whatever you play, except for obvious blunders" and "if you play this sequence of 5 counterintuitive moves, the position will still evaluate as equal, if you miss one, you lose quickly".

@sheckley666 said in #5:

No. 10^80 can easily be refuted. Each of 64 squares can be in one of thirteen states. 13^64 equals 1.96 * 10^71. Even with a factor for castling and en-passant, you end up under 10^80.

True, I think I was thinking of possible games, not possible positions.

@nadjarostowa said in #6:

There are some interesting aspects about evaluations.

Sure, from a game theory standpoint only +1, 0, -1 make really sense. Other values are more representing winning chances.

At the same time, imagine the engine would tell you "draw" in the starting position. For human play, this would be a rater useless information. Even winning would need some more information to make progress (i.e. moves to mate, or similar).

I am not so sure that for the "average player" the latest ultra strong engines are the most useful ones if they want to understand the game better. Their evaluation is based on so deep and refined tactical calculation which often seems irrelevant for human play.

Sure, a good move is a good move, and it's hard to argue with that, but I really wish there would be more refined evaluation, like "positionally, White is really good, but Black has this tactical shot hence Black is better".

Yeah, on my small understanding of the game, engines have to be taken with a grain of salt. Ofc they are picking the best moves, but its function as a tool for us is not that, is to reach a greater understandment of the game, we need to know WHY is the best move.

Yesterday, i was watching the SCC final, a Levy and Dania we're pretty surprised because the best engine move was completely out of their understanding. If an IM and a GM cant follow the engine, it means that in that particular situation is following a logic that, probably, any human can follow, making it not ussefull as a tool.

Of course this are rare cases, but sometimes we need to understand our own limits.

I hope that i maked my point clear, and that it makes some sense hahaha

No it doesn't make any sense.

Of course no one can understand engine moves in a blitz time control but that doesn't mean it's not possible to figure out with more time why some move is the best. Certain correspondence chess players did this regulary.

Of course I do know that some positions in chess, i.e. endgames, are not comprehensible for humans but this does not apply here.