@5997
The word of GM Sveshnikov is good enough.
Chess will never be solved, here's why
Sort:
Oct 10, 2022
0
#4564
Even if you did find a way to "figure" chess out, the Fischer chess 960 would eventually take its place.
However, I think there is simply too much to remember in chess, and only a handful of people would ever be "able to figure it out", even if it was possible.
6000 posts. Nice.
@6000
"I think there is simply too much to remember in chess"
++ Yes, but Kramnik used to study 10,000 games per month.
A player who has memorized 10,000 perfect games with optimal play by both sides from the weak solution of chess has an advantage.
Oct 10, 2022
0
#4566
He'll still get crushed by Stockfish. Chess is too big for humans to get it right.
@6002
"He'll still get crushed by Stockfish."
++ If Stockfish deviates from the memorized games without making an earlier error.
Elroch wrote:
tygxc wrote:
@5997
The word of GM Sveshnikov is good enough.
Absurd.
So after all this, it seems come down to what we mean my "solve". One "definition" (Elroch) involves considering every possible legal move in every possible legal position, so 0% chance of error. The other definition (tygxc) is more of a "for all practical purposes" definition that says "Sure, there's a .0000000000000000000001 chance we're wrong, but we can all just agree that's close enough, and a valid proof."
Everyone else seems to line up with one of these two. I clearly line up with Elroch, assuming I understand what he's saying.
Am I characterizing this correctly?
@6004
"So after all this, it seems come down to what we mean by solve"
++ The definition is clear:
'Weakly solved means that for the initial position a strategy has been determined to achieve the game-theoretic value against any opposition.
The game-theoretic value of a game is the outcome when all participants play optimally.'
https://www.sciencedirect.com/science/article/pii/S0004370201001527
"considering every possible legal move in every possible legal position"
++ No, that would be strongly solving chess to a 32-men table base with 10^44 legal positions.
"Am I characterizing this correctly?" ++ No, the definitions are clear.
There are 2 differences.
1) I interpret any opposition and all participants playing optimally such that white must resist against the draw and thus that white moves that clearly do not resist or clearly resist less and thus clearly are not optimal can be discarded. E.g. 1 e4 e5 2 Ba6?
2) I also take from the above peer-reviewed paper:
'Next to brute-force methods it is often beneficial to incorporate knowledge-based
methods in game-solving programs.'
I also note this MSc. dissertation: A Knowledge-based Approach of
Connect-Four The Game is Solved: White Wins.
http://www.informatik.uni-trier.de/~fernau/DSL0607/Masterthesis-Viergewinnt.pdf
I also note this peer-reviewed paper
Acquisition of Chess Knowledge in AlphaZero
https://arxiv.org/abs/2111.09259
That ranks 1 e4, 1 d4, 1 c4, 1 Nf3 as more optimal than the other 16.
So as knowledge is allowed and beneficial in solving a game, I use the knowledge acquired by AlphaZero with no other input but the Laws of Chess to restrict the weak solution to the 4 most optimal moves 1 e4, 1 d4, 1 c4, 1 Nf3 that oppose most to the draw and to discard the other 16 moves. If the 4 best moves only draw, then the 16 worst moves cannot win either.
I also note that Checkers has been weakly solved with only 19 relevant openings of the 300 imposed openings, thus that it is OK in solving chess to only consider relevant openings only.
@6005
Thanks.....both for taking the time to explain all this and for not ridiculing what was probably an over simplification on my part. I'm trying to see both sides of this, but most of this is very new to me. Interesting nonetheless.
Oct 10, 2022
0
#4571
mikekalish wrote:
Elroch wrote:
tygxc wrote:
@5997
The word of GM Sveshnikov is good enough.
Absurd.
So after all this, it seems come down to what we mean my "solve". One "definition" (Elroch) involves considering every possible legal move in every possible legal position,
so 0% chance of error.
That is close to the standard definition of a strong solution. While I have referred to it, I have mostly addressed the ntion of weak solution (a complete optimal strategy for each side). This is much smaller computationally, rather like if you always play the same opening lines you don't need to know the entire opening encyclopedia, just a small fraction of it (less than 1% in fact!). As you say, the key point is that you have to deal with every legal move the opponent can make, exactly like solving a mate in 2 problem.
The famous and challenging solution of checkers was a weak solution that required years of computation. @tygxc advocates a sort of sloppy version (where large parts of the solution are ignored, for reasons that would not be viewed as adequate by anyone working in this area, and then he still gets the complexity wrong (because each side often has many moves that can't be even sloppily rejected and games can be long).
The other definition (tygxc) is more of a "for all practical purposes" definition that says "Sure, there's a .0000000000000000000001 chance we're wrong, but we can all just agree that's close enough, and a valid proof."
Something like that. This is not satisfactory in a rigorous subject, which is why it took years to solve checkers, rather than a sloppy version being knocked off over weekend.
Everyone else seems to line up with one of these two. I clearly line up with Elroch, assuming I understand what he's saying.
Am I characterizing this correctly?
You are mostly right: I advocate strict rigour - the same as the peer-reviewed literature on solving other games.
It looks interesting as usual. Thank you so much for sharing with us.
Oct 10, 2022
0
#4573
tygxc wrote:
@6002
"He'll still get crushed by Stockfish."
++ If Stockfish deviates from the memorized games without making an earlier error.
10,000 is a tiny number. Not even enough for 14 binary choices! A drop in the ocean.
@6010
"10,000 is a tiny number"
++ 10,000 games is almost a million positions.
TsarKing wrote:
You guys are forgetting a bigger aspect to playing chess.
I SERIOUSLY doubt they are "forgetting" that.
Oct 10, 2022
0
#4576
tygxc wrote:
"So after all this, it seems come down to what we mean by solve"
++ The definition is clear:
'Weakly solved means that for the initial position a strategy has been determined to achieve the game-theoretic value against any opposition.
The game-theoretic value of a game is the outcome when all participants play optimally.'
...
There are 2 differences.
1) I interpret any opposition and all participants playing optimally such that white must resist against the draw and thus that white moves that clearly do not resist or clearly resist less and thus clearly are not optimal can be discarded. E.g. 1 e4 e5 2 Ba6?
...
Strange.
User @tygxc says here
I agree with
"Provide an algorithm that secures a win for one player, or a draw for either, against any possible moves by the opponent, from the beginning of the game. That is, produce at least one complete ideal game (all moves start to end) with proof that each move is optimal for the player making it."
In case of chess that would be:
"Provide an algorithm that secures a draw for black, against any possible moves by the opponent, from the beginning of the game." (my highlight.)
Are there two users with id tygxc?
Incidentally when are you going to post your calculations to determine the theoretical results of these positions and the number of errors in the games. If you do that we can all forget about your proposals and talk about sensible things.
You needlessly insisted that I produce a drawn KRPP vs KRP set of games for you while declining on spurious grounds to produce any yourself. (It's to assess the calculation that you have proposed).
I spent a week doing that. The least you can now do is fulfil your promise and post your calculations.
Oct 10, 2022
0
#4577
tygxc wrote:
@6010
"10,000 is a tiny number"
++ 10,000 games is almost a million positions.
Yes, a TINY number. There are about 10^38 times as many basic chess positions. To put it another way, the number of basic chess positions is much more than the 7th power of this number.
[No need to repeat the mantra about weak humans deciding most of the positions can be ignored based on vague and unreliable heuristics - we've heard it almost as many times as the size of your dataset].
@6014
"Yes, a TINY number. There are about 10^38 times as many basic chess positions."
++ There are 10^44 legal positions and of these 10^17 are relevant.
@6007
"The famous and challenging solution of checkers was a weak solution that required years of computation." ++ Checkers also used only 19 relevant openings of the 300 imposed openings.
"there's a .0000000000000000000001 chance we're wrong"
++ As there are 10^17 relevant positions, 1 error in 10^20 positions is OK.
"why it took years to solve checkers" ++ He had to write his own engine Chinook, he has to generate his own endgame table base, his computers were less powerful. It also takes 5 years to weakly solve chess, with Stockfish available, with a 7-men endgame table base available and with engines available that calculate a billion positions per second.
"I advocate strict rigour - the same as the peer-reviewed literature on solving other games."
++ The peer-reviewed literature shows Connect Four solved with knowledge only.
So weakly solving Chess can use knowledge too.
The peer-reviewed literature shows Checkers solved with only 19 relevant openings of the 300 imposed openings. So weakly solving Chess can discard irrelevant lines too.
There is no reason why solving Chess should be subject to more stringent requirements than Connect Four, Checkers, Losing Chess, or Nine Men's Morris.
Oct 10, 2022
0
#4580
tygxc wrote:
"there's a .0000000000000000000001 chance we're wrong"
++ As there are 10^17 relevant positions, 1 error in 10^20 positions is OK.
...
Apart from the fact that you can't explain why 10^17 and what your positions are relevant to, the figure of 10^20 depends on your "calculation" of error rates. You can check that by applying your "calculation" to my games here where we can check your results with Syzygy.
When are you going to do that as you promised?
I found a stone tablet with "Solving chess requires only 10^17 relevant positions and can be done in 5 years." written on it, so you can be sure we have a reliable basis for that.
Unfortunately, I dropped it, but take my word for it.