Chess will never be solved, here's why

Sort:
tygxc

@6801

"The trick is ignoring most of the lines"
++ Schaeffer weakly solved Checkers with only 19 of the 300 tournament openings.
200 were irrelevant because of transpositions. 81 were pruned.
If he was allowed to prune, then weakly solving Chess may and should prune as well.

tygxc

@6804

"you are just making up stats"
++ No, I fitted a Poisson distribution of the errors / game to the results of the ICCF WC Finals.

"You would need a perfect chess playing chess engine to check the game."
++ No, I do not, I use statistics to conclude that the ICCF WC Finals draws are > 99% certain to be perfect games with 0 errors.

"Stockfish the strongest chess playing chess engine we have.
Is not designed to play perfect chess as a type B Shannon chess engine."
++ It is not designed to do so, but can be used to do so given sufficient time.
If it would calculate 10^44 positions, then it would strongly solve chess, i.e. a 32-men table base, but that would take too much time and storage.
If it calculates 10^17 relevant positions, then it weakly solves chess. That can be done in 5 years.

"Even given infinite time on the world fastest computer." ++ Infinite time solves a finite game.

"Stockfish would not, and could not play perfect chess."
++ That is not necessary. It is enough that the table base optimal move is among the w top moves (e.g. w = 4) of the engine with less than 1 error in 10^17 positions.

"looking deeper then any human chess player"
++ Calculating until the 7-men endgame table base is the deepest that can be looked. 

"human players can outthink stockfish"
++ Yes, in ICCF humans + engines play stronger than Stockfish alone.

Elroch
tygxc wrote:

"Even given infinite time on the world fastest computer." ++ Infinite time solves a finite game.

Even a stopped clock is right twice a day.

Elroch

This possibility has been discussed here. I (possibly overconfidently) tried to sketch how it could be done. The basic idea is to do something like a tablebase construction but all in parallel with superimposed states. Wild! You need quite a lot of entangled qubits that can be manipulated a large number of times without any harm done by noise.

Incidentally, I understand that since thermal noise is such a problem to quantum computing, quantum error correction is key to such ambitious things ever being possible.

1c0nIc
DesperateKingWalk wrote:
tygxc wrote:

@6791

"ICCF WC games are not perfect games with perfect play for both sides."
++ Not all ICCF WC Finals games are perfect games with optimal play from both sides.
The decisive ICCF WC Finals games are not perfect games and contain 1 error,
usually the last move before resignation.
Not all drawn ICCF WC Finals games are perfect, but >99% of them are perfect games with optimal play from both sides. The remaining <1% contain 2 errors that undo each other.

"What device or chess program was used to created the perfect game with perfect play."
++ An ICCF (grand)master with engine and average 5 days per move.

"what device or chess program was used to prove they were perfect games with perfect play"
++ Statistics: fitting a Poisson distribution of errors / game to the ICCF WC Finals results.

"No chess program exist that can play perfect chess."
++ Agreed, but with more time per move it approaches it.
At 1 s / move: 88.2% of games perfect. At 1 min / move: 97.9% of games perfect.

"This thread would not exist as chess would be solved."
++ For all practical purpose chess is already ultra-weakly solved,
the game-theoretic value of the initial position being a draw.
Chess is already in part weakly solved. ICCF WC games show how to draw
against 1 e4, 1 d4, 1 c4, and 1 Nf3, but not yet in all variations.
Also the teams of grandmasters and engines that helped prepare over the board human classical world championship matches have already weakly solved parts of chess.
Chess is already strongly solved for all 7-men positions and for some 8-men positions.

"Not all drawn ICCF WC Finals games are perfect, but >99% of them are perfect games with optimal play from both sides"

Again you are just making up stats. To know if a game was perfect. You would need a perfect chess playing chess engine to check the game. 

But no perfect chess playing engine exist. Stockfish the strongest chess playing chess engine we have. Is not designed to play perfect chess as a type B Shannon chess engine. Even given infinite time on the world fastest computer. Stockfish would not, and could not play perfect chess. 

Type B chess engine were designed to surpass human chess players. By looking a very few line of play, but looking deeper then any human chess player. 

This works greats for surpassing human chess play. But falls painfully short in finding the best moves in a chess game. 

This is why Stockfish in some positions is completely stupid, and human players can outthink stockfish. 

This would not and could not happen to a chess engine that plays perfect chess. 

stockfish couldn't play perfect chess consistently even on the best computer ever, but I can't imagine that all or even most of the moves that stockfish thinks up are going to be different then what a perfect chess engine would play.

LouLomes

Chess will be solved. I read someone where it said that given 1000+ years, computers can look at every possible move. That 1000+ years will come down as chess computers get better 

hrarray
Almost all of the legal positions possible make no sense, so 1000 years isn’t as far off as you think. You don’t need to solve for all possible lines, only those that make sense.
milet220

Right, but we're here for further.

tygxc

@6814

"10^50 is the number of positions that need to be calculated"
++ Please stop such nonsense. There are only 10^44 legal chess positions,
the vast majority of them makes no sense like 3 rooks or bishops per side.
If you calculate all of those then you strongly solve chess i.e. create a 32-men table base.
To weakly solve chess you only need to calculate 10^17 relevant positions. That takes 5 years.

tygxc

@6812

"We do not know what perfect play looks like."
++ We know exactly what perfect play looks like. Here is an example:
https://www.iccf.com/game?id=1164344

"Chess computers have improved every year, year over year." ++ Yes, they make fewer mistakes.

"If chess computers are playing perfect chess. This could not obviously happen."
++ Chess computers do not play perfect chess, but approach it if given more time per move.
Maybe the top 1 engine move is not perfect, but then its top 2, top 3, or top 4 move is.

tygxc

@6820

"No one knows the exact number." ++ We know the exact number of 10^44 legal positions.
https://github.com/tromp/ChessPositionRanking

"The Shannon number is 10^120 of positions."
++ That is wrong. Shannon estimated 10^120 games not positions and that was wrong too. There are between 10^29241 and 10^34082 possible chess games.
https://wismuth.com/chess/longest-game.html 

"lets go with 10^44 as the true number of legal positions needed to solve chess."
++ That would be to strongly solve chess, i.e. generate a 32-men table base.
To weakly solve chess, as Schaeffer did for Checkers, requires 10^17 relevant positions.

tygxc

I try to explain once more so that even those with IQ 60 can understand.

Chess has 10^44 legal positions. Strongly solving Chess would need to visit all of these,
which would take too much time and storage to be feasible.
10^44 means that if we analyse 44 moves deep with 10 choices each time,
then we have the whole of Chess.
10^44 = 2^146 so if we analyse 146 moves deep with 2 choices each time,
then we have the whole of chess.
10^44 = 4^73 so if we analyse 73 moves deep with 4 choices each time,
then we have the whole of chess.

Weakly solving chess needs to find only 1 black move as an answer to each white move.
In the initial position 20 white moves and 20 black answers each give 20*20 = 400 positions.
Now if only 1 black response to each white move, that gives 20*1 = 20 = sqrt (400) positions.
So in essence the difference between strongly solving and weakly solving is a square root.

Assume now we try to weakly solve chess considering all legal white moves, good or bad.
So that would lead to sqrt (10^44) = 10^22 positions, taking 500,000 years.
That is without any limitation: all legal moves considered, no matter how bad.

Let us now modify the Laws of Chess such that a pawn can only promote to either a queen, or a piece previously captured. That is a minor restriction. In the > 1000 perfect games with optimal play from both sides we have from the ICCF WC finals draws, 3 rooks or 3 bishops never occur on one side. That would reduce the restricted legal positions to 10^37 * 10 = 10^38.
If we try to weakly solve chess on that basis, then we need sqrt (10^38) = 10^19 positions.
That would take 500 years.
That is without any pruning, except a minor restriction on pawn promotions.

After inspecting a random sample of 10,000 such positions without promotions to pieces not previously captured, we observe that none can result from optimal play by both sides.
That leads to 10^38 / 10,000 = 10^34 reasonable positions.
If we try to weakly solve chess on that basis, then we need sqrt (10^34) = 10^17 relevant positions. That can now be done in 5 years.

It is possible that GM Sveshnikov envisioned an even smarter and more powerful pruning down to 10^16 relevant positions.

 

hrarray
So far, (since I have joined this conversation) the predictions for the time required to “solve” chess have been: 1000 years, 31 nonillion years, billions of years, and 500,000 years…
tygxc

@6832

"predictions for the time required to “solve” chess have been: 1000 years, ..."
++ That all relates to strongly solving Chess: a 32-men table base with all 10^44 legal positions.
Weakly solving chess just like Schaeffer weakly solved Checkers takes 5 years.

lfPatriotGames
hrarray wrote:
So far, (since I have joined this conversation) the predictions for the time required to “solve” chess have been: 1000 years, 31 nonillion years, billions of years, and 500,000 years…

I've said a few times it will be in about 200 years. Not every possible position kind of solved, but the beyond a reasonable doubt kind of solved. Right now, because we are still in the very early stages of computer infancy, we can only just throw random guesses out. We aren't even close to any educated guesses. 

 

tygxc

@6834

"I've said a few times it will be in about 200 years" ++ Sveshnikov said, I calculated 5 years.

"beyond a reasonable doubt kind of solved" ++ Weakly solved like Checkers.

"we can only just throw random guesses out" ++ We can calculate.

"We aren't even close to any educated guesses." ++ Some guess, I calculate.

chessisNOTez884

Now I get what you get by arguments..

So many useful things 🤣🤣🤣🤣

Ego satisfaction

Time wasting

Energy wasting

Beating someone continuously then gets beaten up indirectly very badly(in terms of argument)

Enjoying the aggressive atmosphere by blabbering trash hearing trash and seeing trash 

Just BE QUIET PLEASE chess forums are for discussion..  7000+ posts OMG.. see in any argument both the sides have their points but just be QUIET pls

 

tygxc

@6836

"You said 5 years" ++ Yes, after Sveshnikov said so.

"you said many things that are false" ++ No.

"It is magical how you know the exact number of legal chess positions"
++ No, Tromp has calculated that. https://github.com/tromp/ChessPositionRanking

"the length of a optimal chess game" ++ We know that too. In ICCF WC Finals games lasted between 13 and 119 moves, 42 moves average, with standard deviation 16.

"It is magical how you know what lines are meaningful, and what lines can be removed."
++ That is no magic, it is logic. Obvious errors can be removed, like 1 e4 e5 2 Ba6?
Also by pure logic 1 e4 and 1 d4 cannot be worse than 1 a4, so 1 a4 can be removed.
1 Nh3 cannot be better than 1 Nf3, so 1 Nh3 can be removed.

"you claim Stockfish can play perfect chess"
++ Not my claim. I claim at 17 s/move on a billion nodes/s engine the table base exact move is among the top 4 engine moves with 1 error in 10^20 positions. 

"a method of making a 32 man tablebase of only 10^17 positions"
++ Not my claim. A 32-men table base of 10^44 positions is beyond current capability.
However, based on 10^17 relevant positions perfect play is possible: black has one path to the 7-men endgame table base draw or a prior 3-fold repetition whatever white does.

"Meaning never losing a game from the starting position."
++ Schaeffer has done the same for Checkers, with only 10^14 relevant positions of the 10^20 legal positions and by analysing only 19 of the 300 tournament openings.

"you are the only one on the planet that possesses such knowledge" ++ Sveshnikov was first.

"We would already have a 32 man tablebase of only 10^17 positions" ++ That is impossible.

"they continue to work on the current tablebase method of using retrograde analysis and looking at every position" ++ An 8-men endgame table base is work in progress.
The idea of weakly solving chess is not to create a 32-men table base, i.e. strongly solving chess, but to calculate from the initial position towards the 7-men endgame table base.

"demonstrate the perfect play of Stockfish"
++ Stockfish does not play perfect chess. It approaches it if given more time/move.

MARattigan
tygxc wrote:

...

"you said many things that are false" ++ No.

...

many+1

(I hope your nose doesn't get longer each time; it could become totally debilitating.)

avramtparra
tygxc wrote:

@6748

"Solving chess means finding an optimal strategy for the game of chess, that is, one by which one of the players (White or Black) can always force a victory, or either can force a draw (see solved game) - Wikipedia"
++ Wikipedia is not the authority. The authority on this is Prof. van den Herik.
"ultra-weakly solved means that the game-theoretic value of the initial position has been determined,
weakly solved means that for the initial position a strategy has been determined to achieve the game-theoretic value against any opposition, and
strongly solved is being used for a game for which such a strategy has been determined for all legal positions."

"The only way to do that is to go through the entire position tree"
++ No, you can ultra-weakly solve Chess without going through any tree at all.
To weakly solve Chess you only have to go through all 10^18 relevant positions.
To strongly solve Chess you have to visit all 10^44 legal positions.

"you don't know if black's position is a starting loss" ++ We do know Chess is a draw.

"you literally don't know if the starting position is a win loss or draw of white" ++ It is a draw.

"what I am talking about is literally solving chess"
++ Ultra-weakly solving Chess merits no further discussion: we know it is a draw.
Strongly solving chess merits no further discussion: 10^44 legal positions is too much.
Weakly solving chess is interesting and the 10^18 relevant positions can be done in 5 years.

"Strongly solving chess, but if you think about its the same case"
++ 10^18 relevant positions is not the same as 10^44 legal positions.

@6775

Can I just revive your comment to say your words right here

""The only way to do that is to go through the entire position tree"
++ No, you can ultra-weakly solve Chess without going through any tree at all.
To weakly solve Chess you only have to go through all 10^18 relevant positions.
To strongly solve Chess you have to visit all 10^44 legal positions."

actually show you have a weak understanding of whats going on. Going through positions is going down a tree path meh, i actually wish there was a facepalm emoji.

if youre not open to to being wrong its like talking to a brick wall, I do not claim to know everything hell look at my mf elo, but as soon as you said "ICF players play better than Stockfish" there is just no hope