Chess will never be solved, here's why

Sort:
MARattigan
tygxc wrote:

@7107

"What on Earth are you talking about?" ++ This is the longest perfect game with optimal play from both sides to reach the 7-men endgame table base draw.
https://www.iccf.com/game?id=1164280 

No it's not.

It's Jon Edwards v Sergey Adolfovich Osipov from the ICCF WC 32, which ended in an agreed draw with 13 men on the board.

This is the same game continued using Arena/Stockfish. It ended with a claim under the 50 move rule on move 236.



It took 119 moves to the 7-men endgame table base draw i.e. 238 ply.

You may as well leave out the endless translations from moves to ply; I think we can all manage it. Especially since you appear to have insurmountable problems counting up to 7.

All the other perfect games reach the 7-men endgame table base draw or a prior 3-fold repetition sooner.

Big red telephone again, right?

"But how does that relate to your proposed method of solution?"
++ If you calculate all reasonable white moves, then 1 tentative black response, then all reasonable white moves, then 1 tentative black response and so on then the 7-men endgame table base draw or a prior 3-fold repetition is reached in at most 119 moves, i.e. 238 ply.

You don't have a reasonable definition of "reasonable".

You claim that all perfect games that are not Jon Edwards v Sergey Adolfovich Osipov from the ICCF WC 32 reach the 7-men endgame table base draw or a prior 3-fold repetition sooner than Jon Edwards v Sergey Adolfovich Osipov from the ICCF WC 32. That's true of perfect games that reach one or other, but only because Jon Edwards v Sergey Adolfovich Osipov from the ICCF WC 32 never reached either. 

But perfect games need only perfect moves, they don't have to include moves that tygxc thinks are reasonable. You make no connection between perfect moves and moves that tygxc thinks are reasonable.

Indeed, earlier in the thread you accused the Syzygy tablebase of trolling in this game.

Is it your contention that Syzygy is trolling and all it's moves are reasonable?

Another explanation: chess has 10^44 legal positions. 10^44 = 2^146.

Here you're a victim of your own misinformation.

Tromp's estimate of the number of basic rules positions is 4.82 x 10^44.  NOT 10^44. 

2^148 < 4.82 x 10^44 <2^149.
So after 146 digital decisions you get the whole of Chess.

By "digital decisions" I assume you mean choices of moves. (Or are you still struggling to count up to 7 on your fingers?)

You make no mention of whether or not the moves are perfect so the continuation of Jon Edwards v Sergey Adolfovich Osipov would be a case in point.

I can't find anywhere in that game the position after 1.e4 e5 2.Ba6. Can you? Wouldn't that be included in Chess?

Neither can I find any of the large number of positions with the same diagram and ply count, say, 13 under the 50 move rule. Would they not also be included in Chess (though of course not in Tromp's number)? 
Checkmates that exceed 146 moves exist, but they must contain a string of forced moves.

So far as I can understand your logic (not very far), that seems to rest on on the obviously invalid assumption that the number of basic rules positions associated with the competition rules positions occurring in all continuations is the product of the number of choices of moves in each such competition rules position or something of the sort. Choice of perfect moves maybe?

That would need Tromp's upper bound rather than his estimate for a valid proof (as well as a different argument). Also you would need to say exactly what you mean by "forced" and how many moves constitute a "string". (Do you include strings of one?.)

Here is a provably perfect (just in case it needs to be) checkmate in 148. Can you indicate some strings of forced moves?

Mainly, I can't see your point even if what you say is true. It is to be expected that with most definitions of "forced move" such strings will occur in a long sequence of moves whether as part of a win or draw. Are you trying to make a relevant point or are you just away wi' the fairies? 

Edit: sentence reinserted for context ->But how does that relate to your proposed method of solution? "Stockfish doesn't do perfect play."
++ Agreed, but that does not matter. Stockfish only needs to generate the reasonable white moves.

In the vanishingly unlikely event that you ever got a sponsor it might matter to him.

How are you going to make it generate only moves tygxc thinks are reasonable?

Stockfish then selects the 1 black response without worry if perfect or not.
If a 7-men endgame table base draw or a prior 3-fold repetition is reached, then that validates all black responses as fit to draw.

Obviously not.

"99.9 % of my games are perfect" ++ No, they are not.

Are so !

"same way you got your ICCF stats." ++ No.

++Yes. So there !

 

tygxc

@7116

"which ended in an agreed draw with 13 men on the board"
++ The transition to a 7-men endgame table base draw is forced.

"This is the same game continued using Arena/Stockfish." ++ Very well, this shows the need for human assistants to cut short such needless calculations and call it a draw.

"All the other perfect games reach the 7-men endgame table base draw or a prior 3-fold repetition sooner. Big red telephone again?"
++ No, statistics. Between 13 and 119 moves, 42 moves average, with standard deviation 16.

"If you calculate all reasonable white moves, then 1 tentative black response, then all reasonable white moves, then 1 tentative black response and so on then the 7-men endgame table base draw or a prior 3-fold repetition is reached in at most 119 moves, i.e. 238 ply.
You don't have a reasonable definition of reasonable".
++ That is just the best first heuristic as used in solving Checkers and Losing Chess.
If the 4 best moves cannot win for white, then the worst moves cannot win either.

"But perfect games need only perfect moves"
++ Statistics applied to the ICCF WC Finals games show they are > 99% sure to be perfect games i.e. they contain optimal moves from both sides.

"Tromp's estimate of the number of basic rules positions is 4.82 x 10^44"
Yes, but the factor 4.82 is irrelevant and should be 1.205 because of up / down symmetry and left / right symmetry after loss of castling rights.

"2^148 < 4.82 x 10^44 <2^149" ++ Yes

"By digital decisions I assume you mean choices of moves"
++ Yes, less than 149 choices between 2 moves, or less than 74 choices between 4 moves.

"You make no mention of whether the moves are perfect or not"
++ That is only legal choice, if they need to be perfect there is even less choice.

"I can't find anywhere in that game the position after 1.e4 e5 2.Ba6. Can you?"
++ It is clear that 2 Ba6? is not optimal play by white.

"Wouldn't that be included in Chess?" ++ Yes, 1 e4 e5 2 Ba6? belongs to the 10^44 legal positions, but not to the 10^17 relevant positions.

"Neither can I find any of the positions with the same diagram and ply count 13 under the 50 move rule" ++ The 50-moves rule plays no role. Games with optimal play from both sides end in draws long before the 50-moves rule would trigger.

"How are you going to make it generate only moves tygxc thinks are reasonable?"
++ Stockfish ranks the legal moves. Then it is the best first heuristic.
If the best white moves cannot win, then the worse white moves cannot win either.

"If a 7-men endgame table base draw or a prior 3-fold repetition is reached,
then that validates all black responses as fit to draw."
++ People here still fail to understand this, though it is simple.
If all reasonable white moves fail to win against a black response, then that black response is optimal. It does not matter how that black response was obtained. It does not matter if other black responses draw as well or not.
If white cannot win against those black responses, then Chess is weakly solved.
If the black responses lead to a table base draw or a prior 3-fold repetition,
then they are optimal in retrospect.

MARattigan
tygxc wrote:

@7116

"which ended in an agreed draw with 13 men on the board"
++ The transition to a 7-men endgame table base draw is forced.

You just shovel the s*** on don't you. 

You have it off the big red telephone that neither side can resign, checkmate or force a triple repetition or dead position before the material reduces to 7 men. How does that affect my statement that the game you linked to doesn't prove that the maximum length perfect game is 119 moves?   

"This is the same game continued using Arena/Stockfish." ++ Very well, this shows the need for human assistants to cut short such needless calculations and call it a draw.

Again, how does that affect my statement that the game you linked to doesn't prove that the maximum length perfect game is 119 moves? Why change the subject?  

"All the other perfect games reach the 7-men endgame table base draw or a prior 3-fold repetition sooner. Big red telephone again?"
++ No, statistics. Between 13 and 119 moves, 42 moves average, with standard deviation 16.

Bad statistics. Not even showing statistically that there's a small probability of a perfect game longer than 119 moves.

If you were to apply your oft cut and pasted analysis to the games shown here and use your previously stated assumptions, you should come to the conclusion that the most probable size of your sample of perfect games from ICCF WC 32 is 0. Why don't you try it?

You can't do much useful statistics with a sample size of 0.

Even if you assumed they were all perfect, it would be interesting to see what you would assign as the probability that there were no longer perfect games in the whole of chess.

I'm sure you could make it 10^17 %. Go on. Give it a go!

"If you calculate all reasonable white moves, then 1 tentative black response, then all reasonable white moves, then 1 tentative black response and so on then the 7-men endgame table base draw or a prior 3-fold repetition is reached in at most 119 moves, i.e. 238 ply.
You don't have a reasonable definition of reasonable".
++ That is just the best first heuristic as used in solving Checkers and Losing Chess.
If the 4 best moves cannot win for white, then the worst moves cannot win either.

By which, I take it, you still don't have a reasonable definition of "reasonable".

"But perfect games need only perfect moves"
++ Statistics applied to the ICCF WC Finals games show they are > 99% sure to be perfect games i.e. they contain optimal moves from both sides.

Only when you apply them, but statistics is obviously not your strong point.

"Tromp's estimate of the number of basic rules positions is 4.82 x 10^44"
Yes, but the factor 4.82 is irrelevant and should be 1.205 because of up / down symmetry and left / right symmetry after loss of castling rights.

The factor 4.82 is relevant because in ignoring it you arrive at the conclusion, "so after 146 digital decisions you get the whole of Chess", whereas if you had not ignored it you might have arrived at, "so after 149 digital decisions you get the whole of Chess", which would have been closer to what you had intended (albeit still flagrant nonsense).

More to the point others may make similar mistakes if you keep posting misrepresentations of Tromp's figure.

It shouldn't be reduced to 1.205 ^10^44 because "the whole of Chess" doesn't mean one position in four.

"2^148 < 4.82 x 10^44 <2^149" ++ Yes

"By digital decisions I assume you mean choices of moves"
++ Yes, less than 149 choices between 2 moves, or less than 74 choices between 4 moves.

For a proof you'ld need an upper bound, not an estimate. More to the point you'ld need a valid argument. Among other things some explanation of why you're using figures for basic rules positions in the context of competition rules.

"You make no mention of whether the moves are perfect or not"
++ That is only legal choice, if they need to be perfect there is even less choice.

"I can't find anywhere in that game the position after 1.e4 e5 2.Ba6. Can you?"
++ It is clear that 2 Ba6? is not optimal play by white.

How is that remotely relevant? Your statement was, "So after 146 digital decisions you get the whole of Chess" (optimal or not, presumably).

"Wouldn't that be included in Chess?" ++ Yes, 1 e4 e5 2 Ba6? belongs to the 10^44 legal positions, but not to the 10^17 relevant positions.

How is that remotely relevant? Your statement was, "So after 146 digital decisions you get the whole of Chess" (included in your silly subset or not, presumably).

"Neither can I find any of the positions with the same diagram and ply count 13 under the 50 move rule" ++ The 50-moves rule plays no role.

The 50 move rule plays no role in your head, but it does in chess under FIDE competition rules. It says so in the handbook.

Moves that could draw under the triple repetition rule from the ply count 13 positions would not draw from the ply count 3 position. 

Games with optimal play from both sides end in draws long before the 50-moves rule would trigger.

Another tip from the big red telephone? You still haven't realised that He feeds you BS because He knows you're not going to listen to a damn word anyway.

How do you define "optimal"?

"How are you going to make it generate only moves tygxc thinks are reasonable?"
++ Stockfish ranks the legal moves. Then it is the best first heuristic.
If the best white moves cannot win, then the worse white moves cannot win either.

So how are you going to make it generate only moves tygxc thinks are reasonable?

"If a 7-men endgame table base draw or a prior 3-fold repetition is reached,
then that validates all black responses as fit to draw."

Obviously not.


++ People here still fail to understand this, though it is simple.
If all reasonable white moves fail to win against a black response, then that black response is optimal. It does not matter how that black response was obtained. It does not matter if other black responses draw as well or not.
If white cannot win against those black responses, then Chess is weakly solved.
If the black responses lead to a table base draw or a prior 3-fold repetition,
then they are optimal in retrospect.

It is simple or you are simple?

If all White moves fail to win against a Black response, but succeed in drawing, while all White moves against some different Black responses lose, most rational people would say that Black response is not optimal. 

 

SirHolland

: D

mpaetz
tygxc wrote:

@7112

"analysis will done on only those opening moves the experts consider relevant"
++ Analysis will not be done on moves that are clearly no optimal play e.g. 1 e4 e5 2 Ba6, 1 e4 e5 2 Nf3 Nc6 3 Ng5 etc. It is to not waste engine time on what is already obvious.

"will only analyze lines from drawn ICCF grandmaster games"
++ The drawn ICCF WC Finals games serve as a backbone of already completed analysis as they each already represent 2 years of engine analysis under guidance of an ICCF grandmaster.
It is to speed up the process.

"This seems to make the entire enterprise reliant on humans and engines that are known to be imperfect." ++ The enterprise relies on the 7-men endgame table base known to be perfect.

"Am I misunderstanding something?" ++ Yes

     So it is true that you will rely on the judgement of imperfect humans to decide what is/is not "optimal play". And it is true that you will ignore many variations to save time. And it is true that you will not consider openings not played to a draw in ICCF games.

     My conclusion is that this enterprise will not satisfactorily answer the question of whether chess is a win for either side or a draw.

MARattigan

And so are you for once.

Eton_Rifles
mpaetz wrote:
tygxc wrote:

@7112

"analysis will done on only those opening moves the experts consider relevant"
++ Analysis will not be done on moves that are clearly no optimal play e.g. 1 e4 e5 2 Ba6, 1 e4 e5 2 Nf3 Nc6 3 Ng5 etc. It is to not waste engine time on what is already obvious.

"will only analyze lines from drawn ICCF grandmaster games"
++ The drawn ICCF WC Finals games serve as a backbone of already completed analysis as they each already represent 2 years of engine analysis under guidance of an ICCF grandmaster.
It is to speed up the process.

"This seems to make the entire enterprise reliant on humans and engines that are known to be imperfect." ++ The enterprise relies on the 7-men endgame table base known to be perfect.

"Am I misunderstanding something?" ++ Yes

     So it is true that you will rely on the judgement of imperfect humans to decide what is/is not "optimal play". And it is true that you will ignore many variations to save time. And it is true that you will not consider openings not played to a draw in ICCF games.

     My conclusion is that this enterprise will not satisfactorily answer the question of whether chess is a win for either side or a draw.

After 7119 replies, this is still a good debate, although enjoyable, a lot does whoosh over my bald swede, but I do have to agree with the above statement. 

    "My conclusion is that this enterprise will not satisfactorily answer the question of whether chess is a win for either side or a draw."

If chess is to be solved, a system must know everything about every position; I believe the term is called a "complete information game".

No university, entrepreneur or business has taken the challenge outlined by tygxc et al. because, in the real world, there's nothing we can learn from solving chess. Real-world problem-solving costs money but ultimately makes money.

 

tygxc

@7120

"judgement of imperfect humans to decide what is/is not optimal play"
++ No, the 7-men endgame table base is the judge: draw / win / loss.
We do not need to decide what is optimal play.
As for the single black response:
if it leads to a 7-men endgame table base draw, then it is optimal.
As for the several white moves: we do not need to decide if one of them is optimal or not,
it is enough not all selected candidate moves are not optimal.
That is the difference between analysis,
which ends in some imperfect evaluation like 'white is slightly better', +=, +0.50.,
and solving, which ends in a 7-men endgame table base draw or a prior 3-fold repetition.
That is also the difference of playing with Stockfish: it has to decide on one move to play,
and solving with Stockfish: it can retain several e.g. 4 candidate moves for further analysis.

"you will ignore many variations to save time" ++ Yes, but only those we are sure about.

"you will not consider openings not played to a draw in ICCF games"
++ ICCF WC Finals games only open with 1 e4, 1 d4, 1 c4, and 1 Nf3 and for good reason.
The same in human top competition: only those four nowadays.

"answer the question of whether chess is a win for either side or a draw"
++ That question has been answered long ago: Chess is a draw.
The real question is: 'How?'
A related question is: 'How long does it take?'
If doing it the stupid way with just a computer and calculating all the junk like 1 e4 e5 2 Ba6? down to the end, then weakly solving Chess needs 10^19 positions and takes 500 years.
If doing it a smart and clever way with humans and computers,
then weakly solving Chess needs 10^17 positions and takes 5 years, like Sveshnikov said.

chaotikitat

This thread is still going? Impressive 

tygxc

@7124

"If chess is to be solved, a system must know everything about every position"
++ That would be strongly solving chess to a 32-men table base,
requiring all 10^44 legal positions, which would take too much time and storage.
Weakly solving chess requires only 10^17 relevant positions, which can be done in 5 years.

"Real-world problem-solving costs money but ultimately makes money."
++ Weakly solving Chess costs 3 million $ of money and makes no money.
The same for weakly solving Checkers or Losing Chess:
these cost money and did not make any money, but were done anyway.

willithius

personally i agree

Elroch
tygxc wrote:

@7124

"If chess is to be solved, a system must know everything about every position"
++ That would be strongly solving chess to a 32-men table base,

Yes.
requiring all 10^44 legal positions, which would take too much time and storage.
Weakly solving chess requires only 10^17 relevant positions, which can be done in 5 years.

Still a fail.

This amounts to an ignorant attempt to redefine the very precise word "solve" to mean something sloppy and very ambiguous, replacing valid deduction with the strong beliefs of unreliable evaluators, themselves based on heuristics generalised from a tiny number (compared to the size of chess) of examples using induction. For example "the side that is a piece up will win except when it is obvious to an unspecified unreliable evaluator that there is a reason they won't"

tygxc

@7128

If you only admit a stupid and dumb way of weakly solving, then it takes 500 years.
A smart and clever way takes 5 years.

charmquark314

That's not why. Solving chess, at least in the sense of strongly solving chess, would be to provide an algorithm that evaluates every legal position to "white mates with perfect play", "black mates with perfect play", or "draw", and provide a move that does not change that evaluation. (Since chess games have a finite length, said algorithm can consistently win any winning position, and force a draw in any drawn position.)

No chess engine we have right now can do that. Such an engine's self-play would result in either consistent wins from one side, or consistent draws.

Yes, that does mean we have solved chess with up to 7 pieces on the board (an algorithm for that is available on this website).

mpaetz

     Still, we all believe that the computers and programs of the year 2100 will far outclass those of the present. Why should we just assume that those entities will not be able to punch large holes in the analyses that the Sveshnikov five year plan might come up with? Couple that with the  unanalyzed lines the human experts judge to be unworthy of consideration and the immense number of possibilities that exist following opening moves that the plan will never examine and the "solution" would be unconvincing.

slaveofjesuschrist

If I walked a quadrillion years, could I reach the end of the multiverse?

BoardMonkey

One-hundred tredecillion possible positions most of which are illegal. It's a lawless multiverse.

BoardMonkey

Game by game? That's an unsurmountable problem. We'll never get a solution. Reminds me of Asimov's short story The Last Question.

Elroch
Optimissed wrote:

A way that doesn't work at all takes 5 years.

It is surprising to me that it is beyond @tygxc to understand that what he is proposing is not a way to solve chess, but a way to replace an uncertain evaluation of the opening position by a less uncertain (but still uncertain) one. We need any random person involved in peer-reviewed research on the subject to explain to him that he has got it wrong (since he can't glean this from their work, like most people can).

MARattigan
charmquark314 wrote:

That's not why. Solving chess, at least in the sense of strongly solving chess, would be to provide an algorithm that evaluates every legal position to "white mates with perfect play", "black mates with perfect play", or "draw", and provide a move that does not change that evaluation. (Since chess games have a finite length, said algorithm can consistently win any winning position, and force a draw in any drawn position.)

"Chess" denotes several different games each with its own set of solutions. Of the FIDE versions only games played according to the post 2017 competition rules are limited to a finite length. Games played under post 2017 basic rules or all pre 2017 rules are not.

That means that for the unlimited games simply providing a move that doesn't change the evaluation is not enough. E.g. for a position with this diagram (with the White king on one of the two squares shown) and White to play ...

an algorithm that recommends moving the king to the other square doesn't change the evaluation but also doesn't solve the position.

And strictly speaking the evaluation is necessary at most for drawn positions in the unlimited games. Just a move will do.

 No chess engine we have right now can do that. Such an engine's self-play would result in either consistent wins from one side, or consistent draws.

True, which is not to say a consistent result indicates perfect play. It may be the case that SF15 would draw against itself no matter how many attempts it made with less than geological think time per move from this Black winning position, for example.

(As opposed to a random legal move generator v SF15 which would probably achieve the mate in far fewer attempts than a monkey on a typewriter would need etc.) 

Yes, that does mean we have solved chess with up to 7 pieces on the board (an algorithm for that is available on this website).

'Fraid not. For example it doesn't do the final position in this competition rules game which is a mate in 16.

(Try using the top move it shows against the computer in "Analysis".)