Chess will never be solved, here's why

But there's also Asimov and the Foundation.
Robots - to do the world's dirty work.  More of it than now.
That's already under way. 
But nothing like to the degree it will be.  (if various things don't cause extinction first)
And there's that concept of 'humaniform' robots.

Thought this had all finished.

     That's unlikely to happen until chess is solved to everyone's satisfaction. You and I are likely too old to see the finale.

In a few million years, probably.

That's presumably why you call yourself Optimissed.

Indeed. I've heard that people die. I've even seen it but I think maybe I can avoid it by concentrating hard or something.

Me too. It has all finished, to my satisfaction anyway. 

Correct.
And those (is it plural) worried that the discussion goes on ... but there they (plural?) are.
The conversation might stop or will.  Because it always does.
Even in worldwide conversations - conversations seem to shut down during the 'night' in north america..
But then it starts up again.  

The conversation will probably go on for another reason.
"Solving" chess is what chessplayers are doing anyway.
The topic relates to chessplayers.  

#2475
“Well then, why don't you write such a paper? “
++ All journals are interested in “Chess is solved”, none in “Chess can be solved”.

“A couple of questions I asked you, are still pending.“
++ I have answered many questions from you and others.
Maybe you should number your questions. I summarize the present state.

Zürich 1953 Candidates’: 210 games = 120 draws + 90 decisive, D = 90 / 120 = 0.43, E = 0.37
Hence
Drawn games with 0 errors (?) = ideal games with optimal moves: 78
Drawn games with 2 errors (?): 29
Drawn games with 4 errors (?), or 2 errors (?) + 1 blunder (??): 4
Drawn games with 6 errors (?), or 4 errors (?) + 1 blunder (??), or 2 errors (?) + 2 blunders (??): 1
Decisive games with 1 error (?): 78
Decisive games with 3 errors (?), or 1 error (?) + 1 blunder (??): 11
Decisive games with 5 errors (?), or 3 errors (?) + 1 blunder (??), or 1 error (?) + 2 blunders (??): 1

ICCF WC 30: 136 games = 127 draws + 9 decisive games, D = 9 / 136 = 0.0662, E = 0.0659
Hence
Drawn games with 0 errors (?) = ideal games with optimal moves: 126
Drawn games with 2 errors (?): 1
Decisive games with 1 error (?): 9

The game-theoretic value of the initial position is a draw.

For the initial position a strategy to achieve the draw against any opposition is to follow an ICCF WC drawn game as long as possible and as soon as the opposition deviates let engines calculate for 5 days per move and whenever the engines disagree let a human ICCF grandmaster decide.

#2485
“An ultra weak solution requires a proof of the result.”
++ 97.9% of AlphaZero autoplay at 1 min / move draws.
Time * 60 yields 5.6 times less decisive games.
This even stays true with rules changed to stalemate = win.
93% of ICCF WC games draw.
ICCF allows table base win claims of > 50 moves without capture or pawn move.
If the game is a draw under less drawish rules, then the game is a draw under standard rules.

“That applies whether "chess" is taken to mean basic rules chess, competition rules chess, ICCF chess (different from either) or the game you have eventually decided you want to solve (none of the above).”
++ It makes no difference. ICCF allows table base win claims > 50 moves without capture or pawn move, but such win claims do not occur. 50 move draw claims are allowed but do not occur in ICCF WC draws. Most ICCF games are over before move 50, average = 39.

“Even if the starting position in ICCF chess is a draw, almost all the draw results are agreed draws.”
++ 74% are agreed, 16% are 3-fold repetition draws, 10% are table base draw claims, 0% are 50 moves draw claims. Of the agreed draws most are because a known drawn endgame has been reached, like with opposite coloured bishops or a rook ending with symmetrical pawns. Some are draws of convenience because of tournament standings. You could argue that maybe one ICCF grandmaster wrongfully accepted or offered a draw in a won position.
Even so draws by 3-fold repetition or table base claim are indisputable.

“Since the ICCF games are not played under the same rules as the game you say you want to solve, it doesn't follow that an ideal game under ICCF rules would be an ideal game under your rules.”
++ The ICCF rules are more decisive than the standard rules, as ICCF allows table base win claims of > 50 moves without capture or pawn move. If ICCF is drawn, then the standard game with less decisive rules is even more a draw. The 50 moves rule is never invoked in ICCF WC, so without consequences can be considered unwritten for simplicity.
I have no special rules: I want to solve chess as it is. I only remark that the 50-moves rule plays no role in solving chess and thus without any consequences can be ignored for simplicity.
I likewise remark that if 2-fold repetition is optimal, then 3-fold is optimal too, and thus 3-fold can without any consequences be simplified to 2-fold.

“I actually posted two positions,”
++ Please only discuss positions right after a capture or pawn move, i.e. with the 50-moves or 3-fold repetition reset to 0. It is pointless to discuss artificial constructs where the optimal move would trigger a draw by the 50-moves rule or the 3-fold repetition rule because of previous non-optimal shuffling.

“I suppose that's a big improvement on your usual accuracy in counting.”
++ There is no need for such wry remarks.

“There are more attributes to a position than just where the pieces are on the board.”
++ No, those are artificial constructs. Start a position right after a capture or a pawn move, i.e. with the 50-moves and 3-fold repetition counter reset to 0.
It is pointless to discuss positions where the optimal move would trigger a 50-move or 3-fold repetition draw because of previous non-optimal shuffling.

“you don't know your arse from your elbow. Instead of just reposting the same junk”
++ There is no need to become rude.

“The second position was taken from the endgame KRPP v KRP which you suggested. Do you think your computation will spend most of its time considering positions from that endgame or were you just trying to find a distribution of material where you thought SF14 wouldn't choose four blunders?”
++ I chose KRPP vs. KRP as it is a common 7-men endgame thus most representative.

“You're supposed to be proving the draws, not just "knowing" them.”
++ Some endgames with > 7 men have already been proven to be draws by human endgame experts. A proof can use previously proven lemmas.

“How do you propose to code your "knowledge" into the program anyway?”
++ The good assistants must once in a while check the progress of the 60 h / move calculation.
If they see an endgame known as a proven draw, then they can halt the calculation to save time.

“what has KRPP vs. KRP got to do with anything?”
++ It is a common endgame, and with 7 men it is closest to the middle game.

“You don't even know if SF14 will produce more or less blunders given 60 hours think time per move on your supercomputer.”
++ I derive that from AlphaZero autoplay at 1 s / move and 1 min / move: time * 60 = 5.6 times less decisive games and thus 5.6 times less error / game.
It also follows from logic: you can do the same or better at 60 h / move than at 1 min / move.

“you would have to try out all alternative moves”
++ No, I do not have to try all alternatives.
Per Capablanca 1 d4 and 1 e4 are the best moves. AlphaZero corroborates that.
If 1 e4 and 1 d4 are proven draws, then I do not have to try 1 a4.
We know that giving up a piece for no compensation loses the game.
1 e4 e5 2 Ba6? loses for white. That is why I do not have to try 2 Ba6?

Clever. Journals are interested in proofs of concept too, as we are, and you "forgot" to add "in five years". The rest of your post is the usual useless repetition, plus hypothesis #8, presented as a fact, and falsity (to not say "lie") #6.

You want to know my questions... You can find the first at the end of this post. I asked it maybe six times and you gave one wrong answer in two words. Answer honestly, and maybe I'll be willing to discuss the other problems in your theory.

You posted this link several times http://library.msri.org/books/Book29/files/schaeffer.pdf.

It's an MSRI publication on the subject “Checkers can be solved” (written before a solution was claimed).

You could try them.

Let us know how you get on.

Nicely done .

Hahahahhaah.

Can you post a game where SF14 plays from one of your tabiya to the 7 man tablebases using only ply count 0 positions, please? I'd be interested.

If you look at the comment I inserted at the bottom of the first example in #2485, the roundish thing after "Ply count =" was meant to represent 0.

Here's another one for you

 The only move to win in your new game is 3...Kc4

Are you definitely sure there's only one of these in every 100,000,000,000,000,000,000 positions?

(Not that it matters of course.)

Curiously, you missed out the last half of the paragraph to which you are responding viz:

"If you look at these games its blunder rate per ply at 1 second per move on my desktop was 0.9% and at 37 minutes per move 4.8% - from the identical position."

An increase from 1 min to 60 h is aroundf 3500.

An increase from 1 second to 37 minutes is around 2000.

How does your "logic" make a distinction between those factors?

(Regarding the 60h per move - we're all eagerly awaiting your response to @haiaku's final question here.)

#2499
“Can you post a game where SF14 plays from one of your tabiya to the 7 man tablebases using only ply count 0 positions, please?”
++ 99% of ICCF WC draws are ideal games with optimal moves. None of those games reach 50 moves without capture or pawn move. The rules of the game are such that optimal play compels a trade or a pawn move within 50 moves, except when the men are sparse like in some endgame table base positions like KNN vs. KP.

“If you look at the comment I inserted at the bottom of the first example in #2485, the roundish thing after "Ply count =" was meant to represent 0.”
++ That position counts 5 men, that is well within the strongly solved 7-men endgame table base and far from the uncharted middle game territory of 8 – 26 men.
The position probably can result from a reasonable game with > 50% accuracy e.g. from an endgame KNNP vs. KBBP where the stronger side sacrificed both bishops to queen its pawn.
As the position is a forced win, it cannot result from an ideal game with optimal moves.
The position probably cannot even result from a reasonable game with > 50% accuracy.
How long did you run your desktop for depth 29? Your desktop is 1000 times slower than a cloud engine of 10^9 nodes/s. Time * 60 gives 5.6 times less error. For the same time your desktop errs 5.6 * log(1000) / log(60) = 9.4 times more than the cloud engine.
SF14 apparently misjudges knight pairs. KNN vs. K is a draw, but it is +6.

“Here's another one for you”
++ Yes, this is a better example. It has 7 men, that is the boundary between the uncharted middle game of 8 – 26 men and the strongly solved 7-men endgame table base.
The position probably can result from a reasonable game with > 50% accuracy.
As the position is a forced win, it cannot result from an ideal game with optimal moves.
How long did you run your desktop for depth 33? Your desktop is 1000 times slower than a cloud engine of 10^9 nodes/s. Time * 60 gives 5.6 times less error. For the same time your desktop errs 5.6 * log(1000) / log(60) = 9.4 times more than the cloud engine.
SF14 apparently misjudges knight pairs. KNN vs. K is a draw, but it is +6.

The most representative examples are drawn KRPP vs. KRP. It counts 7 men, at the boundary between the strongly solved 7-men endgames and the uncharted 8 – 26 men middle games. Rook endings are the endgames that occur most. If it is a draw, then it can result from an ideal game with optimal moves. The side a pawn down must play accurately to avoid losing.
How to draw is not obvious like in say KNNP vs. KNN: sacrifice a knight for the pawn.

#2382
"After spending 60 hours to add to the search only 4 candidate moves for white and one child for each of them (so you said). So after five years how many nodes will have been searched?"
++ I already answered that, but you think my answer was wrong.
If you know the answer, then why do you ask? Are you a teacher?
I now give you 2 wrong answers.

Wrong answer 1:
If we branch b = 4 for m = 39 moves deep, then you could think we need
1 + b + b² + b³ + ... + b^39 = (b^40 - 1) / (b - 1) = (4^40 - 1) / 3 = 10^23 nodes.
This is wrong for 2 reasons:
1) It is clearly too high as there exist no 10^23 legal, sensible, reachable, and relevant positions.
2) Chess has many transpositions
e.g. 1 e4 e5 2 Nf3 Nc6 3 Bc4 = 1 e4 Nc6 2 Nf3 e5 3 Bc4 = 1 e4 e5 2 Bc4 Nc6 3 Nf3

Wrong answer 2:
If to account for transpositions we assume all white moves can be permuted, then we get
1 + b + b²/2 + b³/3! + ... + b^m/m! = exp (b) = exp (4) = 55
This is wrong for 2 reasons:
1) 55 is clearly too low
2) Not all moves can be permuted. E.g. in the above example e4 must precede Bc4.

Two wrongs make one right.
The number of nodes you ask for lies between 55 and 10^23.