Chess will never be solved, here's why

It is a joke that you can prove anything by ignoring the moves that don't superficially appear to be in the top 4 candidates to a weak player.

Anyone who believes that does not know what "solving chess" means.

There are people who refuse to have particular things proven to them ...
but there's a converse too it seems.
People who insist that some things are proven - that are Not proven.

#2024

"Don't repeat the same thing over and over, as if we don't read your posts or we were stupid, just answer the questions, please." ++ The answer is yes. The explanation is for those who did not read or could not understand my previous wording.

"You have stated that countless time, no need to repeat yourself."
++ OK, but that is the answer to your question.

"You stated that you expect to search only 10¹⁷ positions out of 10³⁷ and in order to do that, you want to search only 4 candidates for White and one for Black at any move."
++ Yes, that is correct.

"how do you determine those 4 top candidates and be sure the optimal move is among them?"
++ I determine the 4 top candidates with the Stockfish evaluation function or a simplified version of it. I reckon the optimal move is among them by extrapolation: 1 error in 10^5 positions for the top 1 move, 1 error in 10^10 positions for the top 2 moves, 1 error in 10^15 moves for the top 3 moves, 1 error in 10^20 moves for the top 4 moves. That should do as I plan to consider only 10^17 moves.

"You invoked best-first algorithms as if they were some sort of magical tools."
++ Best first is a heuristic used in solving Losing Chess. For example 1 d4 and 1 e4 are better than 1 a4. Thus if 1 e4 and 1 d4 fail to win for white, then 1 a4 will fail too and needs no investigation. Likewise 1 e4 e5 2 Ba6 is the worst move. If 2 Nf3, 2 Nc3, 2 d4, 2 Bc4 cannot win for white then surely 2 Ba6 cannot win either and needs no investigation.

"You stated that to be sure the optimal move is among those candidates, you would just leave SF calculate 60 hours per move." ++ Yes, that is right.

"If I understand you well, it should do that at any increment of depth; otherwise, you cannot be sure that the top 4 candidates generated in leaf positions before the endgame tablebase are the best, since the engine did not calculate for 60 hours those positions, right?"
++ Yes, the optimal move must be among the 4. Not all 4 must be the best: occasionally top 1, top 2, top 3 will all be mistakes and only top 4 wil be optimal.

"The closer the endgame, the less reliable would be those 4 candidates."
++ No, the closer to the endgame the more table base hits and thus the more reliable.

@MARattigan - I guess we could debate the semantics of 'unique' ...  but that could simply be one of those situations where there's an argument where both parties actually Agree !  (whether both realize it or only one or neither realizes that)  
In other words it could be about what meaning is preferred for 'unique positions'.
Or is 'assigned' for 'unique positions'.

so people like chess and don`t like chess but they`ll try and e4 e5 e6 f6 f7 g8 d4 = good so chess is fun and like acting in a real army, let your friends go and login at chess.com play blizits learn lesson so go and tell them.

but`s what the anwser?

Best-first searches are not a novelty invented to solve antichess.

You don't understand, or you pretend to not understand? Don't be elusive. You answered everything but the last question, that I ask for the third time, I think: how many hours to select those 4 best options at any increment of depth? 1 position = 1 nanosecond is not the right answer, and neither 1 nanosecond per candidate move.

@tygxc is still trying to assign 'time per position' as a constant ??
and as a too small inadequate constant for 'solving' Too ?
He's still on that?
I only had to read @haiaku 's post - to be 'notified' about that 'possibility' ...  

I try to keep my mind open.  Objective to possibilities.
Would it not be funny if @tygxc is proven Right ...  about Everything ? 
Hey it is a possibility?
Chess puzzles and other things can teach us ...
'Things aren't always what they seem to be'.

Regarding 'solving' chess - another concept:
the status of positions - how many true categories are there ?
I'm not talking about the material situations there.
Like for example:  "in this position the supercomputer could not find a forced win for either side - after a default number of hours of time invested in it - but there's Much capacity to 'go wrong' in the position.  Neither side need offer a draw"
And that could have additional 'attributes'.
Like - the supercomputer was able to look at all possible ensuing move sequences - or - it didn't have time for that within the 'default' time assignment.  

I'm too lazy to read this

#2029
"You don't understand, or you pretend to not understand? Don't be elusive. You answered everything but the last question, that I ask for the third time, I think: how many hours to select those 4 best options at any increment of depth? 1 position = 1 nanosecond is not the right answer, and neither 1 nanosecond per candidate move."
++ I am not elusive. I guess I do not understand your question. Do you want to know the time to find 4 candidate moves in 1 node of the solution tree? What do you mean by the phrase 'at any increment and depth'? What precisely do you ask and even for the 3rd time? I am willing to think about it and come up with an answer. Please bear in mind that I have not solved chess, so I cannot tell you how I have done it. I have to think about how it can be done. I even refine my own thinking.

#2032
That is not lazy: 2034 posts...
The essence is this quote by the late GM Sveshnikov:
"Give me five years, good assistants and modern computers, and I will trace all variations from the opening towards tablebases and 'close' chess."
give me = somebody has to pay for it
five years = 5 years
good assistants = to prepare starting positions to launch the search
modern computers = cloud engines of 10^9 nodes / second
all variations = the relevant ECO codes
the opening = the starting positions prepared by the good assistants
towards = starting from the opening and ending at the table base
tablebases = 7-men endgame table base
'close" chess = weakly solve chess, as already done for Checkers (8 * 8 draughts) and Losing Chess

Here is one of the more blatantly obvious of your errors. 

You think that Stockfish makes no more than one losing error in every 100,000 moves (see quote).  AlphaZero beat (an earlier version of) Stockfish 155 times in 1000 games.  According to you (who would surely have said the same thing about that version of Stockfish, which was not much more than 100 Elo points weaker), those games had an average of at least 155 * 100000 / 1000 moves = 15,500 moves.

My recollection is that this is not so, by a factor of more than 100.

You would be foolish to think the incremental improvements in Stockfish have provided that factor of more than 100 (in fact it would have to be a great deal more in practice).

"At any increment of depth", not "and depth". To answer your question, I simply follow your reasoning:

Let's say that after the last Black's move the endgame is reached and it's a draw. The other 3 candidates for White (and the Black's reply for each of them) must be checked. Let's suppose that all of them lead to a draw. So the parent node for those 4 candidates (which is a Black move) at depth -2 (in plies) from the endgame leads to a draw. The engine should check now the other options for White at depth -3 down to the endgame; if Black draws in all these lines, the engine should check the candidates at depth -5, and so on.
If White wins in any of these lines, it would require no less time, because the engine should check White's strategy against all the reasonable alternatives for Black (you say that), not just one. So, in the hypothesis that the game ends in a draw, how many hours would occur (at 60 hours per position) to generate all these candidates between the starting point (which I understood is not the initial position) and the endgame? Is it clearer now?

Well, at least we both think you have to .

But apparently not to write it.

So I was correct in my interpretation of @tygxc's claim? It is reasonable to assume a misinterpretation would have received a response.

Correct, I think, only because you said, "at least 155 * 100000 / 1000 moves = 15,500 moves".

You have to remember AZ was probably making compensatory blunders.

When I tried it out in a situation where I could check the blunders against perfect play via the tablebases, here, SF14 gave me 30 blunders in 497 moves in the basic rules game (which would be the game most akin to what @tygxc now proposes).

That's a little more than 1 blunder in 17 moves. Since there were only five men on the board in a well analysed endgame where SF14's evaluation function should be more accurate than average,  I think it's fair to say that blunder rate is also an underestimate. 

But I think you're naive to expect any response from @tygxc other than to restate his original assertion. 

He could say, totally neglecting pathology in game trees, that those errors occurred because SF didn't spend 60 hours every move to determine those 4 top candidates in the first place.

@tygxc

#2022

"The Troitzky line concept is meaningless with the 50 move rule in effect."
++ The Troitzky line is very useful even in practice: when the blocked pawn is too far, there is no win.

There are wins with the pawn anywhere on its 7th. rank. How far does it have to go?

What I meant was you can't rely on blocking the pawn behind the Troitzky line winning in the competition rules game.

I thought of that objection, but it fails because merely spending 1000 times more time would be unlikely to improve results by more than 200 points. To see this is generous observe that with AlphaZero a bit more than 100 points stronger than Stockfish 8, it was necessary to give a 30:1 time handicap to make Stockfish the favourite. 30:1 to cover 100 points would be 900:1 for 200 points (but the empirical data also suggests the returns get steadily lower as you provide more time).  In truth, @tygxc is assuming incorrectly that he knows how much time Stockfish needs to achieve perfection. Far more likely is that this is the amount of time needed to blunder P% of the time, where P is small but non-zero.