Chess will never be solved, here's why

If someone's too busy wanting to think their own thoughts, they won't understand what they're being told. You have no interest in this subject and you would struggle to discuss it if you tried. All you want to do is back up people who don't understand plain English.

Your level. Just dim the lights a little further and that'll be perfect.

I have commented far longer and more cogently on all the various solving chess threads over the years.  You have little to no understanding of the software/technology involved, and base your entire premise on your human understanding of chess, just like Ponz did.  And you are dead wrong, just like Ponz was (and like Tygxc is now).  You pretend that your arguments have some weight behind them, but you are ponzificating(tm) using circular and subjective garbage with nary a fact or data point to be found, as he did.  The characterizations of your arguments as assumptions is dead on, and will continue to be dead on in perpetuity, since you can no longer do more than skim material and cast aspersions ala your stance on the terminology/definitions of solving games.  It is you that lack understanding, and your definition dodging is quite transparent in that regard.

I think you need to have the final word.
Doesn't stop me wondering how someone can be as clever as you are and still think they're clever. It must take a great deal of doing, to be so clever, when you had so much potential.

Anyway, btickler and mpaetz aside, I've explained that it is not necessary to analyse chess under the assumption that black is capable of winning by force from the opening position. Some people may wish to do so, but a result gained from analysing chess with the assumption that black can't win by force is sufficient to stand alone, as an important step forward, were it achieved. Much more important than the reverse assumption that black may win by force, which nearly everyone rightly discounts in any case.

They do seem to believe that it involves an assumption. How terrible that would be, were it true but, of course, it just means dividing the analysis into two parts and doing the important half first. Clearly, I've been arguing with the intelligentsia of chess.com and naturally, I can't hope to persuade them of anything. Perhaps normal people might understand it sufficiently though, or maybe they already do. I don't think it's going to be achieved any time soon. Perhaps normal people also may understand what "stand alone" means and what a stand-alone result for part of the analysis would mean and imply.  So far as the intelligentsia goes, I can't see much hope. Mental degeneration struck early.

It is true that it is sometimes worth picking a proposition whose value is unknown and dividing a proof into two parts - one assuming it is true and one assuming it is false. However, it is worth noting that this was not done for checkers. Rather a strategy which drew for white and a strategy which drew for black were exhibited with no assumptions.

[I recall that there is a significant mathematical theorem which was difficult to prove, but which was found possible to prove based on each of two assumptions - that the Riemann Hypothesis was true and that the Riemann Hypothesis was false! Of course the two conditional results proved the result, even though it remains unknown if the Riemann Hypothesis is true].

@8631

"He claims that engine evaluations are close to error-free."
++ A cloud engine of 10^9 nodes/ s running for 17 s/move,
or a desktop of 10^6 nodes/s running for 4.7 h make 1 error in 100,000 positions.
However, the solving method does not depend on that, it depends on reaching the 7-men endgame table base with its exact evaluation draw / win / loss.

Again, your one "error" in 100,000 claim is based on imperfect evaluations.  No engine knows definitively how many "errors" are in a chess game. 

Wonder away, it's about as fruitful as anything else you are spending your time on here.

@8644

"one error in 100,000 claim is based on imperfect evaluations"
++ Again, it is not based on evaluations.
It is based on extrapolation from statistical data on game results.
No need for imperfect or perfect evaluation.

Extrapolation of what statistical data, exactly?  Keep following the breadcrumbs back...

@8647

Figure 2

10,000 AlphaZero games 1 s / move: 11.8% decisive games
1,000 AlphaZero games 1 min / move: 2.1% decisive games

Extrapolating:
1 h / move: 2.1% * 2.1 / 11.8 = 0.37% decisive games
60 h / move: 0.37% * 2.1 / 11.8 = 0.067% decisive games

More time leads to less decisive games.
With unlimited time no decisive games: Chess is a draw.
Each decisive game must contain an odd number of errors.

Converting 60 h / move to 10^9 nodes / s: 17 s / move.

Thus 0.067% games with an odd number of errors.
Thus 1 error in 1503 games.
Assuming 100 positions / game: 1 error in 150,300 positions.
Thus 1 error in 10^5 positions.

Thus 1 case in 10^20 positions that the table base exact move is not among the top 4 engine moves at 10^9 nodes/s during 17 s.

Extrapolations are obviously imperfect evaluations. I thought you represented yourself as having been a mathematician. It's such a basic understanding, that extrapolations are risky, that even someone who had been capable at mathematics and is now struggling would retain such a basic understanding. I don't think you have ever been expert in maths.

I'm now sure that puts you in the same class of people as btickler, Brother Fester and mpaetz. People whose thinking is certainly not to be relied upon. The people I mentioned have all made such elementary mistakes in understanding, over the past two days, that they should be described as incompetent. OK for insulting other people when their bad judgements are exposed but that's about it. A sad testimony of people's minds being affected by the aging process, in some cases prematurely.

If you can imagine a graph, with points representing definite co-ordinates which represent observed correspondences between values, say, on the x and y axes, these points being joined up by a line drawn in a way that is intended to represent a likely correspondence between the points representing definite information, an extrapolation is an attempt to extend such a line outside the points that we know about. "Extra-polate" .... that is, "extend outside the points (on a graph)".

Obviously, many extrapolations are guesses. They only cease to be guesses where it is known that all such extrapolations are accurate because the relationship continues to follow its expected course.

If you're solving chess by 'brute force' you need to consider all the possible moves by at least one of the players.Random moves are a statistical sample of all possible moves and are thus well suited to making an estimate of average game length.

The majority of possible moves are in fact of low quality.

Things you read after doing google searches are not always right.

Even if you get to zero, all you will have proved is that AlphaZero plays itself to a draw, not that chess is a draw.  You still try to pretend...but engine play is not perfect.  It is entirely possible for engines to plateau at an imperfect level of play until some major breakthrough, just as humans do.  The number of games played by engines might seem like a lot, and the history of human chess play might seem like a lot, but we're talking only trillions of games.  That is next to nothing at the scale required to solve chess.

Your "each decisive game must contain an odd number of errors" (by which *you* mean moves that provably cause the game state to change to a non-draw) is also incorrect, because your definition of "error" is still circular.  You cannot currently quantify the errors accurately in any game of chess.  That determination will only be available *after* chess is solved, the same way that you cannot determine relative piece values accurately until chess is solved.  You can approximate with the current best engine play, but never prove.

@8651

"you need to consider all the possible moves by at least one of the players"
++ 'Next to brute-force methods it is often beneficial to incorporate knowledge-based methods in game-solving programs.'
To weakly solve Chess it is only necessary to look at the reasonable white moves.
It is not necessary to look at 1 e4 e5 2 Ba6? which does not oppose to the draw.
The question is: 'How to draw as black?' and not 'How to win if white makes an error?'

"Random moves are a statistical sample of all possible moves and are thus well suited"
++ Random moves are not suited as well as good moves.

"making an estimate of average game length"
++ From ICCF WC Finals: average 42, shortest 13, longest 119

You clearly cannot recognize incompetence, or you would recuse yourself from this thread and every scientific thread you currently infest.  If there is an elementary mistake of understanding, produce it and show your work in terms of proving it.  Any one will do.

As always, you will not be able to back up a word of your meandering thought process.  In a decade, you never have done so (successfully).  You can't even find a post from the day before when called upon to do so.  It's about time you recognized whose aging is getting the better of them at 70+.  Not that you were stellar prior...  

@8652

"Even if you get to zero, all you will have proved is that AlphaZero plays itself to a draw, not that chess is a draw." ++ If an engine plays at unlimited time to the 7-men endgame table base draw, then Chess is weakly solved and it is formally proven that Chess is a draw, as we already know.

"engine play is not perfect" ++ No, engine play is not perfect: at 17 s/move on a 10^9 nodes/s cloud engine it makes 1 error in 100,000 positions. That is not perfect, but close to it.

"trillions of games.  That is next to nothing at the scale required to solve chess."
++ Weakly solving Chess takes 5 years on 3 powerful engines piloted by grandmasters.
We know Chess is a draw, but we do not yet know exactly how.
We have > 1,000 perfect games with optimal play from both sides,
but that is only a tiny fraction of the 10^17 positions relevant to weakly solving Chess.

"your definition of error is still circular"
++ No, it is not circular. An error (?) changes the game state from draw to loss, or from won back to draw. A blunder or double error (??) changes the game state from won to lost.

"You cannot currently quantify the errors accurately in any game of chess."
++ I can quantify the errors approximately with a good engine.
Statistics on a sufficiently large tournament of sufficient level can also quantify the errors.

"That determination will only be available *after* chess is solved"
++ It is irrelevant if of 120 ICCF WC Finals draws there are 116 perfect games with 0 errors + 4 flawed games with 2 errors that undo each other, or if it is 115 + 5 or 117 + 3.

"relative piece values" ++ Piece values are only a heuristic.

Just a small quibble. A brute force search isn't practicable. Looking at all permutations of moves takes billions of years, even with the latest computers. So solving chess isn't really possible until and unless an algorithm is developed, which can recognise accurately a bad move and cut it out. It's necessary to leave the fuzzy area which consists of moves that are only probably bad. Of course, a human that can't tell that 1. e4 e5 2 Ba6 definitely loses for white isn't going to help to define good and bad moves. For practical reasons, their lack of willingness to try to cut out bad moves means that this project of solving chess is a non-starter, so they really shouldn't be involved. They're the human equivalent of "bad moves".

For the purposes of solving, a bad move is any move that changes the game state and a good move is one that doesn't. The game state can be defined as the theoretical outcome of any position, given best play by both sides. You say that random moves are a sample of all possible moves and would give an accurate representation of game length; but I don't think that's correct because a game of chess is a different thing from a random permutation of legal moves. The latter isn't suited to solving chess in any meaningful way. A solution won't be viable until an accurate way is found to cut out obvious errors and, of course, most random moves are obvious errors.

There's definitely a lack of focus in thinking about this subject, which is clearly displayed on this thread and which isn't helped by incompetents like btickler and a couple of others getting involved, because they're bored and want to troll. I wouldn't count Mr Desperate among these, although his thinking is not error-proof. Some of the others aren't even capable of considering interesting questions because their determination to troll is so powerful. Unfortunately there's no thread owner to block them. 

     Despite clearly stating that you wish to start with the assumption that black cannot win by force, you go on to claim that your proposed method does not rely on any assumption. Perhaps you are using two different definitions of "assumption". The definition (OED) "taking too much upon oneself, laying claim to undue importance, arrogance" may come into play here.

     Also, you admit that your proposed solution is merely "a first step" and not a conclusive result. Denying the possibility of your assumptions being unsound can only lead to an unconvincing conclusion. 

     Are you proposing analysis that will look at all black responses, no matter how unpromising they might appear at first glance (which does nothing to cut down the extent of the task) or will you first try to find a drawing continuation and then declare that line "solved" because you assume the impossibility of finding any winning line for black? I would see such a "proof" as unsubstantiated and would prefer an investigation into all possibilities.