Chess will never be solved, here's why

#3411

"If humans and A0 (and engines) share a lot of knowledge, how can they be totally different"
++ Because in the vast majority of legal, sensible, reachable, and relevant postions all that acquired knowledge is not sufficient to select a correct move flawlessly.

"if I find that the evaluations of two equally strong engines are correlated, how can their errors (i.e. they play a non-optimal move), which are based on those evaluations, be totally uncorrelated?"
++ I gave the example of equally strong Tal and Petrosian engines. Sometimes the Tal engine wins because the Petrosian engine erroneously allows a correct sacrifice. Sometimes the Petrosian engine wins because the Tal engine sacrifices incorrectly. At the decision neither of them knows if it is correct or not. A bias in allowing or making sacrifices explains the errors.

"Should I do the homework for you?" ++ No, I did my homework, you do yours.

"Can you post the link to the exact page/s you are talking about?"
++ No, I use a link I cannot post here. Google, TCEC superfinals gets you there. https://tcec-chess.com/ is the official site, but a bit difficult to navigate.

"I am trying to determine if the statistical independence can be proven without assuming it as a premise" ++ Well TCEC provides ample data, just like ICCF

We agree on that, but that does not prove they their decisions are totally different and uncorrelated.

But you exclude that their ignorance can lead to a chain of errors by the two, because they both ignore one made a mistake, but to you at least one of them knows how to play perfectly, even with a flawed evaluation...

How convenient for you.

In other words, we cannot know exactly which data you used and how, and that should substatiate your assumptions. Fantastic.

Very precise and unambiguos, as usual.

The discussion is like a chess game in and of itself.
But it is tygxc who is playing it.

He is playing white here.  Its a middlegame.
Nothing personal.  It simply addresses the postings.
There are three pawns not moved -
in front of the castled white King at g1. 
And a rook at f1
If the pawns and that rook can never be taken and no knights can get close then the white king can never be mated.

The fpawn is to substitute 'nodes' for the hardware speed of computers.
Thereby cutting off the reality of their limited speed.
Suggesting to give that up (to wake up to reality) is like telling a flat earther that the horizon recedes.  But for the 'nodes' believer its like a belief in UFO's.
The gpawn is that its okay to 'take the square root' of the number of chess positions to be solved in the task.  In other words to just arbitrarily reduce that number to a millionth of a trillionth of its actual value.
Saying that's invalid (it is Grossly invalid) to a believer is like telling a doubter that water is wet.
The h-pawn is claims like that 1) a4 doesn't need to be considered and that its okay to just consider 4 moves at each point and that it doesn't matter that Stockfish assigns a win to some positions that are obvious draws.

Those pawns and the rook will never be moved nor taken - unlike in an actual game of chess.
The white Queen moves around illegally.  Scooping up any threat.
And is never taken.
Plus white can never mate black ...  too invalid.
So the verbal chess game goes on forever.  

#3418

"We agree on that, but that does not prove they their decisions are totally different and uncorrelated." ++ Well in TCEC they start from the same imposed position and they play it differently and sometimes it is win/draw, thus different moves in the same position and uncorrelated sometimes A wins and draws, sometimes B.

"But you exclude that their ignorance can lead to a chain of errors by the two, because they both ignore one made a mistake, but to you at least one of them knows how to play perfectly, even with a flawed evaluation"
++ If that were the case, then the evaluation functions would wildly fluctuate and they do not.

I don't even know if it does make sense to reply anymore.

Always. Every engine plays a different move... If there are 34 legal moves in that position, the probability for an engine to play any move m is 1/34...

How many some times? It's up to us to discover.

Of course. If two engines play two different moves in one position, all the moves are in general uncorrelated...

Inaccurate, imprecise, hasty generalization.

For reasons known to @tygxc only.

You are right, but it's always annoying to leave the last word to someone who pretends to be scientific, especially when they claim miracles can be done: it's only a matter of money...

#3421

"How many some times?" ++ You can see for yourself. When I link to an article you demand page and line instead of using CTRL F yourself. When I point to a tournament you want all statistics handed to you on a golden platter. Do it yourself then you cannot accuse me of doing it the wrong way.

"If two engines play two different moves in one position, all the moves are in general uncorrelated. ++ The correct moves are more or less correlated, the errors not.

It would be certainly better, instead, if one, following your example, said that he read the articles and found written black and white the exact opposite of what you wrote, without revealing exactly where; or that he made statistics on the data you indicate and they prove you wrong, but he will not provide neither of them. 😏

haiaku has been magnificently patient and attentive ...
constantly and systematically refuting and debunking 'the guy's' constant pseudoscience and presenting real science instead.
Perhaps MARattigan has run out of patience.
How much investment to make?
Elroch - slightly different.  

Its like talking to a flat earth person.
The more you refute - the more they like it.
They will see that as attention and push even more pseudoscience and represent themselves as equal or superior.
Is the pseudoscience here 'better dressed up' than flatearthology?
Are the tactics better calculated?  Yes.
There are internet articles about how to talk to a flat earth person.
But flatearth people are heavily invested ...  not much of a life and they need to get a life.

But here its 'Hey - you invest then nodes/squareroot guy will invest back at you !"

Let me emphasise by elaboration how absurd @tygxc's reasoning from a random sample of 2 is.

Suppose there is a position and a large set of engines. Suppose this is a really difficult position but half of the engines can be said to get it right (let's say they give a high evaluation to white - we'll gloss over the fact that evaluations on a real number scale aren't really compatible with the notion of being "right"). Let's also assume the other half of the engines can be said to get it wrong (say they evaluate it near zero and think it is balanced).

Now let's take two of these engines. If they both say it is a win for white, we have not been misled.  If one says it is a win for white and the other says it is a draw, according to @tygx we can be sure that a single game played from this point will have no crucial errors and the result will prove conclusively the right engine was indeed right. This is of course nonsense. It could easily make a mistake later and the game end up as a draw. According to @tygx, this would prove the second engine was right. Then the third case is that both engines wrongly think the position is a draw. Of course the result may be anything from there, but it is certainly possible it will be a draw. According to @tygx, this will prove both engines were right. Both were actually wrong (according to a hypothetical giant tablebase).

Yes but it must be his way otherwise the whole House of Cards collapses !
The discussion itself is more like a chess game than talking to a flatearther.

Chessplayer dynamics:  'Hey if you've got Colossal spare time to invest in this board game pasttime - then I do too.  And I'll invest more and I'll beat you !'
The contest isn't to see who can be 'Snarkier' (love that word) -
its to see who can invest more time ...
a bit like those drinking contests in bars you see in movies.

#3428

"a random sample of 2" ++ The sample is not random, it are the 2 best engines from the preliminary competition, it is called superfinals for a reason.

"evaluations on a real number scale aren't really compatible with the notion of being "right""
++ Kaufmann gave a translation between the provisional, subjective evaluation and the absolute, objective evaluation:
between -0.70 and +0.70: draw,
> +0.70 white wins,
< -0.70 black wins. 
This translation allows to compare the right and wrong between the provisional, subjective evaluation during the game and the objective, absolute evaluation at the end of the game.

"If one says it is a win for white and the other says it is a draw, according to @tygx we can be sure that a single game played from this point will have no crucial errors and the result will prove conclusively the right engine was indeed right."
++ Yes, the level of these top engines is around 3600. They make an error once in a while, otherwise the intended result of 1.5 - 0.5 would not occur at all. The probablility of both 3600 top engines making several errors in one game is negligible indeed: it is lower than once in the 100 games they play.

"It could easily make a mistake later and the game end up as a draw."
++ Yes, that is thinkable but unlikely as it would require both 3600 top engines to err in the same game. The probability of one 3600 top engine erring is low, the probability of both independent 3600 top engines erring in the same game is lower than once in 100 games and the TCEC superfinals counts 100 games.

"Then the third case is that both engines wrongly think the position is a draw. Of course the result may be anything from there, but it is certainly possible it will be a draw." According to @tygx, this will prove both engines were right. Both were actually wrong (according to a hypothetical giant tablebase)."
++ Yes, that is thinkable, but unlikely as it would require both 3600 top engines to err in the same game. The probability of one 3600 top engine erring is low, the probability of both independent 3600 top engines erring in the same game is lower than once in 100 games and the TCEC superfinals counts 100 games.

Point is, if they would not impose 50 slightly unbalanced openings in the TCEC superfinals, then there would be only draws: 50 - 50 and there would be no competition, just TCE, not TCEC.

Because of his insistencies on nodes instead of the true hardware speed of the computers - 
there's isn't the slightest obligation to read posts based on 'nodes' nor to even consider such notions.
Nor to read posts nor consider arguments based on taking the square root of the number of positions in the task.
Nor to read nor consider posts based on that only four candidates per move to be considered is pretended to be valid.
Nor to read posts based on a premise that the computers are unable to assign a draw to obvious draws is OK.
Any or all such posts can be dismissed.  Not read.
Regarding the refuting posts - better

The truth is signficantly different: it is that decisive games would be more rare. This would be far less so if you replaced one engine by one from a few years different time: there has been a signficant increase in rating of top engines over recent years (most notably Stockfish, with the most dramatic change being the introduction of NNUE).

#3432
"You don't understand what random means"
++ I understand that better than you. When you pick the 2 best engines then it is not random.

"seem to forget that the 2 best engines at one point in time have always become obsolete has-beens at some later time." ++ I do not forget that. Engines get better and make fewer errors. That is why the 50 imposed openings are carefully selected to get 1.5 - 0.5 results.

"the general idea that more material increases chances of winning." ++ Right: 1 pawn wins.

"0.5 does not mean certain draw: rather it means that the probability of winning is equal for each side"
++ This is nonsense. Game theoretically each position is either a draw, a win, or a loss.
That is a property of the position, regardless of the opponent.
The probablility of one side making an error is a property of the opponent, largely regardless of the position.
One side can only win if the other side makes a mistake.
Statistics of the previous superfinals show that a Leela book exit below +0.30 is an almost 100% certain draw.

"what a chess player would regard as a small advantage."
++ A small advantage does not exist. A position is either a draw, a win, or a loss.
If an advantage is large enough to win, then it is not small.
When an advantage is not enough to win, then it is no advantage at all.

"You have a poor grasp on the notion of certainty." ++ I have a good grasp of that.

"there has been a signficant increase in rating of top engines over recent years"
++ Yes, engines make fewer errors now.
They have to be more careful with the imposed openings now to get 1.5 - 0.5.

It has become increasingly like a discussion with a Flearther.

It's not required to oppose every single post, only to consistently debunk the garbage when it becomes the only voice .

Tygxc tosses his opinions in on every post that seems to have the flimsiest opening to spout his faulty premise.  

We have to get used to this, though, the public forums have devolved into a rabble of kids, a significant number of crackpots, and a smaller number of cogent posters that try to hold the line.  For every BlueEmu that posts less, there's a Tygxc that is posting more.  

#3434

"this is also true of positions that are unclear"
++ You cannot evaluate as unclear. Look deeper then.

"equal chances" ++ There aren no equal chances. A position is a draw, a win, or a loss. Maybe you cannot decide between a draw and a win.

"Imperfect means that seeing equal chances can be utterly wrong."
++ No, the provisional subjective evaluation is sometimes wrong as compared to the objective absolute evaluation at the end of the game and that imperfection of the provisional, subjective evelauation often leads to an error.

"All good chess players know that practical chess is about posing difficulties for the opponent so they are more likely to make a mistake."
++ Yes, practical chess is in part about knowing the opponent. The likelihood of making a mistake is a property of the opponent, not of the position. Against a weak endgame player, simplify to an endgame. Against a poor defender, attack. Against a poor attacker, defend.

"A small advantage exists in practical chess." ++ No, it does not. It is one of the myths of the previous century that you can accumulate small advantages.

"no evaluations exist except +infinity, 0 and -infinity." ++ Yes, that is correct. The problem is that the engine must decide on its move based on its evaluation. That is why the provisional, subjective evaluation has more than 3 states so as to rank moves and decide on one.

"It is highly likely that most of the imposed openings have a game theoretic value of a draw." ++ According to the selection some are drawn, some are white wins, some are black wins. Openings that lead to two draws are drawn. Openings that led to two white wins, or to two black wins are won / lost. The interesting openings are those that lead to the desired 1.5 - 0.5 result: are these a win or a draw?

#3436
"For every BlueEmu that posts less, there's a Tygxc that is posting more."
++ And for 1 GM, 65+ World Champion, famous analyst, MSc. with a profound opinion there are 7 weak players that call him a crackpot speaking garbage.