https://www.chess.com/fr/forum/view/general/chess-simple
Chess will never be solved, here's why
Sort:
Jun 1, 2022
0
#2623
'Somebody' wants the 'small fraction' to be a trillionth of a trillionth of the number of positions in the task.
The number of possible positions in chess versus the number of possible games ... well they're both enormous numbers.
But the second ...
try an analogy.
The number of possible chess positions could be something like the number of atoms within the planet Jupiter. The largest planet.
Close to it? Not the point.
The point is its a very very big number - but its finite.
But the number of possible games ?
That could be more like the number of atoms in the universe - whether finite or infinite universe.
Result: the number of possible positions is easier to work with.
And in the endgame tablebases - its even Known for small numbers of pieces !!
Issue: But for 'solving' positions - you can't completely get away from both 'games' and 'elements of games'.
In other words sequences of moves.
Can't be avoided.
That's the central issue.
Getting around the daunting 'game variations' - but in a way that's both mathematically and scientifically valid.
'Scientifically valid' for the computers. Practical.
And for now - there's no way to do that.
And no money from smart people thrown away on such a thing.
But there are related projects though.
Lots of them. One of them ... 'Stockfish'
"By publishing a monograph on the 5...e5 system in 1988, I practically exhausted this variation."
- Sveshnikov
If Sveshnikov did that on his own in 1988 without engines or table bases,
then it is plausible to weakly solve chess in 5 years with modern computers and good assistants.
Jun 2, 2022
0
#2625
Two HUGE errors. Sorry, three.
- It was only ever an assertion in the first place, nothing more than a PRACTICAL opinion.
- Sveshnikov did not claim to have even exhausted one sharp opening. And he was of course right to say he hadn't exhausted it. But the degree to which he failed to do so is important.
His reasoning is the practical chess player's reasoning that if you try a tiny sample of plausible lines, you will generally get a good evaluation. Now we understand that chess engines that examine a billion lines and AI engines that examine a million lines are strictly better - i.e. it quite often happens that a human evaluation is wrong and the silicon evaluation is right. This is so often so that a top silicon player to able to defeat any human almost all the time. - His claim was unquestionably wrong if taken as meaning that he was close to fully analysing the opening. There are certainly vast branches he didn't look at but could only guess about. Much bigger than what he did look at.
On the plus side, you said "plausible", which means you realise you could be wrong, which is entirely reasonable.
#3205
He said:
"Since that time only some details have been developed, without introducing anything particularly new: the evaluations of the main lines have hardly changed." - Sveshnikov
In the Anand - Gelfand match white got nothing with the main line 7 Bg5.
https://www.chessgames.com/perl/chessgame?gid=1666526
In the Caruana - Carlsen match white first avoided the Sveshnikov B33 with 3 Bb5, got nothing and then played the 7 Nd5 line and got nothing either.
https://www.chessgames.com/perl/chessgame?gid=1937913
So it seems that the teams of grandmasters and engines just agreed with Sveshnikov.
So the silicon evaluation just confirmed the human evaluation.
#3207
That is how it works
Here is an ICCF game from the last WC.
It is 99% sure to be a perfect game without any error.
If tic tac toe can be solved, then chess can too!
lmao I'm maybe just kidding
tygxc wrote:
#3207
That is how it works
Here is an ICCF game from le tast WC.
It is 99% sure to be a perfect game without any error.
...according to some imperfect engines that are still incrementally improving each release...
Jun 2, 2022
0
#2632
People used to think Deeper Blue was pretty accurate. It would lose most games to a current top engine.
Jun 2, 2022
0
#2633
tygxc wrote:
"By publishing a monograph on the 5...e5 system in 1988, I practically exhausted this variation."
- Sveshnikov
If Sveshnikov did that on his own in 1988 without engines or table bases,
then it is plausible to weakly solve chess in 5 years with modern computers and good assistants.
We must all agree that. If the antecedent is false then the implication is true.
Jun 2, 2022
0
#2634
Elroch wrote:
People used to think Deeper Blue was pretty accurate. It would lose most games to a current top engine.
This is a good point - but even on that - 'current top engine' (one of them) assigns wins to obviously drawn positions.
Stockfish.
Astutely posted by MARattigan.
But these realities are lost on those rejecting such reality.
Its similiar to flat earthism.
#3211
It is 99% sure to be a perfect game.
I calculated before
17 ICCF (grand)masters with engines, round robin, 5 days / move
136 games
126 drawn games with 0 errors: ideal games with optimal moves i.e. perfect play
9 decisive games with 1 error
1 drawn game with 2 errors
That is the only consistent way to explain the observed data.
Bear in mind that the tournament lasts about 2 years and that the 17 ICCF (grand)masters use one or more multicore engines each, so the result of 136 games represents a century of engine calculation guided by human ICCF (grand)masters.
#3193
"I just think that other explanations are more plausible, but equally not sufficient for a proof."
++ No, you cannot explain that.
AlphaZero autoplay data:
1 s/move: 88.2% draw, 7.72% white wins, 4.09% black wins (a data set of 10,000 games)
1 min/move: 97.9% draw, 1.8% white wins, 0.3% black wins (a data set of 1,000 games)
Assume for argument's sake unlimited time and unlimited engines available.
If the time per move keeps increasing, the draw rate keeps increasing too, to near 100%.
However, if the time per move keeps increasing more, it comes at some finite but huge time/move at a point where it exhausts all finite number of chess positions.
Now if chess were a white win, then the draw rate would keep getting closer and closer to 100% with increasing time per move and then suddenly plummet to 0%.
That is absurd.
This proves that chess being a white win is inconsistent with the observed data.
Likewise for the hypothesis chess being a black win.
Remains: chess is a draw.
#3205
"It was only ever an assertion in the first place, nothing more than a PRACTICAL opinion."
++ No, it was his theoretical opinion. This is not about practical chances in practical play, it is about the theoretical status of the opening.
"And he was of course right to say he hadn't exhausted it." ++ No, he said he exhausted it.
"His reasoning is the practical chess player's reasoning that if you try a tiny sample of plausible lines, you will generally get a good evaluation." ++ Sveshnikov looked at all the important lines.
"Now we understand that chess engines that examine a billion lines and AI engines that examine a million lines are strictly better" ++ Looking at unimportant lines brings nothing.
"it quite often happens that a human evaluation is wrong and the silicon evaluation is right."
++ Yes, but Carlsen and Caruana rented cloud engines during the months to prepare their world championship match. Carlsen confidently played for the Sveshnikov B33 in all games. Caruana first avoided it altogether with 3 Bb5 and then avoided the main 7 Bg5 line. So we can conclude that both Carlsen and Caruana and their teams of grandmasters and cloud engines arrived at the same conclusion as Sveshnikov in 1988: the Sveshnikov B33 draws.
"This is so often so that a top silicon player to able to defeat any human almost all the time."
++ Humans get tired, humans get distracted, humans get nervous in time trouble. ICCF grandmasters fall ill.
tygxc wrote:
It is 99% sure to be a perfect game. I calculated, before, 17 ICCF (grand)masters with engines, round robin, 5 days / move, 136 games: 126 drawn games with 0 errors: ideal games with optimal moves i.e. perfect play [ . . . ]
As usual, you just repeat yourself. Objections have been made to those calculations and their basis: the assumption that the game value is a draw (I sligthly edited your post to make it shorter).
That is the only consistent way to explain the observed data.
To you.
Now if chess were a white win, then the draw rate would keep getting closer and closer to 100% with increasing time per move and then suddenly plummet to 0%. That is absurd.
To you again. The evaluations become usually more stable with depth, but how many times we see an engine change its evaluation dramatically even after a wide and deep search?
"It was only ever an assertion in the first place, nothing more than a PRACTICAL opinion."
++ No, it was his theoretical opinion. This is not about practical chances in practical play, it is about the theoretical status of the opening.
"By publishing a monograph on the 5...e5 system in 1988, I practically exhausted this variation." - Sveshnikov
"His reasoning is the practical chess player's reasoning that if you try a tiny sample of plausible lines, you will generally get a good evaluation." ++ Sveshnikov looked at all the important lines.
The idea is to find a weak solution to ascertain which are the important lines, not to use opinions about which are the important lines to get a non-mathematical solution.
#3218
"Objections have been made"
++ People make all kinds of objections. Try to find yourself an explanation: tell how many of the 127 draws, 6 white wins and 3 black wins contain how many errors under whatever of the 3 hypotheses: draw, white win, black win. It does not even have to be right, if it is plausible then it is OK. I claim you cannot get a plausible explanation aside from chess is a draw, 126 perfect games 1 draw with 2 errors, 9 decisive games with 1 error.
That is just statistics and probability.
Moreover, I can pinpoint the errors in the 9 decisive games.
"That is the only consistent way to explain the observed data.
To you." ++ If not to you, then try to explain yourself: give the numbers of 0, 1, 2, 3, 4... errors for the 136 games under whatever of the 3 hypotheses. If you claim you have an alternative consistent way, then please show it. Facts and figures, not opinion.
Now if chess were a white win, then the draw rate would keep getting closer and closer to 100% with increasing time per move and then suddenly plummet to 0%. That is absurd.
To you again. The evaluations become usually more stable with depth,
"but how many times we see an engine change its evaluation dramatically even after a wide and deep search?" ++ It is not about an evaluation number changeing, it is about the outcome of thousands of autoplay games. To me it is absurd that the number of drawn games would steadily creep to 100% and then magically drop to 0%. To you that sounds normal?
"The idea is to find a weak solution to ascertain which are the important lines, not to use opinions about which are the important lines to get a non-mathematical solution."
++ No, it is allowed to use chess knowledge in the brute force method. Especially chess knowledge in the 3 AlphaZero papers is logically derived from nothing but the Laws of Chess, axioms in your lingo, so it is unbiased knowledge that follows from the Laws of Chess, not some opinion. Sveshnikov by his own account almost weakly solved his Sveshnikov Variation B33 in 1988 alone and without engines or table bases. So much more is possible now with engines and table bases. The teams of grandmasters and engines of Carlsen and Caruana confirmed this: black confidently went for the Sveshnikov and white avoided the main line, so both tacitly agreed that their preparation had come to the same conclusion: B33 draws.
If the good lines cannot win, then the bad lines cannot win either.
It is certain that 1 e4 e5 2 Ba6 does not win for white, it is known that it loses. No need to calculate.
1 e4 most probably draws and 1 d4 most probably draws as well, but 1 a4 is absolutely certain not to win.
tygxc wrote:
"Objections have been made"
++ People make all kinds of objections. Try to find yourself an explanation: tell how many of the 127 draws, 6 white wins and 3 black wins contain how many errors under whatever of the 3 hypotheses: draw, white win, black win. It does not even have to be right, if it is plausible then it is OK.
Already answered: post (1), post (2).
I claim you cannot get a plausible explanation aside from chess is a draw, 126 perfect games 1 draw with 2 errors, 9 decisive games with 1 error.
And I claim otherwise: post(1).
That is just statistics and probability.
Already answered: post(1), post(3).
"That is the only consistent way to explain the observed data.
To you." ++ If not to you, then try to explain yourself [ . . . ]
Already answered: post(2).
To me it is absurd that the number of drawn games would steadily creep to 100% and then magically drop to 0%. To you that sounds normal?
Already answered: post(1). No magic involved.
"The idea is to find a weak solution to ascertain which are the important lines, not to use opinions about which are the important lines to get a non-mathematical solution."
++ No, it is allowed to use chess knowledge in the brute force method. Especially chess knowledge in the 3 AlphaZero papers is logically derived [ . . . ]
Already answered: post(4), post(5), post(6), post(7).
Object the objections, instead of restating the things which raised those objections.
"I think it's almost definite that the game is a draw theoretically" Fischer
"We do not know for a fact that the starting position is a draw, but it does seem like a safe assumption - Rowson 2005.
Hi, I don't think that's "reasonable doubt". Both are saying they think it's drawn. It seems as though they're just coveing themselves; not wishing to be outspoken. It isn't evidence that they really think it may be a win.
Hi. Possible, but I think they at least think that the evidence is not enough for a scientific proof: would they be so cautious, otherwise?
Apart from that, as I said earlier the modern view on science is that the only real proof is a mathematical proof.
Disagree. But who am I to disagree with anyone? Just that mathematics is used as a depiction of observed results and there can be inaccuracies in the observations and also in methodology shortcuts. Normally, things are described as accurate within limits or an error margin.
Agree, but that's another reason to say that, apart from mathematics, strictly speaking science cannot prove things.
But let's assume that we can consider something "scientifically proven" the old way, i.e. a theory is proven, if experiments consistently confirm it. Can we say that it is scientifically proven that the game-theoretic value of chess is a draw? Imo no, because the game-theoretic value is the best outcome that a player can force. If we don't know for sure that the value is forced, the experiments that we can run (games between more or less strong players) cannot confirm nor contradict the statement. The increasing draw rate in games between engines of the same strength, and in particular in autoplay, can be explained with the increasing stability of the evaluation functions, both because of the introduction of neural networks, and because the evaluations become (on average) more stable with depth.
I think you're just describing normal, experimental error. There's no reason to assume that a way is going to be found to eliminate it and when we describe something as "proven" it really can be via the pragmatic method.
There is a difference, though. We can scientifically prove which is the accuracy of the devices used to measure physical quantities, and we have a variety of statistical tools to analyze samples. So, roughly speaking, when we measure a phenomenon we can assume that the mean of the sampled data is the "real" value of the physical quantity we are measuring, within a confidence interval. In our case, I don't think we can say that the outcome we more often observe, is the best that actually can be forced.
For this reasons it becomes more and more difficult to overcome an opponent of the same strength, but that does not mean that the game value is a draw. I think that if the increasing drawing rate depended on the game-theoretic value, we should not see the playing strength and ratings of the top engines increase at the current rate.
I don't follow that so it must be a piece of inductive thinking. It doesn't seem safe or relevant.
Of course, it's not safe (in fact I said "I think")! I wrote that, because @tygxc claims his explanations for the observed data are the only possible ones. I just think that other explanations are more plausible, but equally not sufficient for a proof.
yes
simple!
i say "yes"!