Chess will never be solved, here's why

H

@6636

"Are you saying you can square root the positions because we're assuming we can ignore all non-optimal play by black?" ++ We can even discard optimal play by black. Suppose both 1 e4 e5 and 1 e4 c5 draw. To weakly solve chess it is possible to look only at 1 e4 e5 and discard 1 e4 c5.

"How do you discard non-optimal moves without analyzing them?"
++ By the end result. If all lines end in a draw 7 men-table base endgame position or a prior 3-fold repetition, then that retroactively validates all black moves as optimal.

     I believe there is an old aphorism concerning the color of pots and kettles that might apply here.

@6635

"What weak chess players call "relevant" has no place in weakly solving chess, as indicated by it having no place in the academic literature."
++ Check Schaeffer's solution of checkers: only 10^14 positions relevant of the 5*10^20 legal.
Only 19 of the 300 openings relevant.

@6640

"No matter how eloquently you argue for the 10^17 reduction, after realizing this you have to go back and figure out how it's wrong."
++ 10^17 is not wrong. Chess is just not as deep and as wide as some people seem to think.
However 10^17 is still a huge number.
3 powerful computers working 24/7 during 5 years is huge.

     It might seem that the fact you immediately replied with more of the same might indicate that it does apply. I certainly never believe that some posters are capable of learning much.

     Just pointing out that someone who insists on misinterpreting another poster's use of the term "car", going off on on long sidetracks on that point, and then complaining that someone else noticing said mistake is reacting inappropriately is in fact acting inappropriately.

     It's likely most people here don't care for or pay much attention to those who assume the mantle of arbiter of acceptable posting.

Let's say there are two chess-playing super computers that have equal processing power. Neither has an advantage over the other in CPU speed or in any other parameter by which computers can be compared. They are also running the identical chess engine. What could possibly be the explanation for one computer defeating the other? Is that even possible?

There are plenty of examples of Leela vs Stockfish, AlphaZero vs Komodo, Fritz vs Leela, etc... But I have never seen Stockfish vs Stockfish on identical computing platforms. 

As has been pointed out in the past, engines that are "identical" in software and hardware configuration parameters still run on different hardware under different software instances, and variances will occur.

If you have ever overclocked a CPU, then you will know that CPUs that are supposed to be identical...are not.  They are built to fall within tolerances.  Software running on an OS is subject to variances in resource sharing and interrupts, etc.  The programming *is* the same for identical releases, for the record.

Much of the bad information on this thread comes from people who seem to know diddly and squat about computers, Tygxc included.

IM Levy Rozman did an interesting experiment that comes close to what I was talking about above (Stockfish vs Stockfish) but he forced a particular opening sequence before allowing them to play. He thinks that if he hadn't done that, that the computers would simply play a Ruy Lopez to a draw every time. But, would they? I would find it much more interesting if the engines were allowed to calculate from move #1. And, even if most games resulted in a draw, what could possibly be the explanation for any game ending *not* in draw?
https://www.youtube.com/watch?v=Vq-iWlbqX-0

Cause every hand a loser, and every hands a winner, and the best that you can hope for is to die in yer sleep..

https://youtu.be/7hx4gdlfamo 

@6559

"What could possibly be the explanation for one computer defeating the other?"
++ One making a mistake. It happens, though rarely.
That is why TCEC imposes slightly unbalanced openings to avoid all draws.
Assume a computer calculates 20 ply deep and plays against itself.
Assume there is a tactic that lies 21 ply deep.
So the computer misses it and plays its move. Now the tactic lies 20 ply deep and it finds it.

"I have never seen Stockfish vs Stockfish on identical computing platforms."
++ There is AlphaZero vs. AlphaZero
See Figure 2 https://arxiv.org/pdf/2009.04374.pdf 
At 1 s/move: 88.2% draws
At 1 min/move: 97.9% draws

Another thing that annoys me is the quote @tygxc uses is 15 years old. If 2007 computers could solve chess in 5 years, then current computers could do it in as little as 1 month... and top GMs hire other GMs all the time. Carlsen had more than 5 on his team for the last world chess championship match IIRC. The claim is, essentially, that they had the opportunity to solve chess in 1 month.

Supercomputers are about a million times faster than a top end PC. So, a few seconds should do.

Oh, did Sveshnikov claim he could do it with standard computers? Looking up the quote he merely says "modern computers."

In any case modern supercomputers are at least 60x faster now, so 5 years -> 1 month.

@6665

"You also need to deal with lines like 1. e4 Nc6 etc... " ++ White tries to win, black tries to draw. Once black succeeds in all lines and white fails, chess is weakly solved.
On the first move white has 20 legal moves and black has 20 legal answers, that makes 20*20 = 400 legal positions after move 1. However, to weakly solve chess we only need a strategy, i.e. one strategy for black to draw, i.e. for all 20 white moves 1 black response that draws. Thus a priori only 20 = Sqrt (400) relevant positions after move 1 instead of 400.

We can use game knowledge to further prune this.
From no other input but the Laws of Chess and many boolean operations AlphaZero ranked the 20 legal first moves for white:
d4 > e4 > Nf3 > c4 > e3 > g3 > Nc3 > c3 > b3 > a3 >
h3 > d3 > a4 > f4 > b4 > Nh3 > h4 > Na3 > f3 > g4
If the 4 best moves cannot win for white, then the 16 worst moves cannot win either.
So instead of 400 legal positions or 20 relevant positions after move 1, this can be pruned to e.g.

• 1 e4 e5
• 1 d4 d5
• 1 c4 e5
• 1 Nf3 d5

How does that fit into your summary "If all lines end in a draw 7 men-table base endgame position or a prior 3-fold repetition, then that retroactively validates all black moves as optimal."

++ White tries to win, black tries to draw. If all lines end in a 7-men endgale table base draw, or in a prior 3-fold repetition, then in retrospect all black moves were good: fit to draw.

@6666

"If 2007 computers could solve chess in 5 years, then current computers could do it in as little as 1 month" ++ Maybe the 2007 statement was visionary. I only calculated with present computers.  It took from 1961 to 1969 to land humans on the Moon. By the same logic that should only take one month now.

"top GMs hire other GMs all the time. Carlsen had more than 5 on his team for the last world chess championship match IIRC" ++ Carlsen, Caruana, Nepomniachtchi probably solved part of it during their months of preparation with grandmasters and cloud engines.

@6668

"did Sveshnikov claim he could do it with standard computers?" ++ Yes

"he merely says modern computers." ++ He did not say how many, but he used plural.

"modern supercomputers are at least 60x faster now, so 5 years -> 1 month."
++ I do not know how many nodes/s the best engine in 2007 could calculate.
I can only state that 3 modern cloud engines of 10^9 nodes/s can in 5 years exhaust the 10^17 relevant positions.