Chess will never be solved, here's why

Yes, it would be incorrect to reject the possibility of say a high speed object from outside the solar system destroying the Earth before sunrise tomorrow. Such objects exist, for sure, and we don't have any knowledge of where smaller examples are. The probability of being hit by one in a day is thankfully small.

@11950
"Probability for those 112 games not being perfect is quite abit higher." 
++ That is not precise. What probability do you mean?

'What will the weather be tomorrow?' "The correct answer is don't know."
It is also the most useless.

A true mathematician would say: I have gathered the data of 112 weather stations, on my 2 servers solved the Navier-Stokes equations and arrived at:
clouded, but dry, 20°C, weak wind from northeast. That is useful to farmers, pilots, tourists...

@11952

"You could have said the same about engine of 10 years ago"
++ 10 years ago the ICCF WC finals had 35 decisive games, now none.

"And we KNOW Stockfish is far from perfect." ++ At blitz speed.

"It sometimes makes 6 blunders in a row in tablebase positions."
++ In irrelevant positions and at blitz speed.

"What do you think it does in really difficult positions with many more pieces?"
++ With 5 days per move it can hold the draw.

"Statistics are uncertain"
++ Tromp used statistics to arrive at (4.82 +- 0.03) * 10^44 legal chess positions.

@11956

Weather forecaster: "The data suggests it will rain" 
General Eisenhower: "So it will rain?"
Weather forecaster:
"I don't know whether it will rain, but based on that data it's more likely than not."
General Eisenhower: "Not good enough. I need to know to proceed with D-day or not. Thousands of lives depend on it. Be more precise!"

"a perfect game is a draw" ++ Yes.

"Everyone already understands this" ++ There are still a few who believe otherwise.

"confuse top level ICCF games for perfect games simply because they're high level draws"
++ Not because they are high level draws, but because all 112 out of 112 of the strongest chess on the planet (5 days/move, ICCF WC finalist + 2 servers 90*10^6 positions/s) now end in draws.

"what differentiated imperfect ICCF draws from perfectly played ICCF draws"
++ We cannot know by inspecting the games: the most skilled inspectors: ICCF WC finalists with twin servers 90*10^6 positions/s at average 5 days/move could not find any imperfection.

If you want to talk probability, then assume game 113 is decisive, and a clean win, no clerical error, or hospitalised player. That would leave a probability of 1/113 for a single error, or a probability of (1/113)² = 8*10^-5 of a double error, or a probability of 99.92% for the 112 draws to be perfect games with optimal play from both sides.

"A true mathematician would say: I have gathered the data of 112 weather stations, on my 2 servers solved the Navier-Stokes equations and arrived at:
clouded, but dry, 20°C, weak wind from northeast. That is useful to farmers, pilots, tourists"

thats why a mathematician isnt a weather scientist. a mathematician deals with perfect deduction, not estimates.

tygxc, ive had my statements on mathematical rigor literally verified BY mathematicians.

they all chewed me out for wasting my time as someone as illogical as you.

why arent you addressing this fact?

"If you want to talk probability, then assume game 113 is decisive, and a clean win, no clerical error, or hospitalised player. That would leave a probability of 1/113 for a single error, or a probability of (1/113)² = 8*10^-5 of a double error, or a probability of 99.92% for the 112 draws to be perfect games with optimal play from both sides."

if the 112 draws are proof then its proof regardless of any other game outcome. thats what basic deduction means.

'"Statistics are uncertain"
++ Tromp used statistics to arrive at (4.82 +- 0.03) * 10^44 legal chess positions."

uhhh did you notice how tromp said it was an estimate with a probability of being wrong?

in fact the ± quantity is literally based on the standard deviation of his wrongness.

also dont think that everyone hasnt noticed how you failed to address even a single of llama's posts.

lets put this into perspective here:

tygxc, in response to someone asking for mathematical rigor, responded with the claim that a mathematician working at a weather station wouldnt deal in absolutes and would just give the estimate of the weather.

he really thinks that math is just a bunch of guesses that we think are accurate. LMFAO

Correct me if I'm wrong, but having all draws in the ICCF WC isn't a solution to chess.

A solution to chess would be knowing the exact number of moves (and the precise sequence of moves) to a win/loss/draw from any given position, no?

IOW: a 32-man tablebase ...

I don't buy these probability calculations at all. Once again you assume all errors are similar in likelihood. In reality, errors are not similar at all and they're not independant events either, meaning first error occurring could lead to the second error being more likely than the first one.

After an error there can be a 100 winning lines or there can be 1. In the second case another error is far more likely to follow. It would also make sense logically, that in the starting position there are more drawing lines, than there would be winning lines after a subtle error by a strong engine. This makes it way more difficult to estimate how many errors are there actually in these games or whats the likelyhood of these games being perfect.

You're definitely not wrong.

I did the same thing stockfish drew itself every game

And the fact that it doesn't think ahead in endgames and doesn't understand when opposite colored bishop endgames are winning sometimes

When the dev team wants to make new release, how do they test it?

- Does the new engine beat the old engine over 50% of the time?

- Does the new engine beat other popular engines more often than the old engine did?

- Does the new engine draw itself consistently?

These are the characteristics that produce a TCEC winner. TCEC winning engines and their handful of close competitors survive, all other engines go on the scrap heap.

The very process of "evolution" for computer engines ensures that they are tuned heavily towards their own current playing ability, with no regard to the future, only the past, and certainly no thought towards perfect play or solving chess. They are designed to win most often against other engines, but more importantly to not lose against other engines. An engine that does not draw itself the vast majority of the time is an engine that is vulnerable to losses.

Now, add the other very important factor here:

Stockfish is the king of the heap, but is also open source. Stockfish's competitors can always steal from it in their own development (indirectly if not openly lifting code line for line if the other engine is a commercial one), and if Stockfish does not draw itself at at even higher rate than most engines, then the opposing dev teams can figure out why, and improve their own engines to exploit blind spots in Stockfish's playing. The *only* advantage that Stockfish maintains is the privacy of their current beta version whose source has not been released yet. So, each release *must* change in some way, even if mostly a lateral change in playing ability, to stay ahead of the open source issue.

This creates a neverending cycle where engines play only the same handful of opponents and are tuned to (1) never lose, first priority, and (2) to have as big a shot at winning as they can manage without compromising (1).

So, is it a surprise that engines draw more and more frequently? Is it a surprise that they draw amongst each other? Not at all. Engines have evolved into an incestuous little band of siblings, and this affects their possibilities of approaching "perfect play" at lot more than people think.

Consider the following scenario:

Take 6 top players...Carlsen, Caruana, Nakamura, Nepo, Gukesh, Pragg. Now remove them from FIDE and have them play each other, and effectively only each other, for 20 years. Give them a cutthroat win dynamic...top player gets rich, 2 gets a living wage, 3/4/5 get bread and water, 6 lives in solitary confinement. They have a powerful incentive to learn to claw their way higher by any means, and learn to play each other very, very, well.

Now...what is your confidence level that they remain the top players in the world upon re-integration with FIDE at the end of the experiment? What is your confidence level that their play does not ultimately suffer from the severely reduced playing pool?

This is where the top engines are right now.

I mean assume we had a table base for all 32 pieces and assume its a draw considering the new eval would be draw or win in any given position with no numbers outsdie forced mate isn't a perfect game any game where it never swings to a loss/win? That in mind they could be playing perfect games already in slow time controls with good hardware. In fact I would imagine that's the most likely . So until we get future tech and have that 32 piece table base its not proven but I would assume many perfect games have been played as the best guess have now is it starts a draw and you need to make mistakes to swap from draw to lose and technically anything inside those bounds is perfect. Can make an engine as strong as you want still only 20 moves turn 1 and I personally cant see a future where in 3024 some guy finished the table bases as goes aha sorry the correct move was 1e4 white has forced mate in 408 moves sorry d4 was always a draw.

Ty's smart. very smart. theyre looking at stuff in practice. like 2. Ba6. i mean come on...4real ?? whos dum enuf to say this isnt a completely lost position (w/out compensation !) from move 2 ??...but then it must be some ppl here needta do a delineation. and i say fine. if it makes ur proof gawz happy. but being dn a full bishop THAT early is NOT like saying a lottery tix wont win.

so quit wimping on Ty cuz hes taking the practical side. hes making excellent points that are most likely gonna stand any test. they just needta be confirmed w/a final silicon go-thru. AND U KNOW IT !

Interesting fact, of the n-piece tablebases, the 27 and 28 piece ones are the biggest (source: Peter Österlund). I am not sure which of the two is the absolute biggest.

Guessing is easy, and everyone is free to do it. @tygxc just uses all the wrong terms for his guesses.