Will computers ever solve chess?

Usually?  I find that hard to believe.  I think it may depend a lot upon how one defines "best"  I usually use it to mean "there is no better".

Regardless, I should think positions like that are oddities, not the norm.

Would you care to randomly sample 100 endgame positions, and see how many allow for 12 or more "best" moves?

I have shown several positions where there are over a dozen best moves.

So you admit that clear evidence is lacking,

No, actually, that is your supposition.  

but then still state with confidence that engines trained through self-play will soon transcend those of the conventional type... all while ignoring the following:

Note that a person can consider all kinds of factors in an argument and weigh them as they see fit.  It would be illogical to assume that because a person reaches conclusion X that they have therefore not considered any evidence that might not be in favor of X. 

• Potential for diminishing returns as engines start to encroach upon either a theoretical or practical limit to the standard of play that can be produced by essentially heuristic methods.  Yes, there is a potential.  I'm glad you are using the word that characterizes this best.  Why would I give this more weight than the potential in question dictates?
• The possibility that, after more than a century of fine-tuning (and then ~2 decades of further refinements in the computer era), human valuations just might be a little less flawed than you seem to think.  Well, I have to weigh this in the face of what A0 achieved (severely flawed though it was) in 4 manhours of "training" vs. your century of fine tuning, and with the clear indicator of what happened afterwards, as well.  It sure seems like one of the main Stockfish developers immediately went to work on Leela, while Komodo rushed to build Monte Carlo.  Maybe it's just me, but I am guessing that these guys are pretty good at engine development, and that their enthusiasm for going in this new direction comes from somewhere knowledge-based (no pun/humor intended).  
• Of course, any form of static evaluation is inherently limited in how close it can get to true perfection, but apparently you just assume (without evidence) that decades of dedicated, intensive, and often collaborative study and research efforts by the best chess minds the world has ever had to offer has produced nothing but... useless trash, to be disposed of as quickly as possible.  I haven't actually said that anything is useless trash, but it does seem quite obvious that at some threshold, whether now or later on when they are 800-1200 ratings points past Carlsen or what have you, that engines will need to discard the assumptions inherent in human valuations.  In one of the A0 vs. Stockfish examples shown here (not positive if it was this thread offhand), A0 moves its king up the board despite queens being on the board to shore up a positional lockdown, apparently certain of its safety, in a position that traditional engines are tweaked to be overly cautious in.  It's just a small representative example (keep this in mind if your reaction is to dive into some detail that switches perspective from forest to trees).
• The fact that self-training methods must necessarily involve only a feasible number of games, which means that the number of positions examined during this process will fall short of the total state space complexity by about 30 orders of magnitude. So clearly, there is a ceiling for this approach that can only be raised by significant hardware developments. To illustrate, here is a diagram from https://arxiv.org/pdf/1712.01815.pdf:  I don't see that anyone is arguing about there not being a ceiling...only how high the ceiling will be.
• The subjectivity involved in deciding on what data should be collected during self-play, and how that data should be used. There are many choices to be made when designing the framework for a ML-based system, and the effect they have on the results can be quite dramatic.  Assuredly.  I don't see how this pertains to my ignoring anything, however.  Is your argument that since ML is not perfect, that its flaws will therefore be as great as the inherent flaws in human-derived valuations?  Given the history of such advances, this seems unlikely.

A forced win for White (or Black, for that matter) would not involve just a single line, or even just one narrow set of lines. It would require every single defense for one side to be refuted, which would mean that the best assessments that humanity can currently muster are wrong -- and not just wrong in one opening variation, but in dozens upon dozens of opening variations. Modern theory would have to be wrong about every single line ever analyzed I was about to take you to task for this over-reach, but then you did it yourself -->, in fact, at least within the confines of what Black (or White) can opt for against the hypothetically winning moves (which would necessarily have to include several broad categories of openings, at the very least).

A forced win for one side would also imply that existing theory has been regressing for quite some time now, at least in the "big picture" sense.  I'm not sure it implies that at all...but honestly, I don't see a whole lot of value in exploring the premise, since it seems designed not to prove anything that furthers understanding of the situation, but to merely paint my position with a brush of implausibility.

Most of the significant breakthroughs in the computer era have been of the equalizing variety.

Yes.  Well, they would pretty much have to be, wouldn't they?  In a well-balanced game, the entire point is to narrow the ease with which one person wins over another, because easy wins = boring game.  So, chess rules and play over history evolve towards a draw.  Not in dispute.  The dispute is whether human beings have managed to achieve a game that is a forced draw when the complexity of the game tree is beyond their ken/ability to contain.

In many lines where White was believed to have an advantage, defensive resources and/or counterplay opportunities are being found. Meanwhile, only the most dubious sidelines are even considered refuted, at the moment. As any serious correspondence player would tell you if asked, the paths to even a practical glimmer of an advantage appear to be getting slowly but surely closed off ...in a sphere of knowledge based on the constructs of traditional human-derived engine valuations.  What the new breed of engines will do to centaur chess is all but unexplored, so surely you would want to withhold judgment for a while? 

Usually?  I find that hard to believe.  I think it may depend a lot upon how one defines "best"  I usually use it to mean "there is no better".

WHY I AM CONVINCED CHESS IS A DRAW WHEN NEITHER SIDE MAKES AN ERROR

As already established, human chess players are no longer qualified (and really never were) to determine whether something is truly an "error" in chess or not, so this premise is flawed from the very beginning.  Every major tournament, commentators will show the very best human players how their play was determined to be in error.  You will note that none of them really dispute that engines are wrong save in a few small cases of known engine issues relating to their "horizon".  The point is, every reasonable chess player knows that they are not the arbiter of what constitutes an error...and by extension they are not the arbiters of "best play", either.

1. The grandmasters assume chess is a draw.

Ok (you should qualify this with "most", but anyway)

When they do their analysis from the opening position They know that White has a slight advantage

Ok

but they also know it is nowhere near enough to force a win.

No.  Prove to me that grandmasters do not assume a draw because it is simply expedient for prep/analysis to do so, given that their goal is to win a game/match/tournament, not to solve chess?  If you can't, then stop pretending you know what 95% of GMs think.

Value towards circumstantial proof:   low.

Often when playing other players equal to them--they will try for a win with White and try for a draw with Black.

Ok

But they know chess is a draw with best play.

No.  This does not follow, and it has also been established that most GMs "think" chess is a draw with best play, but that few if any have ever claimed to know.  Thus your foray back to 1890 to find some players that held that opinion.

Value towards circumstantial proof:   low. 

[the stronger the player from Class D to Class A to master to grandmaster--the more likely the player will know chess is a draw.]

No.  The more likely the player will think that chess is probably a draw. 

Value towards circumstantial proof:   negligible. 

Some have declared to me that I do not have proof that the top chess players assume the opening position is a draw. This is not correct. If you are a strong enough player all you need to do is to look at annotated games of grandmasters and supergrandmasters

You have yet to cite a single example of this in practice, and even if you did provide some examples, your ability to extrapolate this to 1600 GMs worldwide is beyond sketchy.

Value towards circumstantial proof:   negligible. 

and you will see this. Or you can listen to interviews of the top players and you will see they assume the opening position is a draw.

Cited examples would also be helpful here...particularly ones where the GMs are specifically addressing the definitive solving of chess, and not just making an offhand remark about the current tournament environment they play in. 

Value towards circumstantial proof:   negligible. 

I remember when checkers [also known as "draughts"] was assumed to be a draw by the top players. Some received a lot of vitriolic for their assumptions but were later proven correct when checkers was "solved".

It's "vitriol".  You've made this error several times now, so I want to let you know lest you continue to indirectly discredit your own arguments in the future. 

The number of positions in Chess is about 25-30 orders of magnitude larger than Checkers.  Your assertion is no better than saying we can make a car that goes 0-60 in <10 seconds, so therefore we can build an engine that will exceed the speed of light.  In fact, it's much worse, since the speed of light is "only" about 4-8 orders of magnitude faster than the fastest man-made vehicles currently.　

Value towards circumstantial proof:   none whatsoever. 

Here is a quote from cobra91 [on chess.com] which helps to sum up this point:

It's quite interesting how you have claimed for so long to have proof of your claims, yet suddenly you decide to put someone else's post front and center when you finally deign to say something about your proof. 

cobra91 replied:

[non-Ponz content removed for brevity]

2. From my 69 years of playing and studying chess, I know chess is a draw.

No, you don't.  You think it's a draw and you are only 99.9999% sure of it.  You can't keep going back and forth to whichever position benefits you in that particular post. 

Value towards circumstantial proof:   negative (discredits all arguments). 

Having played masters and grandmasters reinforces my view. Having studied the games of grandmasters and supergrandmasters reinforces my view.

You have played 4 GMs.  This reinforces your view exactly 4 GMs worth in the eyes of others.  Studying games reinforces your views only to yourself.

Value towards circumstantial proof:   low.　

Having authored chess books and reading chess books reinforces my view. Looking at chess on youtube reinforces my view.

As doing all these things for themselves also reinforces the views of your "opponents", who do not agree with you.

Value towards circumstantial proof:   negligible. 

Everything points one way--chess is a draw.

Except that it doesn't.  A lot of it points one way. 

Value towards circumstantial proof:   negligible.

3. Looking back at World Championship Matches for the past 100 years we can see a progression. There are more and more draws played by these top players. This is because the fewer mistakes there are--the more likely the game will end in a draw.

Value towards circumstantial proof:   reasonable. 

Over-the-board World Championship Matches: There has been a lot of controversy over what was a World Championship match and even sometimes--who was the World Champion? However it has become very clear from looking at matches and World Championship matches for the last 100 years that there have been more and more draws in these matches as players became stronger. This is also a very good indication that chess is a draw when neither side makes an error.

Value towards circumstantial proof:   redundant point, ergo none. 

4. ICCF Correspondence chess is about the strongest chess one can find. This is because there is a combination of a strong chess engine with a strong player.

With the strong player to guide the strong chess engine the play is stronger than a chess engine alone. Top over-the-board grandmasters and supergrandmasters use the games from ICCF Correspondence in their study of chess openings.

Looking at ICCF Correspondence Chess [it is now played with the help of chess engines]. We see a progression of more and more draws.

Value towards circumstantial proof:   reasonable. 

When I played in the 7h USA Correspondence Chess Championship Final round--I had only 1 draw in 14 games. [13 wins, 1 draw, and no losses] 96.5% [I did not use a chess engine]

Value towards circumstantial proof:   negative due to 1 draw out of 14 games.  Maybe keep the chest beating out of your argument when it hurts you. 

Now looking at a more recent winner of this event we find:

16th USA Championship winner had 5 wins, 1 loss, and 6 draws.

[8 points out of 12] 66.67%

Looking at ICCF Correspondence World Championships Finals we find:

13th World Championship Finals winner had 10 wins and 6 draws [13 out of 16] 81.25%

16th World Championship Finals winner had 8 wins and 8 draws [12 out of 16] 75%

22nd World Championship Finals winner had 7 wins and 9 draws [11 1/2 out of 16] 71.8%

25th World Championship Finals winner had 5 wins and 10 draws [10 out of 15] 66.67%

26th World Championship Finals winner had 5 wins and 11 draws [10 1/2 out of 16] 65.6%

27th World Championship Finals winner had 3 wins and 13 draws! [9 1/2 out of 16] 59.4%

28th World Championship Finals winner had 5 wins and 11 draws [10 1/2 out of 16] 65.6%

Value towards circumstantial proof:   low.  Your ICCF draw rates are still only 2-1 with only one outlier at 4-1. 

Here is a cross table of the Final Round of the Latest ICCF World Championship Match [the 29th]

1 RUS 140915 GM Dronov, Aleksandr Surenovich 2676

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 ½ 1 ½ 1 9.5 out of 16

2 POL 420563 SIM Oskulski, Jacek 2528

½ ½ ½ ½ ½ ½ 1 ½ ½ ½ ½ ½ 1 ½ ½ ½ 9.0 out of 16

3 CRO 900070 GM Ljubicic, Leonardo 2604

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 1 ½ 9.0 out of 16

4 POR 390086 GM Neto, Horácio 2602

½ ½ ½ ½ ½ ½ ½ ½ ½ . ½ ½ 1 ½ 1 ½ 8.5 out of 16

5 ENG 211305 GM Robson, Nigel 2619

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 ½ ½ ½ ½ ½ 8.5 out of 16

6 ITA 240090 GM Finocchiaro, Fabio 2606

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 8.5 out of 16

7 ROU 440534 GM Serban, Florin 2635

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 ½ 8.5 out of 16

8 GER 82299 SIM Schwetlick, Thomas 2466

½ 0 ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 1 8.0 out of 16

9 ITA 249221 GM Vassia, Elio 2618

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 8.0 out of 16

10 SLO 480135 GM Borštnik, Aleš 2583

½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 8.0 out of 16

11 GER 83246 GM Mahling, Thomas 2575

½ ½ ½ . ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 7.5 out of 16

12 GER 80888 GM Busemann, Stephan 2606

½ ½ ½ ½ 0 ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 7.5 out of 16

13 GER 81313 SIM Windhausen, Georg 2505

0 ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½A ½ 7.5 out of 16

14 ESP 160468 GM Manso Gil, Ángel-Jerónimo 2567

½ 0 ½ 0 ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 7.0 out of 16

15 POR 390473 GM Silva, António Augusto M. C. 2550

0 ½ 0 ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ 7.0 out of 16

16 RUS 141241 GM Turkov, Vladimir Sergeevich 2533

½ ½ 0 0 ½ ½ 0 ½ ½ ½ ½ ½ ½ ½ ½ ½ 6.5 out of 16

17 GER 81015 SIM Schmidt, Theo 2358

0 ½ ½ ½ ½ 0 ½ 0 ½ ½ ½ ½ ½ ½ ½ ½ 6.5 out of 16

The winner of the 29th ICCF Correspondence Chess Championship had 3 wins and 13 draws and no losses out of 16 games.

The person who came in last place had 3 losses and 13 draws out of 16 games.

Value towards circumstantial proof:   low.  These draw rates are not out of line with top human tournaments.  Ultimately, you are "proving" that draws are common at top levels, which is already a known/given...this part of your "proof" is already known and accepted pretty much everywhere.  

Clearly as we have strong humans with strong chess engines--[known as Centaur Chess]-- Looking at the winners of the USA Correspondence Chess Championships and the winners of the World Correspondencce Chess Championships we notice as time goes by there are more and more draws. This is a clear indication that with chess played at high levels -- there are fewer and fewer mistakes and the results are coming out with more and more draws.

Recent Chess.com chess engine Match between Stockfish [rated 3400] and Houdini [rated 3407]

This was a 20 game match between 2 very strong chess engines. There were 15 draws out of 20 games.

Having 25% decisive games does not favor your argument much.  This match should have been even more drawish...over 85%...if your premise were an absolute.  And future matches would be beating 85% draws more and more consistently.

Value towards circumstantial proof:   low. 

Stockfish won the match by a narrow margin--3 wins and 2 losses and the rest draws.

I have personally looked at a couple of ICCF Corrspondence Chess crosstables between very strong players and found the winners had something like 10 draws and 2 wins and the bottom players of these crosstable had something like 10 draws and 2 losses. And the players in the middle had 12 draws!!

You were doing pretty good at citing examples for the first time, so why abandon that when you get to the peak of your point?  Maybe the data didn't quite tell the story you wanted? 

This is the strongest form of chess--even stronger than the best chess engines. [it is well known that a human guiding a strong chess engine is stronger than just a chess engine without human guidance]

For now...and the human level of expertise is not as big a factor in the equation.  Let's see what anecdotal "proof" Kasparov has on this:

COWEN: You’ve been a pioneer in what’s sometimes called advanced chess, freestyle chess, or centaur chess, where you pair a human being with a computer or a set of programs. Today, 2017, do you still think it’s the case that a human paired with a set of programs is better than playing against just the single strongest computer program in chess?
KASPAROV: There’s no doubt about it.
COWEN: The human will make some mistakes, so the human will ask Stockfish, Komodo, Rybka, “What’s the best move?” Collate the different outputs, make some kind of judgment, explore some lines more deeply. Put that against Rybka Cluster. Is Rybka Cluster really going to lose many games?
KASPAROV: I think so. Again, it depends on the qualification of the operator.
COWEN: Sure, if it’s the best operator in the world, whoever that may be. Maybe yourself, maybe Anson Williams.
KASPAROV: By the way, I exclude myself from this category because I’m not a very good operator. I’m a very good chess player. A great operator does not have to be necessarily a very strong player.
COWEN: What makes for a great operator?
KASPAROV: Someone who can work out the most effective combination, bringing together human and machine skills. I reached the formulation that a weak human player plus machine plus a better process is superior, not only to a very powerful machine, but most remarkably, to a strong human player plus machine plus an inferior process.
At the end of the day, it’s about interface. Creating an interface that will help us to coach machine towards more useful intelligence will be the right step forward. I’m a great believer that, if we put together a good operator — still a decent chess player, not necessarily a very strong chess player — running two, three machines and finding the best way to translate this knowledge into quality moves against Rybka Cluster, I would probably bet on the human plus machine.

All of this is very strong evidence that chess is a draw with best play [or when neither side makes an error]

No.  It's fairly weak evidence compared to anything we already have from other sources.

5. There have been billions of chess games played in the last 200 years. There has not been one game out of all of these billions of games where it has been shown that either White or

Black won by force from the opening position.

You could not even determine this without solving chess.  If such a game had occurred, you would not even be able to prove it.  That's been your error in logic from the get go, that we can talk about error free chess or provable perfect play at all with our current technology and accumulated knowledge.  The only examples of the above that are definitive and provable best play are tablebases, and we all know how incredibly far away we are from ever having a 32-piece tablebase.

This is not a coincidence. Clearly it is good evidence and more evidence that chess is a draw when neither side makes an error.

Nope.  You love ot decalre stuff that you have no basis for. 

Believe me

Not until you start showing some consistency and rigor in your points and analysis.  Admittedly, this is your first foray into anything involving any real facts, so kudos on that.  But the conclusions you are drawing are not supported by those facts. 

if some chess player ever found a forced win from the opening position that person could use his knowledge to become a multimillionaire.[and, sadly, chess itself might be diminished] 

Yes, this is obvious...and?

Now some will say that the billions of chess games played is only a very small percentage of the possible games which could be played.

...which they are. 

This is true but looking at the zillions of chess games which could be played 99.9999999999999999999999999% of these games would be nonsense gsmes where both sides were making errors

You have far too many 9s in that number. 

almost every move. 1. a4 f6 2. Ra2 Kf7 ilk. [and worse as the game continues] Trillions [or more] games with 5 knights or 6 bishops etc.

It doesn't matter, because "trillions" of games is still a tiny fraction of 1% of total games.  You will have a ton of games with 6-9 queens per side, too.  So what?  It doesn't invalidate anything to say that a portion of the games are nonsense, which is why you tried to pawn off that ridiculous and unsupportable percentage.  Yes, there are a gazillion lines that will not turn out to be viable.  There are also a gazillion that will have to be evaluated, far more than can be calculated in our lifetimes.

Some do not like to admit this but chess playing in general has become stronger every decade for the past 200 years. Chess players are understanding more and more about chess.

Please tell us who it is that has disputed the above.  Quote a single person on this thread, if you can.

The top chess players rated 2700 and above know a lot about chess--yes they lose to chess engines rated 3400 but think of the reasons they lose to these chess engines?

A human cannot calculate nearly as fast as a chess engine. A chess engine might calculate at a million positions a second.

This is a great advantage to a chess engine--to make a fair fight--give the human more time for each move--maybe 10 to 20 days for each move? And then there would be a lot of draws...[as happens in todays Correspondence Chess]

All pretty much irrelevant to anything you are talking about here.  

Also a chess engine can go on and on--a human does not have that stamina. The current crop of top chess players know as much as the top chess engines --it is just that they cannot think and calculate as fast.

Not true.  Just because a GM can, in hindsight, explain to you why an engine might have valued one move over another by .2 centipawns does no mean a GM can or would make the same choices over the course of an entire game no matter how much time they had.  I can't count the number of times I've heard a GM state in an interview something like "well, I don't know why the engine would favor this line, I would just simplify down to a won ending here and call it a day...".  The same way that a beginner will mate with 2 rooks using the same rank-by-rank method every time, even when they *could* calculate some thing shorter, because it's a known and it's easy...a GM will do the same thing, just at a much higher level of play...and because you, Ponz, do not fully understand that level of play pasta certain point, you can't tell the difference.  Neither can anyone else here.

Value of 2 digressions towards proof:  negligible.

There are times when a human can solve chess positions better than the best chess engines. A human can be creative and chess engines cannot. At the age of 75 I solved two problems [given by a chess grandmaster on chess.com] that the strongest chess engines could not solve.

Yes, and those "problem" positions are becoming fewer all the time.  Engines have not come into their own until 20 years ago, and even those engines are hamstrung by human-derived valuations.  They have been riding tricycles up until recently when the training wheels got removed. 

Value towards proof:  none. 

6. Here is another piece of evidence that chess is a draw when neither side makes a mistake:

It is sometimes quite hard to win in the endgame even with a pawn advantage. Sometimes being 2 pawns ahead is not enough to win. Sometimes 3 or 4 pawns up is not enough to win.

Most of us know a king and 2 knights usually cannot force a win vs a lone king.

Value towards proof:  good.  Note, though, that I gave this "evidence" on this thread before you actually stated it.  Not that I am saying you didn't already understand it, but that it's not your point to make,  Your best piece of "evidence" in this huge diatribe is one I posted.

There are positions where one side has a bishop and a protected pawn vs a lone king and cannot win.

Yes. 

There are thousands of positions where one side has a king and 4 pieces against a lone king and cannot force a win.

Yes. 

Clearly this is another indication that it takes a lot to win a chess game...

Yes.  Your point?  Who exactly has said that a forced win, were it to exist, would not be hard to achieve?

Guys, this conversation should be a few sentences as most for each post:

Ponz: There is evidence that perfectly played chess is a draw. I believe this proves chess is a draw with best play.

Bitckler: I agree there is evidence. I disagree there is proof.

Ponz: But look at this position, it's a draw OMG!?!?  misrepresents my evidence

Btickler: Ok, but that's both not germane, and not proof.

Ponz: But look at this rare photo of Morphy with a beard! What kind of trimmers did he use? It must be a draw!  misrepresents my evidence

Btickler: Moving on...

the problem is, not enough here ask "why bother?"

why bother figuring out if chess can be solved? if you are smart enough you wouldn't waste your time with worthless pursuits.

My only comment was that it is unknown if there is a perfect game of chess that ends in 16 moves.

If someone knows that it has been answered I would like to see the work, if for no other reason that I'm a mathematically curious person.

But I'm somewhat inclined to not be impressed by guesswork presented as mathematical proof.

Guys, this conversation should be a few sentences as most for each post:

Ponz: There is evidence that perfectly played chess is a draw. I believe this proves chess is a draw with best play.

Bitckler: I agree there is evidence. I disagree there is proof.

Ponz: But look at this position, it's a draw OMG!?!?

Btickler: Ok, but that's both not germane, and not proof.

Ponz: But look at this rare photo of Morphy with a beard! What kind of trimmers did he use? It must be a draw!

Btickler: Moving on...