Chess will never be solved, here's why

#1169
An evaluation function provides a relative truth, the endgame table base provides the absolute truth.
It is better to calculate deeper and hit the table base earlier and more often with a rudimentary evaluation function than to use a more refined evaluation function but not calculating as deeply.
Besides also Stockfish shifted to neural network to refine its simpler evaluation function.

#1171
Yes, I firmly believe GM Sveshnikov: 3 cloud engines can weakly solve chess in 5 years.
I also presented facts and figures in support.
It is just a matter of calculating from the 26-men tabiya to the 7-men endgame table base.
The evaluation function only serves to guide the search. The final evaluation comes from the endgame table base.

I am a big fan of Sveshnikov, and he was wrong in this case. The "just" is virtually as big a problem as solving chess and it certainly does not suffice. The reason is that it fails to allow for every opponent's move in the lines to the tabiya. If these were added, that part alone would surely be bigger than the largest present tablebase.

I am not familiar with the exact quote from Sveshnikov, but it seems highly likely that what he meant was to solve chess FOR PRACTICAL PURPOSES. i.e. not rigorously. The problem with this is that without a proper rigorous solution, it is merely tested by practical results, i.e. inadequately.

#1173
GM Sveshnikov was a world class analyst.
He was called wong on the Sveshnikov Sicilian, on the Kalashnikov Sicilian, on the Alapin and on the French Advance Variation too, but later he was found to be right.
Who are you to tell he was wrong?
What rational arguments can you provide?
At least I provided facts and figures in support.

I respect his ideas for practical play. I do not know whether an oracle would find them valid and nor do you.

I am commenting on his inferred reasoning, which cannot be sound. It's a judgement call, like Ponz saying he knows chess is a draw.

#1173
This is the exact quote from GM Sveshnikov:

"Chess is an exact mathematical problem. The solution comes from two sides: the opening and the endgame. The middlegame does not exist. The middlegame is a well-studied opening. An opening should result in an endgame.... Give me five years, good assistants and modern computers, and I will trace all variations from the opening towards tablebases and 'close' chess. I feel that power."

an exact mathematical problem - i.e. rigorously
the opening - the 26-men tabiya for each of the ECO codes
the endgame - the 7-men endgame table base
Give me - somebody has to pay for it otherwise it cannot be done
five years - 5 years
good assistants - those prepare the tabiya
modern computers - more than one, 3 cloud engines of 10^9 nodes / second
I will trace - i.e. calculate
all variations - i.e. all ECO codes
'close' chess - weakly solve chess, i.e. provide an ideal game with proof that all moves are optimal

The claim does not survive analysis.

The idea that tabiya suffice is ridiculous. You literally might as well say Fine's Chess Endings will serve as a tablebase.

#1176
"develop the required algorithms" ++ the algorithms are already there e.g. Stockfish
"the natural lack of prioritisation of chess" ++ checkers was solved and chess probably has a higher prioritisation, but it is also more complicated
Chess can be weakly solved in 5 years with present hardware and present software, but for it to happen somebody has to actually do it and somebody has to pay for it.

#1179
The tabiya are not even essential, but they speed up and simplify.

The ECO codes partition the big task of solving chess into 500 smaller subtasks.
Chess is most complex after 3 exchanges i.e. 6 captures:
32 men 1.89 × 10^33
31 men 1.71 × 10^34
30 men 1.64 × 10^35
29 men 1.53 × 10^36
28 men 5.46 × 10^36
27 men 1.05 × 10^37
26 men 1.08 × 10^37
25 men 6.14 × 10^36
24 men 3.19 × 10^36
23 men 5.66 × 10^35

Hence it makes sense to start from 26-men tabiya.

I agree.  
And the tablebases don't even allow for castling and en passant.
Also - would going from 8 to seven pieces be more or less difficult than going from 7 to 8 ?
They've 'started work' on 8 pieces.  Whatever that means.
Maybe that's going to take 20 years?  Again without castling and en passant?
How 'weak' are they going to make it?  
It begins to demonstrate and reveal the difficulty of the task that they chose to skip castling and en passant.  The computers just aren't powerful enough to allow for it.

It's over 100 times harder for each additional piece. Will be for quite a while.

The only advantage of aiming for a smaller tablebase would be if you could avoid a large part of the larger tablebase on the way.

You are ignoring all promotions? Or just under-promotions?

#1184
The count excludes excess promotions that is a small error in -.
The count is an upper bound, that is a small error in +.
The count does not apply left/right symmetry when castling rights are lost, that is an error of almost factor 2 in +.
The count does not apply 8-fold symmetry in pawnless positions, that is a small error in +.
The count includes many positions that will not ever occur in a real game, that is a big error in +.
Taken 5 errors together the count should be about right and too high.

'Over 100 times harder for each'
Begins to describe just how daunting it is.
Could that mean that if it takes a year to 'solve' for 7 - its going to take 100 years for 8 ?
Even if its only a ten multiplier each time - that means for 17 pieces - its going to take 10^10 years for that one - plus the time taken to get to that point !  
That's at least ten billion years !   

You keep making this same mistaken logical leap.  Stockfish is to solving chess as a toddler in diapers is to grand unification theory.  We cannot follow the baby, even if that baby is far smarter than we are ...your vast oversimplification and shedding of 20+ orders of magnitude would require perfect selection of the vastly reduced set of possibilities you claim will do the job...and Stockfish will not give you that.

Your whole crusade reminds me of the movie A Bridge Too Far.  At every point you stretch feasibility past its reasonable limits, and then you ask people to buy that all of these infeasible factors will somehow work like a Rube Goldberg machine.

Still - I think of tygxc as 'a kind of good guy' here.
He is bravely doing a kind of 'Devil's Advocate' job as it were.

He puts out these points - giving us opportunities to disagree.
Plus the forum subject itself - is harmless.  
Its not like the 'mass denials' of huge problems in the world.

#1194
"from the opening towards tablebases " means: start at the opening and calculate towards the tablebase
Stockfish is just there to generate future from a previous position in a meaningful way.
Think of the 10^36 positions as position 1 = initial position = draw and position 10^36 = 2 bare kings = draw. In between there are stepstones that are draws with 32, 31, 30... 7, 6, 5, 4, 3 men. Besides the stepstones are pitfalls for black and pitfalls for white.

#1190
Chess is a mathematical problem.
Game Theory is a branch of mathematics.
Combinatorics is also a branch of mathematics.

When chess playing software was first created, everything (naturally) had to derive from human play and human valuations starting with the most basic pawn = 1, knight/bishop = 3, rook = 5, queen = 9, and the software basically used these values to decide what to do next whenever it could not calculate a brute force advantage.  Slowly over time humans tweaked the numbers to experiment, and the software would do better or worse, against humans and against itself and other software built on the same basic building blocks, and best tweaks would stick, and the software got better. 

Still, then and even now, computers would/will get "stuck" into positions where it would move rooks back and forth or otherwise give away tempi because it could not see a way to improve.  This was and is a problem with the human-derived valuations and tweaks and the brute force fading horizon effect.

The machine learning engines will change that, because they build their own valuations from scratch, but they *just* came out, and their play is also "diluted" by playing non-machine learning opponents...essentially the engines learn to beat what they are presented with.