The best players, when facing each other, are often content to draw as Black. As White, you will often see top players deviating from "best" moves in order to increase their winning chances.

Engines have no problem playing for a draw if that's what the best moves lead to. Now, engines have to play matches of up to 100 games to get a decision.

i predict that in 20 years, the best engines won't be able to win a single game in 1000 against equal engines.

That is an interesting prediction. I am inclined to disagree fairly strongly. In simple terms the fraction of won games between top engines takes a long time to halve, and you need to get to a fraction of about (1/2)^10, so it is a fairly large multiple of this long time.

Curiously the 2016 TCEC superfinal (the top 2 engines match) had 25% decisive games (25 from 102), which was a big, statistically significant increase over the previous year! People had been saying draw death was near in 2015, but this threw a spanner in the works.

The rate of draws is far from only being dependent on the quality of the players. The choice between positions that are fairly even with chances and fairly even and safe is independent of simple standard of play, but it has a huge effect on the draw rate. You can see this in human world championship matches, where play with black has become increasingly negative. A computer has no psychological barrier to switching from a cagy start to an aggressive mode when white's aggression creates possibilities.

s23bog asked:

"Is there any work being done on mapping out every possibility?"

The programers of AlphaGo decided against this approach. More possibilities exist in go than chess. Instead they employed a different approach and are expanding their research  with "neural networks". 

AlphaGo is retiring after beating the worlds best go player Ke Jie 3 straight. The 2nd game. Alpha's analysis showed Ke played almost "perfectly" for 50 moves but eventially succumbed to the pressure.

Deepmind will continue AI research in other areas besides go. They may well heading the field, a step ahead of chess programers.

Elroch wrote:

You can see this in human world championship matches, where play with black has become increasingly negative. A computer has no psychological barrier to switching from a cagy start to an aggressive mode when white's aggression creates possibilities.

I disagree with this evaluation. Black has been winning more games than ever before at the top level, relative to the percentage of games White won. When players want to "fight" Black is doing quite well.

I said "in world championship matches", where draws are more common than anywhere else. This is partly the result of different marginal returns to tournaments.

You don't. The best moves are the best moves. 

Who can forget Karpov vs Kasparov, 1984-85? With Karpov leading 4-0 after 9 games, the ensuing 17 draws, a win by Karpov in game 27 to lead 5-0. 

21 more games, 3 wins by Kasparov with 18 draws. The players wanted to continue, but FIDE Pesident Campomanes called the match off.

I think computers have come far enough in chess. They can beat the best human players, so why bother going any further? It's just a matter of calculation power and advanced programming at that point, and I feel it takes the mystery out of the game. Also, I think computers provide enough insight to the game as is, if you want to know where the game took a wrong turn you can always review the game with a computer and see where the plus/minus changes, and I think that's enough technology. I also feel It's only going to result in more cheating.

"AlAlphaGo's algorithm uses a Monte Carlo tree search to find its moves based on knowledge previously "learned" by machine learning, specifically by an artificial neural network (a deep learning method) by extensive training, both from human and computer play." Wiki

Will computers ever solve chess?

In theory. Yes for certain.

In practice. No, seemingly, given the shortage of time, space, and other essential elements.

So, i think the answer is No, with caveat that it could be Yes.

A man with convictions ! Not afraid to be on both sides of the fence ! Has all the angles covered, is never wrong, but is he ever right?

i think the answer is Yes, with caveat that it could be No.

I underestimated the power of distributed computing (posts #1224, 1229 above). I previously said that GIMPS has achieved 150 TeraFLOPS. It is running at 300 TeraFLOPS (simply by volunteers linking PCs on the internet).

Others are doing even better: SETI@Home is running at 681 TeraFLOPS. (searching radio noise from space to see if it has signs of Extra Terrestial Intelligence)

Supercomputers have also made strong gains: China's TaihuLight has achieved 93 PetaFLOPS (but they probably won't use it to study chess).

These are other advances that will help solve chess, compared to computers in 2007 when checkers was solved. I agree with Campter: Yes - chess may soon be solved. with the caveat it could be No.

@vickalan... agree with camter ?

As he so eloquently stated; he might be on the other side of the fence in his previous post !

I'll not straddle the fence !

Chess is solved, now and forever...

It's a Draw !

The point of neural networks is that they are designed to be much better at pattern recognition.

Prof, you could be correct in saying it is a draw, In fact that is a widely held belief.

The second possible outcome is a White Win. Now, that has some credence due to the fact that it is, or at least used to be the most usual outcome.

Th only third possibility is a proposition never taken seriously, that Black has a Win, which could only mean that White was in Zugswang at game start.

The thing that the experts all say is that White is more successful than Black, because White has, wait for it........

the INITIATIVE, whatever that means.

What is this beast called the Initiative? How does one define it, or at least make it a meaningful term?

Chess has no chance of being solved unless we can give a digital, or programmatical meaning.

Or does the Computer not at all concern itself with the idea, and just bash away at the tree of analysis, this tangle of nodes, which make up that tree? Because, unless we have that tree, we have nothing to prune, or discard?

My head starts to spin!   

A former WC from days gone by said...

1. e4 and White is in the throes of defeat !

Ah.. the elegance of a counter-attack !

100 PetaFLOPS is 10^17 floating point operations/sec.  Evaluating a chess position is not 1 operation, mind you, nor is it 10, so let's be kind and say it falls in the 100s order of magnitude, which knocks 10^17 back down to 10^15 positions/second, which is 8.64^19 positions/day, 3.15^22 positions/year.

At that processing rate (assuming infinite memory/storage and ignoring all the issues thereof already laid out) you would solve chess in...3.175^24 years.  I guess you could amortize a loan for the duration on the $273 million for the supercomputer...

Adding in storage, you solve chess...never.

FLOPS is a way to see how computers have improved, but it's not used to measure chess positions. Checkers and chess are analyzed with integer-like operations (floating point numbers with exponents not needed).

Checkers was solved using about 2000 GFLOPS (2 TFLOPS).  Today some supercomputers operate at 50,000 TFLOPS (50 PetaFLOPS). So we  have about 25,000 times the processing power that was used for checkers. You can "take-away" a few orders of magnitude of calculations I showed previously, and it's still easy to see a path to chess being solved. The answer to this thread's question is probably Yes.

A baseball cap with two electrodes attached. Powered by 2 AA batteries. A very small chip that stimulates the brain, custom settings.

No more drugs, alchohol, tobacco. The next big leap in computing technology.