Computers & Chess

Yesterday I met an Expert who was editing an opening book for someone using Houdini.  We had a cordial discussion of the strength of computer engines and chess.

A few interesting questions arose...

1) How long until a computer plays so well that a human can never beat it again? (5 years was my guess).

2) Will computers give credence to Bobby Fischer's lament that "chess is played out", due to the steadily increasing % of grandmaster draws, based in part on computer-proven weak lines which have been pruned from the chess tree? 

3) Is it fair to use how computer-like or engine-matched a human player's moves are in a tournament setting as evidence of their cheating?  (This would seem to me to violate the freedom to choose inherent in chess, which just might align with top engine move picks, however improbable.)

4) If we get quantum computers and chess is one day "solved" as was checkers, will perfect play for both sides result in an automatic draw being agreed?  (This seems likely to me, as forming a mating net is way too complicated not to be foiled by perfect play by Black.)

Questions 2, 3, and 4 are beyond my scope but you're about 5 years off in your estimation for question 1.  A modern engine like Komodo or Stockfish is simply unbeatable by even super GMs.  There have been a whole series of recent attempts right here on chess.com by various titled players to beat Komodo with material odds.  You should try and check them out or just google Komodo if you doubt its' strength.  I wish I could remember Komodo's rating...Maybe 33something?  I know way over 3,000 anyway and at least 3200.  I'm sure someone else will correct me. 

Interesting...

I've heard the new engines are so strong they are inaugurating a new Romantic era in chess, like the age of Alekhine, but instead of spectacular sacrifices that were beautiful psychological blows but not solid objectively, the top engines can now play spectacular sacrifices that are rock solid, and are even now evolving new opening theory with new gambits never seriously considered.

I would be really impressed if there IS a forced mate for white in that position.

1. Qg4-c8 Kf8-g8 2. Bd8-c7 Qe8xc8 3. g6xf7+ Kg8-h8 4. Bc7-e5...

(Sting SF 6, mate in 13)

It didn't take Sting very long to find the mate. DeepFishMZ 4.1 also found the line, but it took almost 50 minutes. Also, Stockfish Matefinder took about 13 minutes.

It's possible someone will make a quantum algorithm that works partly (or fully) as a chess engine, but I doubt it. Quantum computers are definitely NOT very fast digital computers.

Interesting...  never knew/heard that about quantum computers being slower.

I would assume however that's somewhat compensated by Qbits gaining another order of magnitude with every bit for permutational representation -- meaning very large trees/datasets could be explored in a far smaller qbit space than bit space -- but I really am no expert beyond these gray suppositions.

I have no idea how reading said bits, superimposed or whatever, still wouldn't involve a very large Big(O) in that we have to translate them back to non-quantum data.  I theorize that maybe one could send off a bunch of deep tree searches using QBits and read the resultant evaluation only -- but this is a long way off. 

"1) How long until a computer plays so well that a human can never beat it again? (5 years was my guess)."

Computers are already at that point, basically. Of course the answer is statistical (and your question should be, also), but I'd guess the percentage of GM wins against the best combination of hardware and software might be down to less than 5% now, maybe even under 1%.

"2) Will computers give credence to Bobby Fischer's lament that "chess is played out", due to the steadily increasing % of grandmaster draws, based in part on computer-proven weak lines which have been pruned from the chess tree?"

This is mostly a psychological question, based on the effect of computers on human emotions. Computers can excel in chess all they want, but in over-the-board games between humans, chess will always be complicated enough to evade complete human mastery, therefore it will always be popular with humans in its current form.

"3) Is it fair to use how computer-like or engine-matched a human player's moves are in a tournament setting as evidence of their cheating?  (This would seem to me to violate the freedom to choose inherent in chess, which just might align with top engine move picks, however improbable.)"

That method can certainly generate *evidence* but never proof.

"4) If we get quantum computers and chess is one day "solved" as was checkers, will perfect play for both sides result in an automatic draw being agreed?  (This seems likely to me, as forming a mating net is way too complicated not to be foiled by perfect play by Black.)"

People really get sidetracked by quantum computers. Quantum computers are essentially the same stupid computers we already have, just with a lot more parallelism, but blind parallelism is not the key to intelligence. That type of idiot savant parallelism is not even particularly helpful in the domain of tree search problems like chess.

Anyway, in answer to your main question, the experts have long tended to agree that chess is inherently a draw, even with a poor opening move.

----------

(p. 114)

   Quantum computing is not directly applicable, however, to problems such as

playing a board game. Whereas the "perfect" chess move for a given board is a 

good example of a finite but intractable computing problem, there is no easy way

to test the answer. If a person or process were to present an answer, there is no

way to test its validity other than to build the same move-countermove tree that

generated the answer in the first place. Even for mere "good" moves, a quantum

computer would have no obvious advantage over a digital computer.

Kurzweil, Ray. 1999. The Age of Spiritual Machines: When Computers Exceed Human Intelligence. New York, New York: Viking Penguin.

----------

(p. 63)

      - 205 -

   Theoretically, a game should end in a draw

when both White and Black play perfectly.

Chernev, Irving. 1937. Curious Chess Facts. New York: The Black Knight Press.

Appreciate your thoughtful replies.  A few counter points, genially meant.

1)  "Computers are already at that point, basically. Of course the answer is statistical (and your question should be, also), but I'd guess the percentage of GM wins against the best combination hardware and software might be down to less than 5% now, maybe even under 1%."

I suppose it's largely unprovable when computers reach the point that humans can never beat them -- trying to prove a negative -- but lack of evidence (i.e. no human victory for say 10 years) would strongly suggest we'd gotten there.

2)  "This is mostly a psychological question, based on the effect of computers on human emotions. Computers can excel in chess all they want, but in over-the-board games between humans, chess will always be complicated enough to evade complete human mastery, therefore it will always be popular with humans in its current form."

Also though the % of GM games ending in draw has gone up over time, with computer study likely accelerating the curve (i.e. tree pruning of bad branches).  Now of course for 99.99% of chess players, this "played out" canard will remain only that.

But it's interesting whether the very top players will start to assert this more and more.

4)  "People really get sidetracked by quantum computers. Quantum computers are essentially the same stupid computers we already have, just with a lot more parallelism, but blind parallelism is not the key to intelligence. That type of idiot savant parallelism is not even particularly helpful in the domain of tree search problems like chess."

Yes.  But my understanding is if one could have theoretically near infinite tree search capability from quantum computers, then actually said computers would not even have to evaluate moves (the subjective aspect you mention) but only node search to wins and losses.  Since the chess tree of all possible games G is finite -- even though it is immensely huge -- the question that inheres is whether we will ever be able to search it thoroughly, as with Checkers.  

(I'm not sure, contra Kurzweil, there's no way to test it.  Perhaps the method thoroughly exhausting the tree could be proved by something like induction, and we could prove by parts that the tree search was thorough.  In mathematics, e.g., there are all sorts of proofs of properties of infinite sequences, etc., that do not require a full enumeration.  All that's needed is a proof of algorithmic correctness.)

Yeah best play from both sides must be a draw.  If computers had an ability to search the entire game tree G, they would agree a draw on move 1 every time!  :)

 

Incidentally, on a tangential thread, I met Ray Kurzweil once via a local music store in Louisville in his capacity as the piano keyboard inventor that bears his name.  A brilliant guy.

I find this whole futurist movement fascinating, with their AI claims about uploading human consciousness and life extension.

I would tend to find myself somewhere in the middle about what computers can do between yourself and Mr. Kurzweil.  Meaning:  

Because chess is a finite system with strict rules of play and a limited permutational space, computers may eventually play chess perfectly.  

But on the other end of the continuum, the current Google/futurist claim that we are on the cusp (2037) of uploading -- or even simulating -- human consciousness is likely hyperbole as we are still infants in even understanding consciousness, and the issues with reproducing it go far beyond the algorithmic.  

Likely the hardest part of consciousness creation is non-algorithmic (i.e. the qualia of red as consiously experienced versus a robot eye detecting a color and pattern matching it and setting an internal variable to "red".)

Now if one takes a black-box view of the brain, as I often do, then it follows one day we will succeed in creating artificial consciousness, perhaps by using massive parallelism.  I just think the time curve seems way too optimistic, with people jumping from Moore's law to consciousness when even creating an ant would be a stupendous achievement.

Embodiment is needed, which does not involve Moore's law of progress for digital circuits.

 Thank you for this interesting thread, and for your lucid writing. I imagine "consciousness" (synonymous with the term "awareness") as the context or space (degrees of freedom) within which all phenomena occur.  So maybe instead of "creating consciousness", we would be constructing complex adaptive networks through which consciousness can viscerally experience an energetic environment.  Not sure my distinction is real or purely semantic.  But if we don't assume we're trying to create consciousness, then we simplify our task.  We merely need to construct sufficiently complex intelligent devices :-)  

How many years has it been since a top grandmaster beat a top chess program in a serious game?  I don't believe it has happened.  

Not at chess though.

I've googled it and can't find a single game in the last five years where a grandmaster has beaten a top computer program in a serious game.  If you can show me otherwise then I will concede you are right.

s23bog 

Any game without odds.  Show me one.

When computers do solve chess it will be a sad day indeed.  I much more interested in computer AI playing like humans.  Imagine if you could get a computer to analyze all of Morphy's games and really like play like Morphy.  Then virtual reality brings up an image of him and it would be like truly playing the greatest player from the past.  There are so many variables to consider though.  The lack of opening book knowledge could be compensated by playing in a chess960 setting. 

For practical purposes it is solved for humans.  A patzer could use a laptop and find every mistake Fischer or Kasparov ever made in analysis mode. 

For over the board play it will never be solved.

It is solved only compared to humans.  Chess programs will only get stronger but will never "solve" it. 

By reverse engineering. Currently we have a tablebase that can calculate all positions when there are a few pieces /pawns left. If we have enough calculating power which we currently don't have then we can calculate all positions from the starting position. Then chess would be truly solved and the computer can create his own openings.

Nowadays chess engines use opening books created by Grandmasters. This is because engines are dumb and cant figure that out for themselves yet. A bigger tablebase would solve this problem.