How long it will take a Super Computer to solve chess?

[towforce]

Here's a useful schematic diagram:

[pichy]

Here's a useful schematic diagram:

It has not been proven that Deeper Blue made all the moves without human intervention, since GM Kasparov asked for the output log of deeper blue but it was never provided to him.

[rbarreira]

It has not been proven that Deeper Blue made all the moves without human intervention, since GM Kasparov asked for the output log of deeper blue but it was never provided to him.

Here are the logs:

http://www.research.ibm.com/deepblue/watch/html/c.shtml

According to the "Behind Deep Blue" book all the logs were provided to the Kasparov team. The Kasparov team asked for them before the end of the match, but that would be ridiculous as the logs might contain hints towards Deep Blue's weaknesses.

Kasparov also complained that he didn't get enough Deep Blue sample games for his preparation (even though he himself played previous versions). I also find this demand ridiculous, surely the GMs don't send each other the practice games they played one month before their matches (including opening preparations etc.).

In all, I think Kasparov was too arrogant which made it easy for him to become a bad loser...

[pichy]

[quote="pichy"]Solving chess means finding an optimal strategy for playing chess, i.e. one by which one of the players (White or Black) can always force a victory, or both can force a draw.

Recent scientific advances have not significantly changed that assessment. The game of checkers was solved in 2007,[5] but it has roughly the square root of the number of positions in chess. Jonathan Schaeffer, the scientist who led the effort, said a breakthrough such as quantum computing would be needed before solving chess could even be attempted, but he does not rule out the possibility, saying that the one thing he learned from his 16-year effort of solving checkers "is to never underestimate the advances in technology".[6] Assuming computational power continues to increase exponentially, chess would be solved "before 2250".[7]
[edit] Notes

http://talkchess.com/forum/viewtopic.ph ... 48&t=15236

[jefk]

[quote="pichy"][quote="pichy"]Solving chess means finding an optimal strategy for playing chess, i.e. one by which one of the players (White or Black) can always force a victory, or both can force a draw.
>Recent scientific advances have not significantly changed that assessment.

well with the normal Fide rules, like the 50 move endgame rule, and
the human will to solve chess, ie prove its a draw, could be done
within 50 years i suspect. besides fast comps also large
storage capacities would be useful.

but its not so impossible as many think, someone told me once
that the maximum lenght of a chess game considering this
50 move (draw)rule is about 1300 moves. Secondly,
if you look at those lines whereby you try to maintain
a positional advantage for white, you get lots of transpositions.
But storing this moves should better be done in a systematic
way, not by humans , but by an engine, weeding out
the tactical mistakes (losing a piece means a loss,
at least in most circumstances, better say at 23+ ply
search depth a cutoff of eg. minus 6 would mean a loss.

Then store all possible moves, minimax, and voila, its a draw.
As improvements in opening theory stay within the
drawing margin of chess (*)

Other problem would be to remove this Fide 50 move rule,
and then try to see of eg 1d4 could be a forced win.
Probably not, but we cant rule that out.
Well that could take hundred years i suppose,
rough estimate.
As for FRC chess, well many human chess players tend
to be conservative and dont learn so much opening theory
anyway, so i noticed at my chess club they dont like it.
And other chess variants are being liked even less.

But at top level who knows, maybe some sort of FRC
chess could become more popular.

jef
(*) PS example, yesterday received the latest Kaufman book,
a good book, and some of the things i wrote recently i have
to correct. Against Botvinnik semi-slav there still seems to
be small advantage for white. With the Indian systems,
namely with the Nimzo Qc2 (classical) he shows how to
get a small advantage by keeping the two bishops pairs
(and a later move like f3). Also improved a bit on some
of his important main lines, namely Breyer and Gruenfeld
which he is advising for black. Well i found some *small*
improvements and now Sicilian and QGD are my main
defences
As for Catalan, well he didnt give it, and with his Nimzo
line i don't look at it anymore. although i tried to find
advantage with Komodo4 (i like the eval, but its slow
still with 1 cpu and i don't like this complicated aquarium
Idea stuff) but couldnt achieve significant advantage
so i gave up. So its still 1.d4! for me, within drawing
margins, and Sicilian still is *lots* to investigate.
Plenty of chances for both sides.

[h1a8]

My proposal would be to create a perfect strategy to not lose the game for black. The strategy will be vast in complexity and only able to be managed by a super computer using a large book and key algorithms once out of book.

The motivation of this technique was two things:
1. Proving complex mathematical theorems with elegant proofs
2. Proving certain chess endgames a win although the number of possible ways the endgame can go is vast.

Many have posted positions where a win could be proven (where computers can't see). This proof required a concept and algorithm to employ the technique.

The best try so far at my proposal was Pablo (father) and in a book called "The beginner's game". Although somewhat flawed, the book has an ingenious insight on how chess can be solved without the complete use of brute force. Although irrelevant, I even used the techniques of the book to draw against 1900 and higher rated opponents several times when I was only around 1500-1600 level.

[towforce]

..The best try so far at my proposal was Pablo (father) and in a book called "The beginner's game". Although somewhat flawed, the book has an ingenious insight on how chess can be solved without the complete use of brute force.

What was the insight, please?

[jefk]

[quote="towforce"][quote="h1a8"]..The best try so far at my proposal was Pablo (father) and in a book called "The beginner's game". Although somewhat flawed, the book has an ingenious insight on how chess can be solved without the complete use of brute force.[/quote]
What was the insight, please?[/quote]

yes i know this book the beginers game;
it thought its a certain setup with knights on e2 and d2,
and fianchetto of the bishops, you also can do it with black;
and he had some other variations;
well i wasnt impressed, but the idea not to lose
for black indeed looks like a good approach for me.
and then by default if you reach the end, then its a draw
not so easy in practice, but the thought experiment
for me is understandable; even although i have
been called an idiot

eg after 1.d4 dont play e5? or ..g5??
lateron it will get more complicated, i can assure you
as its not only material, losing a pawn, but also
tactics and even positional aspects which will be important.
but probably such a database will be smaller then
storing all drawing variations.. (ie not losing vars
for white)

and against e4 e5! is fine, look at the Kaufman book.
for beginners the Sicilian is not recommended ofcourse.
losing moves against e4 are probably ..f5? and ..b5??

and yes, in endgame situations you have fortresses
and so on, difficult to win by computers, especially
within this 50 move rule.
but i expect the software will also improve the
next fifty years, genetic algorithms seem to be
a way to make progress, although i still have to
study it a bit more.

jef

[JBNielsen]

Solving chess means finding an optimal strategy for playing chess, i.e. one by which one of the players (White or Black) can always force a victory, or both can force a draw.

Recent scientific advances have not significantly changed that assessment. The game of checkers was solved in 2007,[5] but it has roughly the square root of the number of positions in chess. Jonathan Schaeffer, the scientist who led the effort, said a breakthrough such as quantum computing would be needed before solving chess could even be attempted, but he does not rule out the possibility, saying that the one thing he learned from his 16-year effort of solving checkers "is to never underestimate the advances in technology".[6] Assuming computational power continues to increase exponentially, chess would be solved "before 2250".[7]
[edit] Notes

http://rjlipton.wordpress.com/2010/05/1 ... s-one-day/

If you by solving chess is thinking of building a database of all the positions in chess and then connect them like endgame-databases is build, it will take a huge computer.

20 years ago I calculated, that such a computer would have a size 123.000 times bigger than the earth because of the many possible positions. And how long time would it take to connect the positions...???

And what would you end up with? A computer that perfectly wins every won position. But in all the drawn positions (and I assume it includes the starting position), it would only be able to avoid the loosing positions. It would have no idea how to build up some pressure etc.

At this specific part of the game the humans are still superior. Choose an opening or a type of game, that suits you better than the opponent. Play a complicated move when your opponent is in timetrouble. Make a trap when it is possible.

[EroSennin]

If it was possible to use today's best supercomputer all on chess, do you think it would not lose at long time controls? If not, then how about first building a huge opening book with it? I am also assuming there would not be any problem with using all of the supercomputer's resources.