Computer Chess Club Archives




Subject: Re: LCT II Fin4, Deep Thought, and Deep Blue (was Re: LCT II results...)

Author: Robert Hyatt

Date: 14:04:50 01/05/98

Go up one level in this thread

On January 05, 1998 at 12:28:18, Albert Silver wrote:

>On January 05, 1998 at 02:52:50, Howard Exner wrote:
>>On January 05, 1998 at 01:54:04, Bruce Moreland wrote:
>>>On January 05, 1998 at 00:37:50, Howard Exner wrote:
>>>>I have the text for this game and the correct move is h3. Play continues
>>>>Rxh6  a3  2. Rxh3 Ra4 3. Rh1  a2 4. Ra1 Ke7   is drawn. In the actual
>>>>Deep thought played h5? and karpov went on to win (Karpov-Deep
>>>ICCAJ, Mar '90, p. 37-40.
>>>Why did DT blow this position?
>>I was wondering about that also. I thought even back then that its
>>node per second count was higher than today's fastest machines/programs.
>>Here is the entire game for those who would like to look at the finish.
>>[Event ""]
>>[Site "Harvard"]
>>[Date "1990.01.01"]
>>[Round ""]
>>[White "Karpov Anatoly"]
>>[Black "COMP Deep Thought"]
>>[Result "1-0"]
>>[BlackElo ""]
>>[WhiteElo ""]
>>[ECO ""]
>>1. e4 c6 2. d4 d5 3. Nd2 g6 4. c3 Bg7 5. e5 f6 6. f4 Nh6 7. Ngf3 O-O 8.
>>Be2 fxe5 9. fxe5 c5 10. Nb3 cxd4 11. cxd4 Nc6 12. O-O Qb6 13. Kh1 a5 14.
>>a4 Bf5 15. Bg5 Be4 16. Nc5 Qxb2 17. Nxe4 dxe4 18. Rb1 Qa3 19. Bc1 Qc3
>>Bd2 Qa3 21. Bc1 Qc3 22. Rb3 Qa1 23. Bc4+ Kh8 24. Bxh6 Qxd1 25. Bxg7+
>>26. Rxd1 exf3 27. gxf3 Ra7 28. Bd5 Rd8 29. Rb5 Ra6 30. Be4 Ra7 31. Bd5
>>Ra6 32. Rc5 Rd7 33. Kg2 Rb6 34. Bxc6 bxc6 35. Kf2 Rd5 36. Rxd5 cxd5 37.
>>Rc1 Rb4 38. Ke3 Rxa4 39. Rc5 e6 40. Rc7+ Kg8 41. Re7 Ra3+ 42. Kf4 Rd3
>>Rxe6 Rxd4+ 44. Kg5 Kf7 45. Ra6 a4 46. f4 h6+ 47. Kg4 Rc4 48. h4 Rd4 49.
>>Rf6+ Kg7 50. Ra6 Kf7 51. h5 gxh5+ 52. Kf5 Kg7 53. Ra7+ Kf8 54. e6 Re4
>>Rd7 Rc4 56. Rxd5 h4 57. Rd3 Ke7 58. Rd7+ Kf8 59. Rh7 h5 60. Ke5 h3 61.
>>f5 Kg8 62. Rxh5 a3 63. Rxh3 a2 64. Ra3 Rc5+ 65. Kf6 1-0
>>The Louguet 2 endgame #4 picks up on move 59.
>Of course it is strange, but maybe not. A few things come to mind
>regarding it and even Deep Blue as long as we're at it. Deep Thought was
>already so fast that it was thought (perhaps not by all, but certainly
>by many) to be able to literally "solve" such a position, yet obviously
>it couldn't. I don't know if it's so strange. I remember Kasparov's
>crushing 25 mover against it in a 25-minute game, a time control that
>should have been to DT's utmost advantage, and yet it's tactics betrayed
>it. Knowledge is obviously important in evaluating the positions, but
>back then it was believed (Bob, feel free to tell me I'm talking out of
>my a** if I'm dead wrong) that this could be overcome by sheer

I wouldn't be surprised that at this (deep thought) point in time, they
were convinced that speed was everything.  They figured out otherwise as
time went on, leading to the development of the current deep blue chess
processor, which is very sophisticated in the evaluation, more so than
likely any program ever envisioned.

Now they have *both*...  fantastic search *and* fantastic evaluation...

>              Even when I look at Deep Blue's games against Kasparov one
>is left with a strange impression. Sure, it found some brilliant moves
>(but then with that kind of computing power, that should surprise no
>one), but there were also not only the very strange moves (plenty to
>mention, like that queen swap Qxg6 in one game - Don't remember which
>exactly right off hand), but also that incredible opportunity it gave
>Kasparov in game two. So much attention has been given to the fact that
>Kasparov missed it, that the fact that Deep Blue allowed it has been
>almost overlooked. How on earth could it not see it? Especially with the
>depth of calculation it was wielding.

Easy.  They couldn't search the nearly 60 plies necessary to find that
this led to a perpetual.  Of course *no* one could search that deep
so it's not a big issue.  Everyone seems to think that this is a
sort of perpetual that is easy to see.  It isn't.  For simple proof,
that the current world champion missed it as well...  I'm not concerned
DB misses a move that Kasparov missed...

But it wasn't easy to find.  And to date, no program has come close to
finding it...

>                                       I have heard many, many times on
>how smart Deep Blue is. On just how much knowledge it has on the game.
>But there is more to it than just the 'amount' of knowledge, and I'm not
>simply referring to quality, there is also the matter of administering
>it. Just knowing all the factors that constitute a position doesn't mean
>that I will automatically choose the correct ones to judge it and
>subsequently choose my move. There are several programs out there that
>are touted for having a vast knowledge of the game. I remember how
>Vincent Diepeveen proclaimed (I'm probably going to get lynched for
>this) that because his program Diep was seriously weighted down by all
>of it's understanding of the game it couldn't compete properly at normal
>time controls, but that it was quite probably the world champion in
>postal chess. Brave words that weren't confirmed by crushing results
>over all of it's rivals in last year's KK tournament. Even if DB's
>program WERE released, tuned especially for a microprocessor, I honestly
>don't believe it would be Numero Uno. Anyhow, it looks like I've drifted
>quite a ways from the original subject. So much so that I've changed the
>header so that it isn't thought I'm writing this while completely drunk.

I think it would be numero uno by at least 200 rating points, based on
expected wins over the rest of us.  Remember that one processor searches
2.4 million nodes per second, doing a *full* endpoint evaluation that is
most likely better than anything anyone else is (or will be) doing...
would be at least 10X faster than the next closest program, with
100X the knowledge of the fastest micro-based program (I am assuming

>                                    Albert

This page took 0.01 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.