Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Socrates vs Deep Thought2

Author: Robert Hyatt

Date: 14:23:23 11/23/98

Go up one level in this thread


On November 23, 1998 at 11:53:45, Mark Young wrote:

>On November 23, 1998 at 09:44:05, Robert Hyatt wrote:
>
>>On November 22, 1998 at 22:23:04, Howard Exner wrote:
>>
>>>On November 22, 1998 at 11:22:30, Amir Ban wrote:
>>>
>>>>On November 22, 1998 at 10:52:01, Robert Hyatt wrote:
>>>>
>>>>>
>>>
>>>>>Another good one, although I can't tell you where the thing starts, is the
>>>>>game *socrates vs Deep Blue Prototype at Cape May new jersey, 1994.  The
>>>>>critical point starts with both programs castling on the queenside if I
>>>>>recall, and DB prototype initiating a pawn advance on that side.  If someone
>>>>>has the game, or can find it, I might be able to find where the fireworks
>>>>>started, or I'll ask Hsu as he probably remembers.  But this was another of
>>>>>those very deep combinations..  At the point where it happened *none* of us
>>>>>(including an IM and a bunch of good chess players) understood it until about
>>>>>20 plies had unfolded and suddenly Mike Valvo says "oho!  look here, at the
>>>>>end of this, *this* happens and this is crushing for black"...
>>>>>
>>>>>As I recall, this was more obvious once the punch line was found because it
>>>>>was a long sequence that was pretty forced by both sides, once it was started...
>>>>>
>>>>>anybody have the last ACM games?
>>>>
>>>>
>>>>[Event "24th ACM Computer Chess Championship"]
>>>>[Site "Cape May, NJ USA"]
>>>>[Date "1994.06.26"]
>>>>[Round "4"]
>>>>[White "Star Socrates"]
>>>>[Black "Deep Thought II"]
>>>>[Result "0-1"]
>>>>
>>>>1. e4 c5 2. Nc3 Nc6 3. Nge2 Nf6 4. d4 cxd4 5. Nxd4 d6 6. Bg5 e6 7. Qd2 a6 8.
>>>>O-O-O h6 9. Bf4 Bd7 10. Nxc6 Bxc6 11. f3 d5 12. Qe1 Bb4 13. a3 Ba5 14. Bd2 O-O
>>>>15. exd5 exd5 16. Bd3 Re8 17. Qh4 d4 18. Na2 Bxd2+ 19. Rxd2 a5 20. Bc4 b5 21.
>>>>Rxd4 Qe7 22. Bf1 Qe3+ 23. Rd2 b4 24. Qd4 bxa3 25. Qxe3 axb2+ 26. Kxb2 Rxe3 27.
>>>>Rd6 Rb8+ 28. Kc1 Ra3 29. Rxc6 Rxa2 30. g3 Ra1+ 31. Kd2 a4 32. Bg2 Rd8+ 33. Ke2
>>>>Rxh1 34. Bxh1 Ra8 35. Rb6 Nd5 36. Rd6 Nc3+ 37. Kd3 a3 38. Kxc3 a2 39. Rd1 a1=Q+
>>>>40. Rxa1 Rxa1 41. Bg2 Rg1 42. Bh3 Rh1 43. Bc8 Rxh2 44. g4 Rf2 45. Bb7 g6 46.
>>>>Kd3 h5 47. gxh5 gxh5 48. Be4 h4 49. Ke3 Rg2 50. Bf5 Rg5 51. Bh3 Rg3 52. Bf1 h3
>>>>53. Kf2 h2 54. Bg2 Rg7 55. f4 f5 56. Kf3 Kf7 57. Kf2 Rg4 58. Kf3 Ke7 59. Kf2
>>>>Rg8 60. Kf1 Kd6 61. Kf2 0-1
>>>
>>>Aren't other programs of today playing the same winning moves as Deep Thought
>>>here? Starting with 22. ... Qe3+ and ending with 31. ... a4(Rd8+ looks like
>>>a direct transposing of moves would be considered as good if followed by the
>>>a4 advance).
>>>
>>>Or is the point that Deep Thought ran these lines deeper, which I don't
>>>doubt that it did.
>>
>>the point here is that deep thought had a +2 (or so) score vs *socrates for
>>about 10 moves *before* *socrates saw they were losing.  At the point where
>>this started, DT said +2, *socrates said "about even" and the discussion was
>>on king safety.  Everyone was "assuming" that DT depended on piece/square
>>tables, which it did to a point, that reflected lots of material on the board.
>>But following it's PV showed that pieces were coming off right and left, but
>>the last part of the PV wasn't visible due to the hardware search on the end,
>>and (I suppose) some hash table stuff making recovering the PV difficult (I saw
>>this in my mtd(f) experiments for example).  But each move, * socrates said
>>"even" and deep thought said +2, until about 10 moves later when *socrates
>>started saying "not to even", "significantly worse than even" and finally
>>-2 and the game was busted wide open...
>>
>>It is certainly possible that the programs of today might play the same
>>moves.  But would they see they were winning?
>
>The program's I tried all say they are winning at move 22 black. They showed
>around +1 or better in a very short time. The only program that would not play
>22.. Qe3+ was Junior 5. It perfered b4 but still showed +1.
>

back up to move 17 and see what your programs say.  at this point, before
DT played d4 *socrates thought it was somewhat ahead.  After d5 was played,
DT was about 1 pawn different from what *Socrates saw...

after move 20 (b5) was played DT was well over +1....

Other notes here say that at white's move 33, the game was over...  The
interesting position then may be after white's move 20.  I don't get +1 here.
although I haven't yet run for many minutes...

However, in the words of Hsu, this was not as much about a single tactical
shot as it was of outplaying the opponent.  But the evals were so different
it caused a lot of interest...

more when I have some time...



>
>Hardly.  IE several programs
>>would play the c5 in the dt/CB game... but have no idea at all that they are
>>winning the game...  and varying anywhere along the way might miss the "win"
>>totally...
>>
>>I've asked Hsu if he remembers exactly where this started. No answer yet, but
>>he had previously said he'd be "doing something" for a few days, which might
>>involve a trip home over the Thanksgiving holidays or something...  once we
>>get info on where this started, we can begin to analyze to see what it would
>>take to see what they saw...
>>
>>
>>
>>>
>>>I still think that the best demonstration of the project's strength
>>>are games 2-5 of the second match with Kasparov. I'm still guessing
>>>that these games really wore Kasparov down. These were great games
>>>and games 3, 4 and 5 showed how extremely difficult it was
>>>for the best chess player on the planet to convert his advantages in these games
>>>to wins.



This page took 0 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.