Author: Robert Hyatt
Date: 08:41:07 07/23/02
Go up one level in this thread
On July 23, 2002 at 10:55:24, Uri Blass wrote: >On July 23, 2002 at 10:33:15, Robert Hyatt wrote: > >>On July 23, 2002 at 09:41:10, Ed Schröder wrote: >> >>>On July 22, 2002 at 14:11:23, Robert Hyatt wrote: >>> >>>>On July 22, 2002 at 11:52:35, Ed Schröder wrote: >>>> >>>>>The two you entirely miss the point of Kasparov's suspicion, sigh... >>>>> >>>>>Lesson 61, now pay attention my pupils :) >>>>> >>>>>[d]r1r1q1k1/6p1/p2b1p1p/1p1PpP2/PPp5/2P4P/R1B2QP1/R5K1 w - - id DEEP BLUE - >>>>>Kasparov,G; >>>>> >>>>>Position before 36.axb5 >>>>> >>>>>Here DB for a long time showed the following main-line: >>>>> >>>>>36.Qb6 Qe7 37.axb5 Rab8 38.Qxa6 e4 39.Bxe4 Qe5 40.Bf3 Rcd8 41.Qa7 Qc3 42.Bh5 >>>>> >>>>>Now let's have a look at the main-line shall we? >>>>> >>>>>After 39...Qe5 we get: >>>>> >>>>>[d]1rr3k1/6p1/Q2b1p1p/1P1PqP2/1Pp1B3/2P4P/R5P1/R5K1 w - - >>>>> >>>>>As you can see black has sacrificed 3 (!!) pawns for a king attack. And this my >>>>>pupils is what Kasparov could not believe, a computer sacrificing 3 pawns and so >>>>>he started asking questions how that could be. >>>>> >>>>>The refusal of IBM to answer Kasparov's questions made Kasparov suspicious and >>>>>from one thing came another. >>>>> >>>>>This is *the* heart of the discussion. >>>>> >>>>>Ed >>>> >>>> >>>>So? If two _current_ programs think axb5 and Qb6 are equal, what do you >>>>conclude then? That we all have similar king safety? That we all like >>>>the same move (axb5) but for different reasons? >>> >>>Again, axb5 or Qb6 is irrelevant. What is relevant is the main-line. Black has >>>nothing for the 3 pawns. You don't need to be a GM to see that, you don't need >>>the complete main-variation, after 39...Qe5 the picture is clear that black has >>>nothing for the 3 pawns, see the above diagram. Yet the mainline is a very human >>>approach to establish a) a draw by eternal check or b) looking for counter play. >> >> >>Do you think sacrificing the three pawns is the _only_ way to proceed after >>axb5? Is fritz showing the same sacrifice in its PV after it gets the same >>score for both moves? If so, is fritz getting help too? Remember that >>Kasparov did _not_ get to see the partial PV from deep blue. And remember >>that we don't know what happens in the last N plies of the PV since we can't >>see them. Perhaps it found something different than tossing three pawns, >>because I don't believe Fritz would sac three pawns and say "equal to winning >>a pawn outright" myself. >> >> >> >> >>> >>>Kasparov *KNEW* about the mainline (just check your records), exactly *THE* >>>reason he started to ask questions. >> >>He knew about what _he_ saw. Not about what DB saw. He knew what fritz >>could see after a few minutes. Not about what it could see after weeks of >>computing (at that time, since it took almost three days at today's speed to >>get the two scores equal). >> >> >>> >>>Bob, this is an issue about chess, not computer chess, try to understand 36.Qb6 >>>Qe7 37.axb5 Rab8 38.Qxa6 e4 39.Bxe4 Qe5 as a chess player, not as a computer >>>chess programmer. >> >> >> >>Try to understand why programs are saying axb5 is _just_ as good... which >>Fritz did. Crafty got to within .1 after 2.5 days... >> >> >> >>> >>>I know you entirely want to blacken Kasparov for the public scandal but then you >>>should tell history as it happened, here is: >>> >>>1) Kasparov doesn't understand DB's main-line 36.Qb6 Qe7 37.axb5 Rab8 38.Qxa6 e4 >>>39.Bxe4 Qe5 and starts to ask questions how a computer who usually sits on every >>>pawn suddenly can sacrifice 3 pawns in a row. >> >>OK... but two points: (1) kasparov didn't see db's main line. The main line >>is incomplete in any case so what happens to the last N plies is anybody's >>guess. (2) fritz and Crafty are _both_ saying that axb5 and Qb6 are identical >>after several days of searching. That suggests that either fritz and crafty >>at DB speeds would produce the _same_ mystery, or there is something else going >>on. At least we can prove that his statement is false. > >The fact that they show the same score does not mean that they are going to >choose deeper blue's move because they need to see bigger score for axb5 >relative to Qb6 in order to choose axb5. With alpha/beta equal is good enough. That means that had axb5 started off better, it would have stayed better, because we prune on >=. It is also pretty clear that for both fritz and crafty _every_ iteration sees axb5 get better and Qb6 get worse. One more iteration might do the trick for crafty as it is now .08 worse (axb5)... Unfortunately I now get to start this test over as a bad battery and a power outage took the cluster down completely during the night. > >I also believe that deep fritz and crafty of today can see deeper than deeper >blue at the same speed thanks to null move pruning. I also believe they (fritz/crafty) make more mistakes. And I also believthat DB extends enough that it sees far deeper than we do at a given ply... six of one, half-dozen of another... > >Uri
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.