Computer Chess Club Archives


Search

Terms

Messages

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

Author: Robert Hyatt

Date: 17:12:02 01/05/98

Go up one level in this thread


On January 05, 1998 at 19:56:33, Bruce Moreland wrote:

>Augh.  I was hoping to avoid a "Deep Blue is awesome" thread as well.
>
>All I want to know why it played ... h5, since micro programs seem to
>find ... h3 pretty quickly.
>
>bruce


No idea.  there have been numerous cases of their having significant
bugs.
Remember that they were "new" in the parallel search arena, and believe
me, with parallel searches, *anything* can happen.  I remember them
having
bugs such as screwing up castling status so that they hated to castle,
or
wouldn't castle, or they would, on rare occasion, play the *worst*
possible
move...


Parallel search bugs are difficult to predict, difficult to detect, and
then nearly impossible to find.  I reported a while back about the worst
example I ever had.  We had just won the 1983 WCCC in new york, using a
really poor parallel search algorithm on a two-cpu Cray XMP.  That
Summer,
Cray told us they would have a 4 cpu machine ready for the 1984 ACM
event.
I requested time for a Labor-Day weekend tournament in Mobile, Alabama,
to
serve as a tune-up/test for my latest (at the time) parallel search
code.
In the first game, Cray Blitz played a move BxP check and mate.  Our
opponent
promptly took the bishop and we lost.  Turns out there were three legal
moves
and we had 4 processors.  Deep in the tree, when searching the replies
to
BxP+, 3 processors found good moves to search, the 4th said "I'm mated"
and
this erroneously got backed up to the root.

That happened after running the entire win at chess suite with no
problems,
plus playing several blitz games to mate...  It happens.  It isn't
unexpected.
And it is *absolutely unpreventable*...



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.