Author: Vincent Diepeveen
Date: 08:27:04 08/15/00
Go up one level in this thread
On August 15, 2000 at 11:11:10, Matthew wrote:
>On August 14, 2000 at 22:16:24, Uri Blass wrote:
>
>>On August 14, 2000 at 21:16:55, Martin Grabriel wrote:
>>
>>>Fritz
>>>Tiger
>>>Nimzo
>>>Rebel
>>>Junior
>>>
>>>My pick of top 5 (in that order)...
>>
>>Did you consider the fact that Crafty has a better hardware(probably at least
>>twice faster than the opponents)?
>>
>>I guess that Crafty will be in the top 5 but I do not guess which program will
>>be number 1.
>>
>>Uri
>Hi, others here might be more than I do, but it seems to me that unless all the
>programs are played against each other using exactly the same
>configuration/hardware, the results of the programs playing each other proves
>nothing (except that one person can afford a better computer).
> It brings to mind the argument I heard the other day in a chess news group
>about whether Deep Blue's program would be able to beat fritz6. My question
>is...would the program win or the hardware? To test whether deep blue's PROGRAM
>could beat fritz, both would have to be on the same hardware..imagine if fritz6
>was able to use the hardware that Deep Blues program had at it's disposal! Matt
to answer your questoin about deep blue: there are log files. You can see
them. Deep Blue searched between 11 and 13 ply. Considering it extended
all checks and other things getting 11 to 13 ply fullwidth is not bad
with 480 processors that all do not communicate with each other and
do not have hashtables at all and all do a 6 ply search.
So 11 to 13 ply was not bad.
However at nowadays fast cpu's and duals with shared memory you can
easily search way deeper than that, so if the deep blue lemma would be
true: that outsearching a program wins, you already win it on that.
Secondly, deep blue showed pathetic understanding of pawn structures. I
do not understand why kasparov didn't use that in the games against
deep blue.
Friedel sent me a cd he made about it. There a press conference a
reporter asked: "what if you lose Mr. Kasparov?". Big laughter in the
press room. Kasparov clearly not serious answers the question.
AFter the first game where deep blue made many bad positional moves
h6? ==> real bad only 2 programs or so, not surprisingly Zarkov that's
probably having a very similar evaluation to Deep Blue plays many
mistakes deep blue also played
Qa5?
Bc7?
g5?
g4?
That all in game 1. So the pawn structure understanding of Deep Blue and
the overall understanding of deep blue was real bad.
Seirawan very clearly writes something that's probably very true
to explain Qa5? : "deep blue probably counts the number of squares
the queen occupies". That is not by accident also something that's in
Zarkov/Gnuchess (4.0, the raped bitboard version i don't know anything
about) where all squares are counted with a mobility function.
So based upon evaluation function Deep Blue doesn't make much chances
either.
Now we didn't talk about book yet, as that's something i assume which
was OK in deep blue as a GM made it.
However many books made by strong players and many opening preparations
that work against humans, usual do not work against computers.
Great example is the game Diep-Nimzo in dutch championship which
is supposed to be playable against humans with white, but practical is
dead lost for white if black is a computer as the only compensation you
have is tactical compensation, and that doesn't work against a
program.
So there are 2 terrains at which Deep Blue would get clearly destroyed,
the book thing is something which is less important as you can force always
programs to play with the same book.
Greetings,
Vincent
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.