Author: Vincent Diepeveen
Date: 06:46:36 10/11/02
Go up one level in this thread
On October 08, 2002 at 20:59:39, Gregor Overney wrote:
>Well, it seems that Kramnik controls this PC program and, should he continue to
>perform at his current level, Kramnik will most certainly win.
>
>The "only" disturbing thing is that now some folks will indeed proclaim that the
>human brain outperforms any computer system the human brain is capable of
>building. The answers is no. Of course not! (See the game of 1997 as an
>example.)
>
>Although some journalists try to manipulate the opinion of the general public
>that Deep Blue and the program Fritz running on a multiprocessor PC are indeed
>comparable in strength, for us it is very obvious that this cannot be the case.
>It is easier to find flaws in certain evaluation functions than to outthink
>brute force searches with decent algorithms (such as implemented in Deep Blue).
>
>In any case, these games proof that PC programs are not yet winning against top
>players at standard time settings.
>
>Gregor
I hope you realize it is utter nonsense
a) openingsbook of fritz is a few hundreds of thousands of hand tuned
moves
b) in middlegame fritz gets like 15-17 ply, versus deep blue 12.2
c) deep blue was a preprocessor, fritz isn't
d) Fritz is having a lot of knowledge which deep blue doesn't.
Let's just give the classical doubled pawn on g2,g3 which deep blue
never understood well to be a great thing to have.
e) Fritz is 100x stronger in endgame than deep blue, that kramnik is
even better than that says something about kramnik
f) Even knowing something about programming doesn't say you know something
about computerchess. In fact even most chessprogrammers who feel very
well the problems DB had and solved in a way which was acceptible in
1997, but not in 2002; from those chessprogrammers majority doesn't even
see the pathetic setup it had with regard to parallellism.
g) the focus of deep blue in 1997, to quote its marketing department was
not anything of above, but the number of nodes a second only. Deep blue
DID get a lot of nodes a second. This was the ONLY important thing in 1997
for its marketing department. It seems you have forgotten this
h) deep blue only played a few pathetic games against kasparov. Now again
kramnik we see play random lines within his own opening. Completely
lost lines kramnik plays. However kasparov has done an even worse job
in 1997. First game he just plays using 3 rows. Normally completely
suicidal. He was asking for a zero there. But didn't get it at all.
Second game he again plays 10 blunders in an opening he doesn't know.
So he resigned, of course to his horror to find out that he could still
draw it. third,4th and 5th game are complete amateur level. I would be
ashamed playing such games. I'm just FM. 6th game he plays a kind of
anti-chess which is too pathetic for words. Note i have a book written
by kasparov and shakarov here which says this line kasparov plays there
is completely lost for black. The book was written years before the
match.
So based upon 6 pathetic games from kasparov, where only 1, game 6,
decided the match, you conclude that deep blue was ok?
i) The last so many years when ordering big numbers of elements, many
new algorithms have improved upon older ones. I bet bubblesort
was one of the first to use. Then obviously merge sort and quicksort
and others came in.
Now imagine that there were great algorithmic improvements in computer
chess, which also
were known in 1990 for computerchess. Killer moves (invented around
1975?) and hashtables (definitely proven to be working very well
around 1984?).
Hashtables give a few ply more (depending upon search depth) and
killermoves 20-40% node reduction. Yet Deep Blue didn't use any of
that in its hardware search. Isn't that pathetic?
Of course i have to add 2 notes. When you use them, your nodes a
second drops. Of course that note isn't proving that the real reason
to not use these *trivial* algorithms, was that the IBM directives
said that the only holy thing was a higher nodes a second...
I could fill another entire list of things DB didn't do and DB did do
which we see as dubious now. but the above 2 factors definitely were
well known in 1997 as well...
Best regards,
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.