December 21st, 2008
If the world's best computer chess program plays against itself, will
the advantage of having the first move allow white to win? Or will
every game be a draw?White will not always win; indeed, sometimes Black will win. White
would win more often than Black, however.
Even though one might think that a draw would be the inevitable
result, given that both sides in such a contest are ipso facto of
equal strength, in fact this is not the case. Typically computer
programs work by searching to some fixed depth, then using heuristics
to extend the depth to which they search by a certain amount, and then
finally using positional and material heuristics to evaluate the
quality of the resulting positions. This process has some probability
of returning an incorrect move evaluation, and thus there is some
probability that in any given position the computer will make the
wrong move. Once a computer reaches a position which is bad enough, it
will lose, even against itself.
These are general considerations, now let me be more precise:
The best current computer chess program is generally thought to be
Fritz 7. Fritz 7 is at the top of the Swedish computer chess rating
list at
http://w1.859.telia.com/~u85924109/ssdf/ . This is generally the most
respected computer chess rating list.
(It is true that the very strong Junior just one the Maastricht
computer olympiad ( http://www.cs.unimaas.nl/Olympiad2002/ ) .
A good site on computer chess is:
http://www.xs4all.nl/~verhelst/chess/programming.html .
Now for your specific question, the closest I could come was a
"schachduell", a chess contest, in a German site at:
http://www.heise.de/ct/schachduell/ .
In this site two strong programs, Fritz and Shredder, played each
other on slightly different hardware (an AMD chip and an INTEL chip).
The translation of the site is:
http://translate.google.com/translate?sourceid=navclient&hl=en&u=http%3A%2F%2Fwww%2Eheise%2Ede%2Fct%2Fschachduell%2F
by the way.
Anyway, I could not find a tournament crosstable on this site, but
there is a record of games, say:
http://www.chessbase.de/heise/games/heisematch.pgn .
If you look through the games, you can see that sometimes even when
Fritz plays itself, black does win. (I actually own Chessbase, so I
was using the similar format:
http://www.chessbase.de/heise/games/heise.cbv ) . For example, in game
503 Fritz running on an Intel chip beat Fritz running on an AMD chip
as black, but in game 504 Fritz AMD beat Fritz Intel!
So you see, sometimes White wins, sometimes Black, sometimes draw. It
depends on the opening and even on luck to some extent. Computer chess
programs are not infallible!
Although it does not apply directly to computer chess, a famous result
by Dana Nau on game tree pathology shows that sometimes the DEEPER you
search in a game tree, the WORSE you play! See
D.S. Nau. "How to Do Worse by Working Harder: The Nature of Pathology
on Game Trees." In Proc. 1983 IEEE International Conference on
Systems, Man, and Cybernetics, Dec., 1983 ,
or some comments at http://www.cs.umd.edu/~nau/papers/forward-aaai94.ps
(postscript format, text version at:
http://216.239.51.100/search?q=cache:0oHlL6iq1fQC:www.cs.umd.edu/~nau/papers/forward-aaai94.ps+dana+nau+game+tree+pathology&hl=en&ie=UTF-8)
Nau's home site is: http://www.cs.umd.edu/~nau/publications.html .
Search strategy:
As I got my Ph.D. in computer chess, I already knew most of these
links. Try http://www.chessbase.com or google search on computer chess
or swedish rating list. The best computer chess message board is:
http://www.talkchess.com although registration is required.
One interesting question that I am not sure of, and would be kind of
hard to know: what is the exact percentage of White wins, draws, and
Black wins? Here though, the answer will depend to a large extent on
things like the exact time control, the opening book used, and various
parameters of the programs. But be assured that yes, Black will win
sometimes (not as often as White though).#If you have any other info about this subject , Please add it free.# |
|
Posted in linkdatax.com | edit
