[extropy-chat] computer chess again

[Alejandro Dubrovsky]

On Sun, 2003-11-09 at 16:32, Spike wrote:

> Perception management on the part of IBM after
> the Kasparov-Deep Blue match of 1997 has caused
> many people to believe that chess computers are
> currently stronger than the best humans, however
> since 1999, there have been 7 matches between
> computers and human players with Elo ratings 
> over 2700 (the top 10 to 15 human players).
> 
> Remarkably all seven of these matches have ended
> in a tie score.  These seven matches represent
> a total of 49 games.  The fact that the top humans 
> and computers are dead even over all that time
> suggest that the humans are getting better at
> exactly the same rate as the top chess software.
> Perhaps someone can suggest a different explanation
> for the even score since 1999.  
> 
I'll try.  First, the cutoff at 2700 is completely arbitrary, and if
games vs slightly lower ranked players is included, the results are
nowhere near as neat. This is then, i think, a statistical fluke.  But
more importantly, the games won by computers and the games won by GMs
are different in nature.  Computers win by GMs blundering, or by the
game being open and tactical.  GMs otoh, squeeze computers to death.  (A
few exceptions on both sides, but great majority adheres to that).  What
i think this means is that a year 2000 program would have won and lost
the same games that the year 2003 program won or lost.  It then makes a
bit of sense for the GMs to be not getting much worse in matches since
they are mainly playing against themselves.  If they are tight and play
well, they'll win/draw, otherwise they'll lose, and they seem to be
making blunders and letting computers open the position up at about the
same rate than they used to.   Based on this, some predictions:  if
matches of these kind become more common (ie human GMs resign to the
fact that they will be beaten and play more often in public without
demanding six figure sums for the effort), then 1) some GMs will do much
better against computers than their ELO would suggest, and some will do
worse.  2) Kasparov will be part of this second group.
alejandro