Path: utzoo!mnetor!tmsoft!torsqnt!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!decwrl!pa.dec.com!shlump.nac.dec.com!merlyn!billmers
From: billmers@merlyn.enet.dec.com (Meyer Billmers)
Newsgroups: comp.ai
Subject: Re: Computer chess
Message-ID: <20223@shlump.nac.dec.com>
Date: 14 Feb 91 13:56:06 GMT
References: <26892@uflorida.cis.ufl.EDU>
Sender: newsdaemon@shlump.nac.dec.com
Reply-To: billmers@merlyn.enet.dec.com (Meyer Billmers)
Organization: Digital Equipment Corporation
Lines: 56

> Path: shlump.nac.dec.com!rust.zso.dec.com!pa.dec.com!decwrl!sdd.hp.com!uakari.primate.wisc.edu!samsung!caen!uflorida!webb.psych.ufl.edu!turner
> From: turner@webb.psych.ufl.edu (Carl Turner)
> Newsgroups: comp.ai
> Subject: Computer chess
> Message-ID: <26892@uflorida.cis.ufl.EDU>
> Date: 13 Feb 91 18:59:07 GMT
> Sender: news@uflorida.cis.ufl.EDU
> Reply-To: turner@webb.psych.ufl.edu (Carl Turner)
> Organization: University of Florida Psychology Department
> Lines: 17

> 
> What is the current level of expertise in the best chess playing machines?
> Is the current machine champion still at CMU?  And (for anyone familiar
> with the current chess efforts) how would you characterize the emphasis
> in making machines play better chess--faster machines, better algorithms
> to reduce search, storing patterns of "book moves," etc?

Well, I'm not sure what's the current champion in an all-computer tournament,
but as far as the chess world is concerned, Deep Thought (which comes from
CMU) is the most serious palyer. There have been improvements in search
algorithms and such, and research done by people like Berliner and his students
in chunking to improve analysis. These are interesting: some years ago, a 
chunking algorithm found a win in an endgame line that was previously believed
to be a draw in all the endgame literature. But interesting as this stuff is,
it mostly remains research. Most improvements in playing ability have come
from increases in processor speed. It is estimated that every doubling of
compute power increases rating by 200 points, and some extrapolate that this
will happen every year to 18 months. 

> 
> The reason for my asking: in a recent AI class several people insisted
> that machines are playing at "grand master" level.  Is this the case?
> 

Deep thought certainly is playing at or around GM level. At the US Open in
Boston a few (2 ?) years ago, it pulled a major upset against an International
Master, who proceeded to withdraw from the tourney. It then played a
Grandmaster,
and though it lost, it had at least a draw in the opinion of it's opponent,
except that a bug in its evaluation algorith (since fixed) caused it to miss
the correct line.

Since then, DT has play both Karpov and Kasparov. Though it lost to each, 
Karpov had to work hard enough to win that he stormed out of the room
afterwards,
refusing to talk to press or spectators. Kasparov won more easily, and declared
that computers would never be able to beat the likes of him. Most observers
don't believe this, though, unless they have egos of like size. 

DT also commented on some of the games in the recent K-K match, finding some
interesting lines that it claimed were a win for one side or the other (but 
which the players missed). Though there's technical debate about the
correctness
of DT's analysis, there's no doubt that it is now "approximately" a GM.
I think it's rating is over 2500.