Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/17/84; site hplabsb.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!houxm!vax135!cornell!uw-beaver!tektronix!hplabsb!bl
From: bl@hplabsb.UUCP
Newsgroups: net.games.chess
Subject: Re: computer chess
Message-ID: <3216@hplabsb.UUCP>
Date: Thu, 2-Jan-86 13:13:12 EST
Article-I.D.: hplabsb.3216
Posted: Thu Jan  2 13:13:12 1986
Date-Received: Sat, 4-Jan-86 04:59:12 EST
References: <3199@hplabsb.UUCP> <623@oakhill.UUCP> <1063@cvl.UUCP>
Distribution: net
Organization: Hewlett Packard Labs, Palo Alto CA
Lines: 45

> In article <623@oakhill.UUCP> davet@oakhill.UUCP (Dave Trissel) writes:
> >In article <3199@hplabsb.UUCP> bl@hplabsb.UUCP (Bruce T. Lowerre) writes:
> >>In case you haven't seen it, there was an article in the October
> >>28 issue of Time about the latest computer chess machine, Hitech.
> >>
> >>The machine is currently rated at about 2250 which is master level
> >>but Hans feels its actual rating is about 2500.  The machine will
> >>be entered in more turnaments to determine its actual rating and
> >>to find the level of player that can consistently beat it.  In
> >>the mean time, its weaknesses are being patched.
> >>
> >
> >I find the 2500 claim a little hard to believe.  As the ratings get higher
> >the more chess knowlege is required to be embedded in the program.  (Just
> >examine how large the opening libraries have grown from year to year in the
> >best programs.)  And, after leaving opening book the more chess knowlege
> >embedded in the evaluation algorithms the slower which means fewer moves
> >examined which leads to weak tactical play blundering.
> >
> >One obvious method of attack is to so specialize the program that it has
> >a well defined goal decided at the time of exit from the opening book.
> >But this would take a substantial effort and the results would still be
> >inflexible.  Anyone know any details of Hans system?
> >
> >Skeptically - Dave Trissel  {ihnp4,seismo}!ut-sally!im4u!oakhill!davet

The evaluation of positions after each move could be expensive if one
were to include a large amount of chess knowledge in the evaluation.
However, Hitech, like its predecessor Tech, uses a very simple evaluation
metric, material gain or loss.  However, with this type of evaluation, the
program would be willing to play P-KR4 just as much as P-K4 for the opening
move.  Thus, the "chess knowledge" is put in to sort, or prioritise, the
moves at the top of the search tree and the search is used to evaluate the
tactical feasibility of the ordered set of moves.  The kind of "chess
knowledge" used depends on the position, i.e., opening book, middle game,
king safety, end game, etc.

> 
> 
> On the other hand, who would know better than a rated GM...  Think
> Hitech beats Berliner at games?

I asked Hans how he plays against Hitech.  Since he knows how the machine
plays, he has an advantage.  However, he says he played one game with it
and drew.