Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site alberta.UUCP
Path: utzoo!watmath!clyde!burl!mgnetp!ihnp4!alberta!jonathan
From: jonathan@alberta.UUCP (Jonathan Schaeffer)
Newsgroups: net.chess
Subject: Re: What makes a strong computer chess program?
Message-ID: <322@alberta.UUCP>
Date: Wed, 4-Jul-84 17:10:48 EDT
Article-I.D.: alberta.322
Posted: Wed Jul  4 17:10:48 1984
Date-Received: Thu, 5-Jul-84 00:38:10 EDT
References: <988@eosp1.UUCP>
Organization: U. of Alberta, Edmonton, AB
Lines: 57

Here are a few comments on Tobias Robison's remarks on what it
takes to have a really good chess program with emphasis on how
it applies to my program, Phoenix:

1) Opening book.  Agreed.  One cannot ignore a century of extensive
   analysis of the openings.  But from the programing point of view,
   opening analysis is purely mechanical - I get a friend to type
   in the latest analysis from the latest chess journals.  Phoenix
   does not have an extensive opening book (a shortage of friends
   who have the patience to type all that junk in).  On the other
   hand, Ken Thompson typed in approximately 350,000 moves from ECO
   for his chess machine Belle.  One can envision in the near future
   chess opening book databases in which ALL the latest theory will
   be available in machine readable form.

2) Determinism.  Chess programs should not be deterministic, but
   regrettably, most are.  In most programs (including Phoenix), non-
   determinism comes from two sources.  First, we can randomly select
   our opening line.  This offers some variety, but there is still the
   danger of repeating a game.  Second is the clock.  In tournament play,
   even if two games follow the exact same move sequence, the times on
   the clock will never be identical.  Differing times may mean the
   program will search more/less and possible result in a different move.

   There is actually a third form of non-determinism.  Rarely does a tour-
   nament game go by when I do not detect an inaccuracy in Phoenix and 
   fix it before the next round.   In that case, the next opponent is
   playing a slightly different program which will possibly make
   different moves.

3) Regarding open positions.  It is true that most programs play open
   positions well.  Blocked positions require more chess skill to play
   correctly.  However, with enough chess knowledge, there is no
   reason why programs cannot play such positions well.  Phoenix runs
   on a VAX and therefor has considerable difficulty being competitive
   with a Belle or Cray-Blitz, both of which run on much faster hardware.
   Phoenix uses knowledge to compensate for depth.  In fact, Phoenix
   avoids open games and prefers closed ones, contrary to most chess programs.
   I noticed Robison criticized Phoenix for playing a French defence,
   but the fact is that Phoenix has never lost the black side of a French
   in tournament play.

4) Variable depth.  Almost all brute-force chess programs I know search
   to variable depths.  For example, most programs extend the search tree
   an extra ply for every check encountered along a line of analysis.
   It is true that the program is still brute-force and not selective (such
   as the chess program Awit).  Given that you want to search 5 ply,
   Phoenix will search a line for a minimum of 4 and a maximum of 8 ply
   (plus extensions for capture sequences) depending on what "features"
   of the position it sees during its analysis.  Phoenix does have the
   advantage of variable depth and the advantages of brute-force search-
   ing (along with some of the disadvantages too I'm afraid).

In summary, I think your points are too general.  I can argue that all the
deficiencies you see in computer chess programs today do not really exist
for many programs.  As to what it will take to have a really strong program,
well thats another article...