COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub[let,jmc]" source
C00007 ENDMK
C⊗;
.require "let.pub[let,jmc]" source
∂AIL Professor M. M. Newborn↓School of Computer Science
McGill University↓Montreal, Quebec H3A 2K6↓CANADA∞

Dear Monte:

	Many thanks for the invitation to give away the awards.

	Let me take this occasion to make some suggestions concerning
the tournament.  I don't think they can be applied to the this year's,
but I would like to urge them for the future.

	It has always bothered me that the scores of computer chess
games are annotated just as though they were games between humans,
and there were no way of determining what the program was thinking
about.  This is particularly bothersome when the sponsor of the
contest is the ACM, which should be alert to opportunities to advance
computer science.

	Therefore, I would like to propose that it be a condition of
entry to the ACM tournament in 1980 that programs have some minimal
facilities for printing what they were thinking about.  I don't have
recent experience with chess programs, but I would like to suggest
some facilities that we used in an ancient program for the game
of Kalah.  The program ran on a PDP-1 computer that had only 4096
18 bit words, so costly analysis features could not be used.  I
think these proposed facilities are well within the capability of any of
the recent programmers of chess.

	After each move the program printed the following:

1. The first two plies of the move tree examined.

2. From each end point of the first two plies, the subsequent
principal variation to the maximum depth of search.  (This was
sometimes distorted by alpha-beta cutoffs).

3. At the end of each such variation, the value of the endpoint,
also sometime distorted by cutoffs,
and the amount of effort that went into the subtree.  The effort
was the number of positions examined, but only relative values
are wanted, so any such measure, e.g. computer time would also
be acceptable.

4. It would also be worthwhile to print the values of %2alpha%1
and %2beta%1 with which each of the variations was entered.

5. It would be further worthwhile if the programs stored the values
of the parameters used to make each move, so that the move
could be re-evaluated for the benefit of a commentator, who could,
for example, ask whether a particular variation was considered.

	The ACM might not require that this information be printed with
the move although there would probably be time to print it while
the opposing program was thinking about its next move.

	In my opinion, imposing some such requirements on entries to the
ACM tournament would raise the scientific level of the whole enterprise,
would educate the public to the fact that we really do know something
about how these things work and to the differences between human thought
and current programs, and would lead to better programs by leading to more
informed criticism of existing programs.

.sgn

cc: Professor Ben Mittman, Computation Center, Northwestern University,
David Levy.