p.2 A primitive scoring function would then be
material + (0.2 x mobility) = score
p.4 Shannon's paper was written in 1948 and published in 1950. He did not
describe an actual chess program but he did suggest many useful ideas which are still in use. He realized the necessity of
having a good scoring function (a chess master's most valuable asset is his ability to assess the merit of a position). He
also pointed out that a scoring function... could only be applied in quiescent positions.
p.31 In my opinion it is at least as important to teach programs the ideas behind the openings as it is
to teach them the openings themselves.
p.50 [Levy is commenting on a move just played in a game, 9...BxN] Normal
play for a chess program, doubling the opponent's pawns even though it requires giving up bishop for knight. Herein lies one
of the fundamental problems of computer chess - how to teach programs when particular heuristics apply and when they should
be considered less important than others... Most programs seem to believe that the disadvantage of doubled pawns is greater
than the advantage of bishop over knight. In fact this is generally the reverse of the case.
p.56 In this position [Levy is commenting on another game played between computers] the TREEFROG programmers
discovered that their program was not analysing the correct position (possibly a couple of White's most recent moves had been
incorrectly typed on TREEFROG's terminal). Since Black's position has been hopelessly lost for some time the programmers resigned
for their program.
p.63 I think I probably spent about three seconds thinking time on each move [during Levy's simultaneous
exhibition, where he played 12 machines at once], except for one or two occasions where more thought was required.
p.66 [Levy comments on how a computer program cannot accurately assess a position in a game he is playing
against a machine where his opponent is up a rook but cannot bring the piece into play] So I am a rook down but
Black's [the computer's] rook is out of play... Without the ability to conceptualize no program will ever play Grandmaster
chess.
p.76-77 [information about the chess program Chess 4.3] The depth of the last iteration is determined either
by a parameter setting or (as in a tournament) by an automatic time control mechanism. With the mechanism in use in a tournament,
[search] depths range from 3 to 5 ply, in opening and middle game, up to 10 ply in the endgame... On the average, the program
scores about 300 positions per CPU second on a CDC 6400.
p.84 There are several groups in Europe who are working on chess programs, some of which are investigating
extremely unusual ideas. In England, for example, there is an attempt being made to represent the game of chess in 63-dimensional
space [JLJ - perhaps they got their idea to work, and are relaxing in the newly-discovered 27th dimension... at least now
I don't feel that my ideas are unusual.]
