davet@oakhill.UUCP (Dave Trissel) (08/02/85)
Even though I have never seen the program I can make a good educated guess
since in the early days my own program made just such moves.
The p-r4 is a classic computer move generated by programs with simple
evaluation routines. Since the implications of the pawn move could not
be determined by a computer unless it could search 10 or more moves ahead
(20 ply in computer parlance) the computer fails to see this as a weakness.
Most programs search anywhere from 4 to 9 ply (2 to 5 moves) ahead. Since
there are no obvious attacks or defences to be had, the computer finds no
"important" move in its move list evaluation to take. When this happens
smaller values in the evaluation function tend to dominate, one being that
it's a generally good thing to move ones pieces forward.
Remember, the computer normally doesn't really "understand" a position. It
just computes a point value made up of things like piece mobility, pieces in
the center 4 squares, king attacks/defends, pawn chains unbroken and the like.
On that basis a short-sighted look at p-r4 looks like a viable move since a
later p-n3 will allow a pawn chain to be made and no piece is seen to be made
immediately endangered. This is no different than what frequently happens
with raw novice players.
I solved this problem with my own program early on by forcing penalities
in score evaluation depending on how many first-rank men are still in
place. This allowed full deployment of the more important pieces and only
then would pawns start to forge ahead. Yet another procedure, I have three
evaluation routines to weigh things differently for beginning, middle and
endgame situations. Originally pawn movements are given small bonuses in the
early game but progressively bigger bonuses in the middle and endgame
calculations.
I hope this simple analysis helps to elucidate some of the "thinking" behind
the more primitive chess computer programs.
-- Dave Trissel {seismo,gatech,inhp4,ctvax}!ut-sally!oakhill!davet