...Pell
Parts of this work have been supported by RIACS, NASA Ames Research Center [FIA], the American Friends of Cambridge University, Trinity College (Cambridge), and the Cambridge University Computer Laboratory.

...plays.
Instead of playing a game, the program now plays a game about game-playing, hence the name meta-game playing.

...power.
Recall from Section 1.3 that pieces in this class may move and capture in different ways.

...possessed.
This analysis would have to be improved substantially to capture the complexities of possessed pieces in games like Shogi.

...instead.
If we take the sum instead of the maximum here, a piece with many promotion options could be more valuable than its combined options, and thus it would never be desirable to promote it. For example, a pawn on the seventh rank would sit there forever enjoying all its options, but never caching them in.

...position.
A more sophisticated version of this feature, not fully implemented yet, takes into account the value of each victim, as determined by other static advisors.

...goals.
The net effect of this is that pieces which help a player to achieve arrival goals receive positive points, and those which help the opponent receive negative points.

...below.
Thus if a piece gets one point for each piece it can possibly capture, and there are 5 distinct piece names, it is possible to score 10 points if the piece can capture all pieces.

...1.
It should be noted that this is not the same as setting all piece values for a given game to equal value. The general knowledge still imposes constraints on the relative values of individual pieces in a game. For example, even a random setting of weights will cause METAGAMER to value queens above rooks [\protect\citenamePell, 1993b].

...knowledge,
I am thankful to Nick Flann for serving as a checkers expert.

...Chinook,
In our experiments, Chinook played on its easiest level. It also played without access to its opening book or endgame database, although it is unlikely that the experimental results would have been much altered had it been using them.

...level,
In the experiments, GnuChess played on level 1 with depth 1. This means it searches 1-ply in general but can still search deeply in quiescence search. METAGAMER played with one minute per move, and occasionally searched into the second-ply.

pell@ri.arc.nasa.gov
Thu Jan 6 15:54:24 PST 1994