Previous: Preliminary Results
Up: Preliminary Results
Next: Chess
Previous Page: Preliminary Results
Next Page: Chess
We are comparing the performance of METAGAMER in checkers by
playing it against Chinook [\protect\citenameSchaeffer et al. ,
1991]. Chinook is
the world's strongest computer checkers player, and the second
strongest checkers player in general. As it is a highly optimised and
specialised program, it is not surprising that METAGAMER always
loses to it (on checkers, of course!). However, to get a baseline for
METAGAMER's performance relative to other possible programs when
playing against Chinook, we have evaluated the programs when given various handicaps
(number of men taken from Chinook at the start of the game).
The preliminary results from the experiments are that METAGAMER is around even to Chinook, when given a handicap of one man. This is compared to a deep-searching greedy material program which requires a handicap of 4 men, and a random player, which requires a handicap of 8. In fact, in the 1-man handicap positions, METAGAMER generally achieves what is technically a winning position, but it is unable to win against Chinook's defensive strategy of hiding in the double-corner.
On observation of METAGAMER's play of checkers, it was interesting to see that METAGAMER ``re-discovered'' the checkers strategy of not moving its back men until late in the game. It turned out that this strategy emerged from the promote-distance advisor, operating defencively instead of in its ``intended'' offensive function. In effect, METAGAMER realized from more general principles that by moving its back men, it made the promotion square more accessible to the opponent, thus increasing the opponent's value, and decreasing its own. Of course, as discussed in Section 1.1, we cannot make an unbiased claim about originality of the program, as this game was known to METAGAMER's designer beforehand, but this development can still be taken as an indication of the generality of the advisors across a variety of games.