Previous: Bias when using Existing Games
Up: The Problem
Next: SCL-
Previous Page: Bias when using Existing Games
Next Page: SCL-

Metagame

The preceding discussion only serves to emphasize a point commonly acknowledged in AI: whenever we design a program to solve a set of problems, we build in our own bias about how the program should go about solving them. However, while we cannot eliminate bias totally from AI research (or any other for that matter), we can to some degree control for human bias, and thus eliminate a degree of it.

This motivates the idea of Meta-Game Playing [\protect\citenamePell, 1992a], shown schematically in Figure 2.

Rather than designing programs to play an existing game known in advance, we design programs to play (or themselves produce other programs to play) new games, from a well-defined class, taking as input only the rules of the games as output by an automatic game generator. As only the class of games is known in advance, a degree of human bias is eliminated, and playing programs are required to perform any game-specific optimisations without human assistance. In contrast with the discussion on existing games above, the human no longer mediates the relation between the program and the games it plays, instead she mediates the relation between the program and the class of games it plays.