Previous: The Problem
Up: The Problem
Next: Metagame
Previous Page: The Problem
Next Page: Metagame

Bias when using Existing Games

However, even this generality-oriented research is subject to a potential methodological bias. As the human researchers know at the time of program-development which specific game or games the program will be tested on, it is possible that they import the results of their own understanding of the game directly into their program. In this case, it becomes difficult to determine whether the subsequent performance of the program is due to the general theory it implements, or merely to the insightful observations of its developer about the characteristics necessary for strong performance on this particular game. An instance of this problem is the fixed representation trick [\protect\citenameFlann and Dietterich, 1989], in which many developers of learning systems spend much of their time finding a representation of the game which will allow their systems to learn how to play it well.

This problem is seen more easily when computer game-playing with known games is viewed schematically, as in Figure 1.

Here, the human researcher or programmer is aware of the rules and specific knowledge for the game to be programmed, as well as the resource bounds within which the program must play. Given this information, the human then constructs a playing program to play that game (or at least an acceptable encoding of the rules if the program is already fairly general, like HOYLE [\protect\citenameEpstein, 1989b]). The program then plays in competition, and is modified based on the outcome of this experience, either by the human, or perhaps by itself in the case of experience-based learning programs. In all cases, what is significant about this picture is that the human stands in the centre, and mediates the relation between the program and the game it plays.