TOWARD AN IDEAL TRAINER

This paper demonstrates how the nature of the opposition during traini ng affects learning to play two-person, perfect information board game s. ft considers different kinds of competitive training, the impact of trainer error, appropriate metrics for post-training performance meas urement, and the ways those metrics can be applied. The results sugges t that teaching a program by leading it repeatedly through the same re stricted paths, albeit high quality ones, is overly narrow preparation for the variations that appear in real-world experience. The results also demonstrate that variety introduced into training by random choic e is unreliable preparation, and that a program that directs its own t raining may overlook important situations. The results argue for a bro ad variety of training experience with play at many levels. This varie ty may either be inherent in the game or introduced deliberately into the training. Lesson and practice training, a blend of expert guidance and knowledge-based, self-directed elaboration, is shown to be partic ularly effective for learning during competition.