Über das Lernverhalten des Automaten Robbi

Summary

Robbi is an ER 56 computer programmed to play a simple nXn (2≦n≦9) board game against human opponents or against itself, that is, to play both sides of the game. Of special interest in the playing program is the learning routine, which improves the performance of the system as it gains experience with the actual game. The efficiency of the machine-learning procedure combined with various tactics adopted by the players has been studied in some detail. The results of each run are tabulated, i.e. the number of plays and the machine-playing time necessary to achieve the criteria of learning, the number of losses, the number of corrections, and the number of board positions stored in the memory of the computer during a run together with their associated lists of possible alternatives, the number of which is reduced by those which have been proved to be incorrect responses. Both the efficiency of the learning system and the characteristic form of the learning curves depends on the combination of tactics used by the players.