An analysis of decision quality of minimaxing vs. product propagation

The minimaxing approach to backing-up heuristic values has been very successful, e.g., in computer chess for making move decisions on the world-champion level, by employing very deep searches and effective pruning algorithms. From a theoretical point of view, however, it is not yet clear whether minimaxing or product propagation is better in terms of decision quality without deep searches. We present a systematic analysis of game trees with depth 2, where a single application of the competing back-up rules each (in a given branch from the root of the search tree) reveals their pure decision quality. Interestingly, product propagation tends to make better decisions per se more frequently, under realistic assumptions modeled after real game-playing programs. So, its decision quality may still make it a viable alternative for game trees where only shallow searches are affordable.