Reasons for the effects of bounded look-ahead search

Some light is shed on the issue of why bounded look-ahead search can be beneficial. While this question has received some attention in the context of two-player perfect-information games, the authors also consider it for single-agent problem solving. Primarily, they focus on what makes minimaxing this useful in game-playing practice (especially in computer chess). They investigate a class of models based on domain-independent definitions of quiescence, observing more and more realistic behavior with more and more realistic definitions. As a global result, these models show beneficial behavior based on the specific properties of the tree and especially without the need for improvements of the evaluation function toward the end of the game or even error-free evaluations. In contrast, the model investigated for single-agent problem solving needs improved accuracy of the static values with increasing depth to make look-ahead beneficial