A Search Strategy For Tutoring in Game Playing

In this paper we introduce some search strategies under the general term SST (Search Strategy for Tutoring) intended to be used in tutoring games. Contrary to principles of tutoring game-playing programs stated elsewhere, we argue that tutoring systems not necessarily should interrupt games or provide explicit advice. In addition, unlike earlier tutoring strategies , SST does not depend on a prior model of the opponent's strength and knowledge. After an intuitive example in TicTacToe we formalize the idea of SST and give three types of SST strategies, based on minimum and maximum loss limits. Moreover, depending on the opponent's response to a move, a meta-SST strategy is deened for each of the three SST strategies. The SST strategies are then illustrated by easy-to-implement algorithms, which are exempli-ed by a sample search tree. The main conclusion is that SST provides a vehicle for realization of hidden tutoring in game-playing programs, adapting automatically to the user's strength.