Introduction. According to the analysis of grandmaster-like strategies in Shogi [Iida and Uiterwijk 1993], it is important for a teacher, at the beginning stages of teaching, to intentionally lose an occasional game against a novice opponent, or to play less than optimally in order to give the novice some prospects of winning, without this being noticed by the opponent. Such a strategy is called a tutoring strategy in game-tree search.

In this work we consider a loss-oriented search strategy (LO-search for short), by which a player attempts to lose a game, and we show an algorithm for LOsearch based on the minimax strategy and OM-search (opponent-model search; see [Iida et al. 1993]). We further describe characteristics of LO-search, including the concept of intentional error. We next discuss the situation in which a player will notice such an intentional error. We then describe a tutoring search (TUsearch for short) by which a player attempts to lose a game without this being noticed by the opponent.

LO-search is based on OM-search, since LO-search also takes the opponent model into account. OM-search is predicated on the assumption that one has complete knowledge of the opponent's strategy and that this strategy is completely contained in the opponent's evaluation function. In OM-search, two values are computed for all positions in a search tree. For clarity, the players are distinguished as a max player and a min player.

Loss-Oriented Search. In order to play games, the minimax strategy or variations thereof have been used by many game-playing programs.