Learning with Fuzzy Definitions of Goals

This paper explains a method to learn from fuzzy definitions of goals. This method has been applied to learn strategic rules in the Game of Go and decision rules for the management of a firm. The learning algorithm uses a representation of knowledge mainly based on the predicate logic. The goal of this paper is to extend this method of learning to systems using fuzzy logic. It is not useful to have gradual knowledge in order to learn tactical knowledge, but it becomes necessary when learning strategic knowledge. Strategic knowledge is knowledge about long term and global goals, it is fuzzy by nature. I give a method to learn using explanations of how the achievement of a gradual goal has been influenced by an action. This method is supported by an example of strategic learning in the game of Go. I show how this method can be applied in complex domains.