Estimating a user model with the action sequences in the mind reading game between a human and a life-like agent

This paper proposes an adaptive interaction between a human and a life-like agent. We present the method to estimate the partner´s context with the nth order MDP model. In this framework, an agent plays a mind reading game with a user in which the agent estimates the user´s current mind state by the previous game context. The online identification of the model for mind transition rules of a user using the mind transition sequence and its usage in the iterated mind reading game between a user and the agent. From the experimental results, the properties of the mind transition rules of a user used in the game are consistent, individual, almost deterministic and partial.