In this paper we present a probabilistic approach to the Human State Problem (HSP). In HSP a robot with a set of sensors, actuators and a set of intelligent computational resources has for task to provide the user with such behavior as to maximize the user's happiness. We formalize the HSP as a Hidden Markov Chain and analytically provide a solution that is the base for the proposed algorithmic solution. We also describe the mechanism called Adaptive Functional-Module Selection (AFMS) as a method of controlling the robot agent. The AFMS is shown to be controlled by a probabilistic method as described in an example. Finally a machine learning approach is presented as a realistic solution to the HSP problem.