A commonly held idea is that people engaged in guessing tasks try to detect sequential dependencies between the occurring events and behave accordingly. For instance, previous accounts of the popular Rock Paper Scissors game assume that people try to anticipate the move an opponent is likely to make and play a move capable of beating it. In the paper we propose that players modulate their behavior by reacting to the effects it produces on the environment, i.e., that they behave exactly as they do in non competitive situations. We present an experiment in which participants play against a computer controlled by different algorithms and develop a procedural model, based on the new ACT-R utility learning mechanism, that is able to replicate the participants' behavior in all the experimental conditions.