Stochastic Modeling of Human Learning Behavior

Abstract

A stochastic model of human learning behavior in a manual control task is described. Regulation of the state of a double integral plant to minimize the integrated absolute error is the operator's task. Subjects given this task were instructed to drive the process from an initial state to the null state using a two-position relay controller and a visual display. A subject is conceptualized in the model as a sequential data-processing system. A sensor, a decision maker, and an effector are the three serially connected components making up the system. Each element requires a finite time to either process or transmit information, and thus a delay is incurred between the reception of the visual stimulus and the execution of a motor response. Response decisions are based on the a priori estimate of the probability that the control polarity should be switched, given the current state of the plant. Patterns in the resultant phase trajectory are used as evidence by the decision maker to revise the prior estimate with an algorithm according to Bayes' theorem. Behavior of this model is compared with subject behavior in the motor skill experiment, and the model's characterization of the time-varying random nature of human learning is brought out by this comparison. Also discussed are the applications of the concept of this model to other manual control tasks.