Learning to Control Dynamic Systems: Information Utilization and Future Planning

Dynamic systems involve states that change both autonomously and as a consequence of the learner’s actions. Research has shown surprisingly poor performance when people learn to control dynamic systems. Many researchers have proposed that learners often misperceive the feedback provided by the dynamic system, although it is still unclear how the feedback is misperceived and what can be done to improve learning. In two experiments, we studied learning behavior in a dynamic system called the beer game. We found that performance did improve through repeated practice, but subjects had a strong tendency to ignore the temporal dynamics in the system. Concurrent verbal reports indicated that performance improved mostly through better utilization of information related to the temporal dynamics of the system. As a consequence, subjects learned to anticipate changes in the system. In the second experiment, we provided only information that was critical for their decisions and found that initial performance was significantly better, indicating faster learning. It is concluded that poor utilization of information that is critical to the temporal dynamics of the system and insufficient anticipation of system changes are the major problems for learning in dynamic systems