Proceedings of the Annual Meeting of the Cognitive Science Society

n the current study we present an air traffic control (ATC)task in which we measured pupil dilation to automaticallydetermine high and low workload periods. We manipulatedworking memory (WM) requirements across three conditions:a no WM condition, a passive WM condition in whichinformation was accumulated, and an active WM condition inwhich information had to be added to and removed from WM.Results showed that no WM resulted in the least dilation, butthat passive WM and active WM did not differ. Next, we usedthe pupil data to train a range of classifiers to differentiatebetween high and low workload periods with the ultimategoal to create an online task-independent interruptionmanagement system (IMS). The best predicting features werethe median and a second-order polynomial fit, going back 12seconds from the to-be-predicted moment. Using thesefeatures, our classifier was able to predict workload at highaccuracy (77%). We conclude that pupil dilation can be usedto create a reliable IMS.