Adapting training in real time: An empirical test of adaptive difficulty schedules

Adaptive Training (AT) has been shown to be an effective technique for training tasks in multiple domains. Despite the promise AT has shown as a training technique, researchers remain unsure of the specific qualities that improve learning. In this experiment, we examined how adaptation schedule affects the efficacy and efficiency of difficulty adaptation in computer-based training. We used Cognitive Load Theory to guide our predictions about performance gains. In the reported study, we hypothesized that an adaptation schedule that adapts more frequently would lead to superior performance. To test this, we examined two types of difficulty adaptation (i.e., Within-Adaptive & Between-Adaptive) schedules using an audio-visual change detection task over five 10-minute scenarios. The Within-Adaptive condition adapted difficulty throughout the scenario based on trainee performance in real time. The Between-Adaptive condition adapted difficulty of subsequent scenarios based on previous scenario performance. We compared these two conditions to a Control condition, which maintained a constant difficulty throughout the experiment. We identified performance benefits for the Within-Adaptive condition, particularly for individuals whose performance was initially poor. A closer examination of the results suggested that average difficulty was the driving factor for performance gains in the Between-Adaptive condition. The data support that a within-scenario adaptation schedule effectively manages cognitive load to facilitate learning gains.