Proceedings of the Annual Meeting of the Cognitive Science Society

Recent studies suggest that eye movements may be used to monitor task-specific mind wandering (MW) episodes during online learning. We examined whether we could replicate these lab-based findings in real-life virtual cognitive stimulation therapy (vCST) sessions using eye movements estimated from video by machine learning methods without eye trackers. We found that lower joint attention was a reliable indicator of MW for tasks involving well-defined strategies. For tasks involving well-learned visual routines, in contrast to previous studies, higher rather than lower eye movement consistency was associated with MW. This may be because real-life vCST sessions involved less structured discussions with more varied task demands, where lower eye movement consistency may reflect active engagement. Our results suggest the feasibility of using machine learning to monitor eye movements from video for MW detection, and raise the issue of generalizability from lab research to real-life scenarios due to potential differences in task demands.