Modeling of anticipation using instance-based learning: application to automation surprise in aviation using passive BCI and eye-tracking data

Human-centered artificial intelligence (HCAI) needs to be able to adapt to anticipated user behavior. We argue that the anticipation capabilities required for HCAI adaptation can be modeled best with the help of a cognitive architecture. This paper introduces an ACT-R cognitive model that uses instance-based learning to observe and learn situations and actions in the form of mental models. These mental models enable the anticipation of the behavior of individual users. The model is applied to a use case of automation surprise in commercial aviation to test how anticipation can best be modeled for cockpit applications. Empirical data from a flight simulator study including behavioral, neurophysiological and eye-tracking measures from 13 pilots were used to evaluate the model. Results show that the accuracy of the model is significantly higher than chance, demonstrating that combining context information, user state data and a cognitive model can enable HCAI adaptation based on anticipated user behavior.