Gaining Physiological Insight into Satisfaction with XAI Explanations: A Call for Research

The staggering performance of prediction models based on machine learning (ML) algorithms has led to a boom in research interest in their application, but also led to the question how black box algorithms arrive at their results. To open the black box, explainable AI (XAI) approaches have been developed, which create approximated models that make the results of ML black box models more transparent to human proponents. This transparency is important for a multitude of reasons (e.g., to trust automated decision making), but not all explanations are equally well received by human explainees. Thus far, explanation satisfaction is mainly measured through self-reports, and the application of neurophysiological measures in this specific context is widely lacking. We review the existing research and make suggestions for future research directions, calling for NeuroIS research into measurement approaches that could be applied in this domain.