Continuous Calibration of Trust in Automated Systems

摘要

This report details three studies that have been conducted in order to explore user calibration of trust in automation. In the first, we discover that all-or-none thinking about automation reliability was associated with severe decreases in trust following an aid error, but high expectations for automation performance were not. In the second study, we examine predictors and outcomes of calibration of trust. We measured calibration in three different ways. We found that awareness of the aid's accuracy trajectory (whether it was getting more or less reliable over time) was a significant predictor of calibration. However, we found that none of the three measurements of calibration had strong associations with task performance or the ability to identify aid errors. We also describe the conceptual premise and design of our third and final study. This study examines the development, loss, and recovery of trust in a route planning aid in a military simulation context. The results of this study will be presented in our final report.