Practical Guidance for Evaluating Calibrated Trust

Abstract

As automation, autonomy, and AI become more prevalent, human factors engineers are called to evaluate whether users trust that automation. We argue that the true question is whether users trust the automation appropriately. In other words, do they trust it as much as it deserves to be trusted? When automation performance decreases, are users aware so they decrease their momentary trust, and more importantly, their reliance on the automation? There are few metrics that focus specifically on calibrated trust, and the trust literature can be daunting for human factors professionals who are not experts in trust. We offer two aids to human factors practitioners tasked with evaluating trust. The first is an easy-to-use calibrated trust framework that simplifies the aspects of trust into Belief, Understanding, Intent, and Reliance. The second is the introduction of Calibration Points, a way to classify situations in which the automation excels or situations in which the automation is degraded. By identifying these Calibration Points, human factors practitioners can evaluate whether human trust is aligned with automation performance. This approach allows the practitioner to leverage the rich set of evaluation techniques that have been developed to evaluate trust.