In the Rough: Evaluation of Convergence Across Trust Assessment Techniques Using an Autonomous Golf Cart

Abstract

As automated and autonomous systems become more widely available, the ability to integrate them into environments seamlessly becomes more important. One cognitive construct that can predict the use, misuse, and disuse of automated and autonomous systems is trust that a user has in the system. The literature has explored not only the predictive nature of trust but also the ways in which it can be evaluated. As a result, various measures, such as physiological and behavioral measures, have been proposed as ways to evaluate trust in real-time. However, inherent differences in the measurement approaches (e.g., task dependencies and timescales) raise questions about whether the use of these approaches will converge upon each other. If they do, then the selection of any given proven approach to trust assessment may not matter. However, if they do not converge, it raises questions about the ability of these measures to assess trust equally and whether discrepancies are attributable to discriminant validity or other factors. The present study used various trust assessment techniques for passengers in a self-driving golf-cart. We find little to no convergence across measures, raising questions that need to be addressed in future research.