Effects of Failure Types on Trust Repairs in Human-Robot Interactions

System performance is the central determinant of user trust in human-machine communication; however, performance failure is inevitable. This study develops a three-fold typology of performance failures (i.e., logic, semantic, and syntax) commonly observed in human-robot interactions based on the differences between the expected and actual outcomes. Herein, 1027 observations are collected from an online experiment to elucidate how the three types of failure and four repair methods (namely, internal attribution apology, external attribution apology, denial, and no repair) impact user trust while examining blame attributions as an underlying mechanism. The results reveal that despite some similarities, the interactions between trust violation types and repair methods differ in robot-to-human trust repair from those in human-to-human trust repair, which contradicts previous findings. Logic failures are found to be the most detrimental category of performance failures, and the internal-attribution apology is the optimal repair strategy. Notably, participants report greater levels of competence-based trust beliefs if they believe that the situation is jointly controlled by the human interactant and robot.

Automated summary

System performance is critical in determining user trust in human-machine communication. This study develops a typology of performance failures in human-robot interactions and investigates how they impact user trust, with a focus on blame attributions. The results highlight the importance of logic failures and internal attribution apology as effective repair strategies. Participants also report higher levels of competence-based trust beliefs when they perceive joint control between humans and robots.