How users adjust trust in automation

Abstract

The present study examined how users adjusted their trust towards an automated decision aid. Results revealed that a valid recommendation of the decision aid increases whereas an invalid one reduces trust in automation. The magnitude of trust decrement is greater than that of trust increment. More importantly, this study showed that trust adjustment is not benchmarked strictly against predetermined objective criteria, that is, the decision aid’s recommendation quality. Rather, users’ ability of performing a task themselves and final task outcomes moderate the effects of recommendation quality. A valid recommendation is less appreciated if users are more capable of completing a task by themselves. An invalid recommendation is less penalized if the final task performance is not harmed, as if the invalid recommendation is “forgiven” to a certain degree.