Time Pressure Increases Automation Reliance in a Face Matching Task

Automated facial recognition (AFR) systems are commonly used to help verify the identity of individuals, often in a one-to-one face matching task. The AFR system may be a decision aid, such that a human operator must ultimately approve the final identification decision. Previous research has shown that time pressure, a common operational factor in many applied settings, impairs human face matching accuracy. We investigated whether time pressure influences human reliance on AFR decisions in a face matching task, predicting that participants would show greater automation reliance as time pressure increased and the task became harder. Each participant (n = 129) completed a face matching task under blocks of high (2s), medium (5s), and low time pressure (10s), where the stimuli disappeared from the screen after the allocated time. For each pair of faces, participants made an initial identification decision without assistance. Participants were then shown an identification decision from a simulated AFR system (92.3% accurate), before they were asked to submit their final (assisted) decision to the same trial. As predicted, human accuracy improved with AFR assistance in all time pressure conditions, with the greatest improvements occurring under higher time pressure. But replicating previous studies, average aided human accuracy was below that of the AFR system alone. Moreover, assisted accuracy fell in all conditions when the AFR system provided an incorrect decision, suggesting participants struggled to correct system errors. Our results have implications for human oversight of AFR systems in applied face identification scenarios.