The absent P3a. Performance monitoring ERPs differentiate trust in humans and autonomous systems

Abstract To address suggestions that human brain responses to autonomous system errors may be used as brain-based measures of trust in automation, the present study asked participants to monitor the performance of either a virtual human or an autonomous system partner performing a novel, complex, real-world image classification task. We predicted visual feedback of partner errors would elicit the feedback-related negativity and P3 ERP components, and that these components would differ between the human and system groups. Behavioural results showed that while participants calibrated their trust in their partner according to our intended manipulation of error rates, no group differences were found. The ERP data, however, revealed FRN and P3 effects for both groups, modulated by accuracy and error rate. An unexpected finding was that the P3 topography differed between groups, with both a frontal P3a and posterior P3b component seen for the human condition, while for the system condition only the posterior P3b was observed and the P3a was completely absent. We suggest that this selective absence of the P3a may reflect reduced frontal attention during system monitoring in passive task conditions potentially resulting from reduced social and emotional processing for the system partner. This study demonstrates the potential for EEG-based measures of trust in automation to surpass the sensitivity of traditional measures of trust, while additionally uncovering a potential neural signature of automation complacency in the absent P3a. This identifies potential boundary conditions under which the application of human correlates of performance monitoring may not apply to the monitoring of an automated system.