Accuracy is not enough: The roles of accuracy, reliability and personality in the trust process

Artificial intelligence and machine learning (AI/ML) systems are increasingly integrated into human-machine systems, making it essential to understand how specific performance characteristics shape user trust and interaction behaviors. Although system performance is known to influence trust, research has largely conflated accuracy and reliability, limiting guidance for applied system design. The present study experimentally examined the independent effects of AI/ML accuracy and reliability on trustworthiness perceptions, trust, and monitoring behavior in an image classification task. In addition, individual differences in propensity to trust technology (PTT) and risk aversion (RA) were examined to better understand variability in user responses over time. Changes in trustworthiness perceptions, trust, and monitoring behavior were modeled using growth analyses. Results demonstrated accuracy and reliability exerted distinct effects on user perceptions and behavior. Higher accuracy primarily influenced trust and trustworthiness perceptions, whereas reliability was more strongly associated with monitoring behavior over time. PTT predicted initial trust perceptions prior to task experience, whereas RA moderated changes in trust and trustworthiness as users accumulated performance information. Findings demonstrate accuracy and reliability are separable performance properties with different implications for user trust calibration and monitoring behavior.