Experience With AI Algorithms And Algorithm Aversion In AI-Supported Medical Care

Despite the growing deployment of AI-based decision support systems in professional settings, experts frequently override algorithmic recommendations, particularly under conditions of high uncertainty. Prior research offers mixed evidence on whether experience with AI-based algorithms mitigates such algorithm aversion and provides limited insight into how different forms of experience matter over time. We address this gap by examining how experience with an AI-based algorithm shapes experts’ acceptance of algorithmic advice across decision contexts characterized by varying levels of uncertainty. Leveraging a unique longitudinal dataset of more than 600,000 real-world treatment recommendations generated by a certified AI system used in nephrology care, we study nephrologists’ acceptance of algorithmic recommendations in therapeutic decisions involving substantial clinical discretion and risk. We distinguish between time-based experience and volume-based experience with the algorithm. Our analyses show that time-based experience—rather than the sheer volume of interactions— increases acceptance of algorithmic recommendations, and that this effect is concentrated in decision contexts marked by the highest uncertainty. We further show that time-based experience is associated with reduced therapeutic inertia and greater alignment with the algorithm’s treatment optimization logic, accompanied by improved clinical performance.