Examining Factors in Medical Experts’ Advice Taking: Designing Advice that Overcome Algorithm Aversion

Despite computational algorithms outperforming humans in many tasks, algorithmic advice is adopted less than human advice (algorithm aversion). In healthcare, where AI systems outperform medical experts, medical experts’ algorithm aversion critically reduces healthcare quality. However, few studies examine medical experts on medical tasks. Our behavioral experiments revealed that, among the factors affecting advice use, the distance between the initial judgment and provided advice consistently demonstrated the strongest influence on both medical experts’ and laypeople’s advice use, both in statistical analyses and subjective evaluations. We thus proposed an individualized advice presentation method that adjusts advice according to the distance between initial judgments and provided advice. The method significantly improved medical doctors’ advice use in patient prognosis prediction. This study constitutes a foundational step toward overcoming algorithm aversion among medical experts.