How trust shapes human reliance on AI decision support in healthcare: A simulation-based behavioral study

Artificial intelligence (AI) based decision support systems are increasingly used in high-stakes domains such as healthcare, making it important to understand how trust relates to the use of AI recommendations. Although trust can support effective human–AI collaboration, miscalibrated trust may contribute to inappropriate recommendation acceptance and reduced decision performance. This study examined associations among perceived AI performance, trust, AI recommendation acceptance, and decision accuracy under different levels of AI reliability. In a simulation-based experiment, 100 medical students completed clinical decision-making tasks with AI assistance under high- and low-reliability conditions. Perceived AI performance and trust were measured after each block, while recommendation acceptance and decision accuracy were recorded at the trial level. Mixed-effects models showed that perceived AI performance was positively associated with trust. Higher trust was associated with greater overall acceptance of AI recommendations and greater acceptance of incorrect recommendations. Conversely, higher trust was associated with lower odds of rejecting correct recommendations, indicating more frequent rejection at lower trust. Greater misalignment between normalized trust and objective AI reliability was associated with lower decision accuracy. These findings indicate that calibrated trust is important for effective AI-assisted decision making. They also support the development and evaluation of AI interfaces that communicate reliability and uncertainty to help users align trust with system capability.