Trusting the Middle in the TNO Trust Task: How Medium Risk and Reliability Shape AI‐Assisted Advice‐Taking

As artificial intelligence (AI) becomes increasingly integrated into decision‐making processes, understanding how people trust and rely on AI versus human experts is essential. In particular, the complexities of medium‐level task risk and advice reliability require further exploration. Using the judge–advisor system (JAS) model, this study investigates the effects of advice source (human vs. AI), advice reliability (low, medium, and high), and task risk (low, medium, and high) on subjective trust, advice reliance, and eye‐tracking patterns. Results show that medium‐risk tasks elicited the highest subjective trust but the lowest attention to advice, while reliance and final fixations were higher than in low‐risk tasks. Importantly, trust in AI relative to human experts depended on both reliability and risk: AI was favored in medium‐risk, medium‐reliability conditions but was trusted less than humans in high‐risk contexts. These findings highlight the often‐overlooked role of medium levels of risk and reliability in human–AI collaboration, offering insights for optimizing AI‐assisted decision‐making.