Cognitive Friction and the Illusion of Control: An Empirical Study on Sense of Agency and Responsibility Attribution in Medical AI Interactions

Medical AI interfaces can shape physicians’ agency, diagnostic behavior, and responsibility judgments. In a 2 × 2 within-subjects experiment, 60 radiologists interacted with prescriptive or dialectical interfaces featuring instrumental or placebo controls. We assessed explicit agency, intentional binding, rejection of erroneous AI recommendations, and responsibility allocation, supplemented by interviews. Placebo control was associated with high explicit agency but attenuated intentional binding. In the prescriptive-placebo condition, physicians attributed 62.0% of responsibility to themselves. Dialectical interaction increased rejection of erroneous AI recommendations from 31.8% to 70.3%. These findings show condition-dependent differences in agency measures, diagnostic decisions, and responsibility attribution. Because functional control and feedback predictability varied together, the cause of attenuated intentional binding remains uncertain. Intentional binding and responsibility attribution were analyzed separately; their co-occurrence does not establish mediation. The results support further testing of how interface design affects control and accountability in medical AI.