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

Medical AI systems increasingly shift clinicians from direct decision-makers to supervisors of algorithmic outputs, raising critical concerns about cognitive load, automation bias, and ethical responsibility. This study reveals how prevailing frictionless interface designs induce an illusion of control that dissociates explicit and implicit sense of agency (SoA), while simultaneously inflating physicians’ self-attributed responsibility through moral crumple zone effects. Using a 2×2 within-subjects experiment (N=60 radiologists) combining intentional binding, diagnostic performance, responsibility matrices, and grounded theory interviews, we demonstrate that (1) placebo controls produce paradoxically high explicit SoA alongside near-complete collapse of implicit SoA; (2) dialectical (adversarial) interaction protocols significantly increase rejection of deliberate AI errors (70.3% vs. 31.8%, OR=4.85); and (3) illusory control drives disproportionate self-attribution of responsibility (62% vs. ~40% in other conditions), even when physicians possess no meaningful corrective authority. These findings provide the first quantitative evidence in medicine that frictionless design actively undermines professional judgment and creates asymmetric accountability. We propose a paradigm shift from “seamless” to “honest” interaction—one that preserves productive friction to maintain genuine agency and fair distribution of responsibility.