Human-AI Collaboration

The widespread application of medical artificial intelligence (AI) raises a critical question: Who should be held accountable when AI-assisted diagnostic decisions result in errors? Based on the Stimulus-Organism-Response (S-O-R) framework, this study employs a scenario-based survey experiment to examine how three responsibility attribution models (AI-primary, shared, and physician-primary responsibility) influence physicians' AI adoption intention and verification behavior through risk perception, internal moral attribution, and trust, while exploring the moderating role of AI literacy. Analysis of data from 487 clinical physicians reveals three key findings. First, physician-primary responsibility significantly reduced internal moral attribution (β=-0.804) and risk perception (β=-0.436), while enhancing benevolence trust (β=0.257). Second, internal moral attribution emerged as the strongest mediator affecting both adoption intention (β=-0.258) and verification behavior (β=0.298), surpassing risk perception. Third, MGA results also show group differences. AI literacy moderated the effects of ability trust and risk perception on adoption intention, with stronger effects observed in the high literacy group. The study contributes by introducing responsibility attribution as an institutional stimulus into medical AI acceptance research, revealing the "responsibility attribution→tool-ification cognition→behavioral change" mechanism. Findings suggest that physician-primary responsibility models coupled with AI literacy enhancement programs can optimize human-AI collaborative decision-making in healthcare.