Medical artificial intelligence (AI) has transitioned from algorithmic validation to clinical deployment, confronting systemic challenges in liability attribution, cost-effectiveness evaluation, and payment incentive alignment. This study examines three representative applications, bone age assessment, AI-assisted gastroscopy (GRAIDS), and traditional Chinese medicine (TCM) tongue diagnosis, spanning a standardization spectrum to analyze structural differences in physician duty, liability allocation, and reimbursement mechanisms. Integrating tort law’s economic analysis framework with medical AI’s technical characteristics, we demonstrate that task standardization critically determines institutional design: highly standardized domains enable straightforward duty establishment, while intermediate or low-standardization contexts impose dual pressures of tool adoption and clinical judgment maintenance. We propose a stratified liability framework grounded in the least-cost avoider principle, distributing responsibility across algorithm developers, clinical practitioners, and healthcare institutions based on their respective capacities to prevent harm. Economic evaluation should account for hidden costs including legal risk exposure, reputational harm, and erosion of physician–patient trust, particularly acute in face-to-face diagnostic modalities. Payment mechanisms must decouple AI deployment from quality achievement via outcome-contingent incentives. TCM AI requires multi-center data infrastructure and real-world validation before cautious implementation.
Tian et al. (Wed,) studied this question.