As artificial intelligence (AI) systems increasingly outperform human clinicians in specific diagnostic tasks, legal debates have turned to whether such statistical superiority should create new obligations in medical practice. This article proposes a two-stage transparency framework, distinguishing 'pre-deployment transparency' from 'post-deployment interpretability', to clarify when clinicians may, must, or must not use or rely upon AI systems. It argues that duties to adopt or rely on AI arise only where institutional endorsement and meaningful transparency enable doctors to make informed, context-sensitive judgments. Legal responsibility in AI-assisted care must rest on institutional validation and explainability, not on statistical performance alone. The article further shows that, consistent with existing case law, courts may draw adverse inferences from evidentiary gaps created by AI opacity, particularly when a party fails to preserve or disclose information within its control. This framework preserves clinical judgment and patient trust while ensuring that overall statistical gains do not mask systematic harms to minority groups. It concludes with recommendations for adapting medico-legal standards to the growing role of AI without displacing the clinician's role as the legally accountable decision-maker.
Amelie Sophie Berz (Sat,) studied this question.