The WHO's "Ethics and Governance of AI for Health" guidance (2021, updated 2024), the AMA's Augmented Intelligence in Health Care policy, and a substantial peer-reviewed literature in NEJM AI, The Lancet Digital Health, and JAMA all converge on a common position: AI in healthcare must support, not replace, the clinician-patient relationship.
A human clinician must remain reachable in any consequential care decision, with AI playing the role of decision-support rather than gatekeeper. Triage, diagnostic, and treatment-recommendation AIs operate within a framework where the human carries ultimate clinical responsibility — which the AI's design must facilitate, not foreclose.