When prediction is not explanation: AI and the ethical transformation of medicine

Abstract Artificial intelligence (AI) is evolving from a collection of discrete applications towards a technology increasingly embedded in the infrastructure of medicine. This development may improve diagnosis, prediction, monitoring, and the use of clinical information, but technological adoption is advancing faster than the governance structures needed to accompany it. At the level of the individual patient, this transformation creates a specific ethical problem. Patient information enters computational systems and emerges as probabilities, classifications, predictions, or recommendations that may influence care. The black box provides a metaphor for the intervening computational process when clinically relevant elements cannot be sufficiently examined. Yet opening the black box does not resolve the entire problem: an intelligible and validated prediction does not itself establish that a proposed intervention will benefit a particular patient. I propose a Patient-Level AI Code of Conduct governing the transition from AI output to clinical action. The Code requires examinability, validation, patient-level applicability, independent clinical justification, and meaningful human responsibility before an AI output materially influences care. Its purpose is to preserve a fundamental distinction as AI becomes increasingly infrastructural to medicine: computational capability can inform clinical decisions, but it cannot by itself justify action on a patient.