When prediction becomes prescription : AI, risk, and the humanistic right to moral deliberation.

Artificial intelligence is increasingly used to estimate diagnosis, prognosis, treatment response, deterioration, mortality, and other clinically relevant risks. These predictions may improve care, but they can also acquire prescriptive force: an estimate of what is likely to occur becomes a recommendation about what ought to be done. Existing ethical discussions describe opacity, bias, automation bias, responsibility gaps, threats to autonomy, and the need for meaningful human control. These concerns are important, but they do not fully explain how an accurate, transparent, and statistically fair prediction can still narrow moral judgment. This article introduces normative compression, the process by which heterogeneous clinical evidence, uncertain probabilities, institutional priorities, and patient values are condensed into an apparently self-justifying course of action. Normative compression can occur through outcome selection, uncertainty suppression, value substitution, interface design, and institutional embedding. It is ethically significant because medical risk is never only a probability. Its acceptability depends on what outcomes mean for a particular patient, which burdens are bearable, which relationships and commitments matter, and who may legitimately decide. Drawing on relational and structural accounts of medical humanism, shared decision-making, relational autonomy, and recent work on medical AI, the article defends a humanistic right to moral deliberation. This is a moral, not necessarily legal, right to a clinical process in which prediction remains distinguishable from prescription and in which patients can participate meaningfully in interpreting, contesting, and revising the reasons for action. A seven-part PREDICT framework is proposed to operationalize this right. The article concludes that human oversight is insufficient when humans merely ratify an algorithmic recommendation. Humanistic medicine requires preserving a space in which evidence informs judgment without pre-empting it.