The Ethics of Artificial Intelligence in Clinical Diagnostic Decision-Making

Abstract

Artificial intelligence is increasingly embedded in the diagnostic process, from autonomous screening tools
to decision-support systems that flag findings for clinician review, raising ethical questions that go well
beyond whether these systems are accurate enough to trust.
This perspective examines how the ethical stakes of diagnostic AI vary with its deployment model, and
surveys recent literature on algorithmic bias and the limits of treating accuracy and fairness as
interchangeable goals.
It then turns to the black-box problem and the unsettled question of who is accountable when an opaque but
well-validated system contributes to a diagnostic error, and to informed consent in an AI-mediated diagnostic
encounter that patients are rarely told about.
We also consider evidence on automation bias and the deskilling of clinicians who come to rely on
diagnostic AI, and the regulatory apparatus -- centered on the FDA's evolving framework for adaptive
software as a medical device -- that has been slow to keep pace with either concern.
We conclude that no single deployment model or regulatory fix resolves these tensions, and that the ethical
questions AI raises for diagnosis are best addressed deployment model by deployment model rather than as
a single undifferentiated debate about "AI in medicine."