Artificial intelligence, particularly large-language models, increasingly informs clinical decision-making, from triage to treatment recommendations. While promising efficiency and objectivity, these systems can encode and amplify historical biases present in clinical practice and documentation. Names, language, and other implicit social signals can trigger inequitable recommendations, formalizing discrimination at scale. Mitigating this requires technical, data-centric, and institutional interventions, including counterfactual testing, equitable dataset design, and fairness audits. With deliberate oversight, AI can not only improve care but also reveal and address persistent healthcare disparities.
Om M. Patel (Wed,) studied this question.