Clinical judgments and computational harms: the origins and ethics of ableism in evolving clinical NLP applications

Artificial intelligence (AI) is often positioned as a force for social inclusion, with disabled people cited as key beneficiaries of these technologies. While some AI applications have produced meaningful benefits for disabled people, I argue that the integration of AI into healthcare risks entrenching ableism in new and more durable forms. Specifically, I examine how electronic health records, authored within a healthcare system dominated by ableism and the medical model of disability, inform and bias the training of clinical AI systems. Tracing the institutional and epistemic roots of ableism in healthcare, I argue that current documentation and medical education practices reinforce ableism and harmful understandings about disability, which harm disabled people. Grounded in frameworks of disability justice and bioethics, this article offers an ethical analysis of the problem. I propose interventions, including counter-narrative annotations authored by disabled people and reforms to electronic health record documentation and medical training.