As concerns grow that artificial intelligence may replicate or intensify epistemic injustice (EI) in medicine, this paper argues that such warnings require demonstrable evidence on two fronts: that EI is actually present in a given AI system, and that this EI, if present, makes the system's effects normatively undesirable. Without evidence on the first point, the EI framework is, by its own standards, not applicable; without evidence on the second, it remains unclear whether epistemic justice should outweigh other values at stake.
The paper also flags a further complication: the very discourses used to assess these two points can themselves become sites of epistemic injustice. Two concrete steps are proposed , distinguishing AI's effects on how medical problems are constructed from its effects on the ability to address them, and centering assessments of AI-induced EI on relative accuracy and real-world impact on quality of care.
To the publication
