Interpretive Competence as Epistemic Infrastructure in AI-Mediated Knowledge Production

When large language models participate in scientific inquiry, the wording of a request helps shape the knowledge claims presented for appraisal. This conceptual article examines the interpretive demands that follow. Technical improvements cannot remove all such demands: even accurate responses require framing appropriate to the inquiry, and generated text can sound authoritative or cautious without adequate evidential support. We propose four interpretive competencies: framing awareness, uncertainty detection, evidential coherence judgment, and interactional reflexivity. The framework organizes them by what users examine—their own contribution or the generated representation—and whether they examine a bounded instance or relations among instances. Medicine and biomedical research provide the worked examples. Illustrative runs with three language models show how several vulnerabilities appear in generated responses; they do not validate the proposed competencies. We distinguish the framework from established approaches to evidence appraisal and AI literacy by its object and level of analysis, and outline tasks for cultivating and assessing the competencies. We argue that interpretive competence can become part of the epistemic infrastructure of AI-mediated knowledge production when embedded in shared training, appraisal, and assessment practices.