AI-generated health intelligence: Rethinking evidence and governance in healthcare systems

Abstract Artificial intelligence is being used in healthcare to support a range of tasks such as diagnosis, screening and clinical decisions; however, the information produced through its use may also inform decisions beyond individual patient care. The interpretation and use of this information across healthcare systems have received less attention than the performance and implementation of AI technologies. To examine these processes, we propose a conceptual framework built around three interconnected propositions: the epistemic continuum, which traces how the meaning and use of AI-generated information may change across clinical, organisational, and policy settings; shifting epistemic roles, which considers how different actors contribute to producing, interpreting, and authorising that information; and the AI-generated health intelligence flywheel, which explores how information produced through clinical AI may feed back into decisions about its continued use or expansion. Using AI-supported chest X-ray screening as an illustrative case, we show why AI-generated classifications, screening activity, and confirmed clinical outcomes must be distinguished when assessing healthcare services. We argue that tracing how these forms of information become evidence is important for understanding the governance of clinical AI, particularly where decisions about healthcare priorities, resources, and technology deployment depend on them. The framework offers a basis for empirical research into these relationships without assuming that they necessarily result in inappropriate decisions or self-reinforcing patterns of AI adoption.