This Conceptual Note examines a recent finding that adding a medical knowledge base to a reasoning-based medical AI system did not necessarily improve performance and, in one model, reduced accuracy. The note argues that retrieval-augmented medical AI should not be understood as a model with a passive document library attached. Once connected to clinical reasoning, the knowledge base becomes part of the intervention architecture. From a Universal Resonance Model perspective, the key governance issue is not only whether information is available, but whether the system can select the right information for the right patient state, at the right moment, under uncertainty. The note proposes that medical AI should be evaluated as a full clinical operating system, including the model, retrieval layer, knowledge source, workflow integration, escalation logic, and failure modes. It highlights state-sensitive interpretation as a core requirement for safe clinical AI governance.
Anita Domargård (Wed,) studied this question.