AI alone will not transform healthcare: Learning Health Systems as the framework for responsible AI implementation

Artificial intelligence (AI) tools show strong technical performance in research settings, yet their uptake into routine healthcare is inconsistent and often inequitable. Recent hospital-level evidence highlights clustered adoption, lower uptake in high-need communities and a striking lack of routine evaluation for accuracy and bias. These patterns reveal a core implementation challenge rather than a technical limitation. AI differs from many prior healthcare innovations in that its performance is dynamic, shifting with data, workflows and population, thereby requiring continuous, rather than episodic, evaluation and adaptation. This commentary argues that the Learning Health System (LHS) provides the framework for successful AI translation. Implementation science supplies the tools: the Consolidated Framework for Implementation Research (CFIR) as a determinant framework for assessing readiness, workflow fit, governance and equity; and the Expert Recommendations for Implementing Change (ERIC) compilation, alongside the LHS learning cycle itself, as implementation process frameworks for selecting and enacting strategies. The LHS supplies what implementation science alone does not: permanent operational infrastructure that embeds these frameworks, plus evaluation tools such as the Translational Evaluation of Healthcare AI (TEHAI) and the Monash LHS-Maturity Matrix (LHS-MM), within routine care, so that monitoring, recalibration and coproduction continue after project funding ends. We outline the infrastructure, workforce and governance this requires; the operational burden it entails; and mechanisms by which LHSs can narrow, rather than widen, the emerging “AI divide.” With these conditions in place, AI can contribute to safer, more reliable and more equitable care. Without them, AI risks superficial deployment, increased burden and amplification of existing disparities.