This chapter examines the epistemic value of (purely) predictive ML models for public health. By discussing a novel strand of research at the intersection of ML and economics that recasts policy problems as prediction problems, we argue—against skeptics—that predictive models can indeed be a useful guide for policy interventions, provided that certain conditions hold. Using behavioral approaches to policymaking such as Nudge theory as a contrast class, we carve out a distinct feature of the ML approach to public policy problems: the ML model itself may turn into a cognitive intervention. In underscoring the epistemic value of predictive models, we also highlight the importance of taking a broader perspective on what constitutes good evidence for policymaking. Moreover, by focusing on public health, we also contribute to the understanding of the specific methodological challenges of ML-driven science outside of traditional success areas.