Are predictive models enough?

This article discusses the debate on whether predictive models are sufficient for causal inference, arguing that while good predictive models provide conditional distributions, they fall short without causal inference theory. It explains how causal graphs help identify necessary variables and avoid modeling irrelevant ones, using examples like randomized experiments and observational data. The author contrasts modern causal methods with older regression approaches, noting pitfalls like adjusting for intermediate variables. The piece emphasizes that predictive models alone cannot determine appropriate conditioning sets, making causal reasoning essential for accurate effect estimation.