Burkov recommends Peng Ding's causal inference textbook to fix AI's correlation-only blind spot

burkov · x · 2026-09-05

ML author Andriy Burkov recommends A First Course in Causal Inference by Peng Ding (2024). His argument: today's AI systems excel at detecting associations but can't reliably answer interventional or counterfactual questions — exactly the ones that arise when models recommend, treat, rank, price, or act.

The book covers the potential-outcomes framework, randomized-experiment analysis, confounding and identification in observational data, propensity scores, matching, doubly robust estimation, sensitivity analysis, and instrumental variables. These tools help practitioners separate stable causal mechanisms from spurious correlations, run correct A/B tests, and reason about distribution shift and fairness.

Original post →

More from Research

Research channel →