Automated Causal Effect Estimation through Self-Evolving AI

Causal inference is central to scientific discovery, yet choosing appropriate methods remains challenging due to the complexity of statistical methodology and real-world data. Inspired by the success of artificial intelligence in accelerating scientific software, we introduce an evolutionary framework that uses large language models to discover and iteratively refine causal methods. Across benchmarks, our estimators consistently outperform established baselines: our best estimator lay on the Pareto frontier of 58 human submissions for a recent community competition. We also extend the algorithm to achieve competitive results in settings with estimated rewards. Analysis of the evolutionary trajectories shows that agents progressively discover sophisticated strategies tailored to unrevealed data-generating mechanisms. Our findings suggest that language-model-guided evolution could be used in scientific settings with partially observed rewards such as causal inference.