Causal AI Scientist: Facilitating Causal Data Science with Large Language Models

Causal AI Scientist, or CAIS, takes a natural-language causal query such as whether a job-training program raises income, together with a dataset and its description, and works through causal effect estimation end to end using large language models. It profiles the dataset's column types, missing values and statistical distributions to identify treatment, outcome and covariate variables, then traverses a rule-based decision tree that evaluates properties such as randomization, temporal structure and the availability of instruments to select a valid causal inference method rather than leaving method selection to an opaque LLM judgment call.

Supported methods include econometric approaches such as difference-in-differences, instrumental variables, ordinary least squares and regression discontinuity design, as well as causal-graph-based backdoor and frontdoor adjustment. Once a method is selected and executed with predefined code templates, CAIS validates the results with diagnostic checks and interprets the numerical output in the context of the original query, aiming to make causal effect estimation usable without requiring the user to already have deep methodological expertise.