CAREER: Automated and Robust Causal Inference for Data-Driven Decisions

Stanford University, Stanford CA

Investigators

Abstract

Thank you! This CAREER project advances the state-of-the-art in causal artificial intelligence (AI) methods (the blending of causal inference with machine learning and AI methods). Building causal AI systems is a complex pipeline of activities that requires eliciting assumptions from domain experts, developing estimation strategies, constructing uncertainty estimates, performing analysis to understand how robust the conclusions are, and conducting real-world experiments for validation. By automating key components of this causal analysis pipeline and developing robust data-driven estimation procedures, this project will enable more decision-makers to leverage causal AI systems. Integrating education and research, this CAREER project will develop open-source software tools, foster collaborations between academia and industry, and create educational materials, including coding tutorials, lecture notes, and textbooks. This research should also facilitate adoption by practitioners, educates young researchers in the field, and exposes students in K-12 education to foundations of the field. The project explores the following key research directions: i) precise finite sample analysis of estimation procedures for complex causal quantities, such as heterogeneous treatment effects and dose-response curves, in complex scenarios, ii) automated construction of confidence intervals for arbitrary causal quantities in high-dimensional settings, iii) model selection and hyperparameter tuning with rigorous guarantees for both estimation accuracy and uncertainty quantification in the causal setting, iv) aiding users in providing domain assumptions through the use of Large Language Models, v) developing algorithmic approaches for the automated identification of causal effects under parametric or semi-parametric restrictions on the data-generating process, vi) providing approaches for finite-sample sensitivity analysis that provide sharp bounds on parameters of interest, vii) addressing statistical problems related to adaptive data collection either to enhance accuracy or to correct uncertainty estimates. The ultimate goal is to provide accessible tools and educational resources that reduce the barriers to entry to applying causal AI in various domains. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.