Can Large Language Models Help Experimental Design for Causal Discovery?

2025-03-03

AI Summary

In causal discovery, accurately identifying causal structures from observational data alone remains challenging, while intervention-based experiments suffer from high costs and low sample efficiency. To address this, we propose the first interpretable intervention-target recommendation framework that integrates large language model (LLM) priors with numerical optimization. Our method embeds LLMs’ world knowledge into causal graph modeling and active learning, and employs a multi-round intervention feedback mechanism to enable end-to-end LLM-augmented experimental design. Evaluated on four real-world benchmarks, our approach achieves significantly higher causal structure identification accuracy and sample efficiency than conventional uncertainty- or gradient-driven methods—and even surpasses human expert judgment. This work provides the first empirical validation of substantial, generalizable gains from LLMs in scientific experimental design.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Designing proper experiments and selecting optimal intervention targets is a longstanding problem in scientific or causal discovery. Identifying the underlying causal structure from observational data alone is inherently difficult.Obtaining interventional data, on the other hand, is crucial to causal discovery, yet it is usually expensive and time-consuming to gather sufficient interventional data to facilitate causal discovery.Previous approaches commonly utilize uncertainty or gradient signals to determine the intervention targets. However, numerical-based approaches may yield suboptimal results due to the inaccurate estimation of the guiding signals at the beginning when with limited interventional data. In this work, we investigate a different approach, whether we can leverage Large Language Models (LLMs) to assist with the intervention targeting in causal discovery by making use of the rich world knowledge about the experimental design in LLMs.Specifically, we present oursfull (ours) -- a robust framework that effectively incorporates LLMs to augment existing numerical approaches for the intervention targeting in causal discovery. Across $4$ realistic benchmark scales, ours demonstrates significant improvements and robustness over existing methods and even surpasses humans, which demonstrates the usefulness of LLMs in assisting with experimental design for scientific discovery.

Problem

• Optimizing intervention targets for causal discovery.
• Reducing cost and time of obtaining interventional data.
• Enhancing accuracy in causal structure identification.

Innovation

• Leverage LLMs for intervention targeting
• Augment numerical approaches with LLMs
• Improve robustness in causal discovery