CausalSteward: An Agentic Divide-Conquer-Combine Copilot for Causal Discovery

2026-07-02

AI Summary

High-dimensional causal discovery faces significant challenges, including the frequent violation of core assumptions, weak identifiability, and difficulties in integrating domain knowledge. This work proposes a human-in-the-loop, multi-agent divide-and-conquer framework that recursively clusters high-dimensional variables into manageable subproblems. By synergistically combining retrieval-augmented generation, conditional independence tests, and collaborative multi-agent reasoning, the approach deeply integrates data-driven methods with prior domain knowledge. The incorporation of human oversight further enhances the trustworthiness of inferred causal structures. Empirical results demonstrate that this paradigm substantially improves both the accuracy and interpretability of causal models, thereby validating the efficacy and delineating the boundaries of multi-agent architectures in high-dimensional causal inference.

Abstract

Learning causal models from high-dimensional data is a significant challenge, particularly in real-world settings where violations of core assumptions lead to causal identifiability issues. Although massive amounts of prior knowledge are available, and contain valuable causal information, effectively integrating this knowledge into the causal discovery process remains an open problem. We introduce CausalSTeward (CAST), a novel human-in-the-loop framework for interactively assembling large causal models. CausalSteward is a multi-agent collaborative system that tackles high-dimensional causality through a divide-and-conquer approach where large clusters of variables are iteratively partitioned and then separately analyzed. Our framework fuses prior knowledge with a data-driven approach by using tailored tools such as retrieval augmented generation and conditional independence tests. Finally, we use this work to examine the capabilities and limitations of causal reasoning in multi-agent frameworks, and how the human-in-the-loop can contribute to accurate and trustworthy results.

Problem

• causal discovery
• high-dimensional data
• causal identifiability
• prior knowledge integration
• causal reasoning

Innovation

• causal discovery
• multi-agent system
• divide-and-conquer
• retrieval-augmented generation
• human-in-the-loop