Abstract

To perform effective causal inference in high-dimensional datasets,initiating the process with causal discovery is imperative, wherein a causalgraph is generated based on observational data. However, obtaining a completeand accurate causal graph poses a formidable challenge, recognized as an NP-hard problem. Recently, the advent of Large Language Models (LLMs) has usheredin a new era, indicating their emergent capabilities and widespreadapplicability in facilitating causal reasoning across diverse domains, such asmedicine, finance, and science. The expansive knowledge base of LLMs holds thepotential to elevate the field of causal reasoning by offeringinterpretability, making inferences, generalizability, and uncovering novelcausal structures. In this paper, we introduce a new framework, namedAutonomous LLM-Augmented Causal Discovery Framework (ALCM), to synergizedata-driven causal discovery algorithms and LLMs, automating the generation ofa more resilient, accurate, and explicable causal graph. The ALCM consists ofthree integral components: causal structure learning, causal wrapper, andLLM-driven causal refiner. These components autonomously collaborate within adynamic environment to address causal discovery questions and deliver plausiblecausal graphs. We evaluate the ALCM framework by implementing twodemonstrations on seven well-known datasets. Experimental results demonstratethat ALCM outperforms existing LLM methods and conventional data-driven causalreasoning mechanisms. This study not only shows the effectiveness of the ALCMbut also underscores new research directions in leveraging the causal reasoningcapabilities of LLMs.