Reproduce “Causal Agent based on Large Language Model” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Kairong Han, Kun Kuang, Ziyu Zhao, Junjian Ye, Fei Wu

A Causal Agent framework empowers large language models (LLMs) to perform robust, data-driven causal reasoning on tabular data by integrating specialized causal analysis tools, a structured planning process, and a memory module. This approach successfully addresses the challenge of applying LLMs to complex causal problems, achieving high accuracy across a new CausalTQA benchmark and outperforming previous state-of-the-art models on the real-world QRData dataset.

Key insights

  • Integrating external, specialized causal tools with an LLM via an agent framework effectively bridges the gap between natural language understanding and complex tabular causal analysis.
  • Hierarchical decomposition of causal problems allows LLMs to systematically address different aspects of causality, from basic statistical dependencies to quantitative causal effects.
  • The iterative Thought-Action-Observation loop, coupled with non-textual memory for causal graphs, enables robust multi-turn reasoning and tool chaining, outperforming direct code generation by LLMs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.