Large Causal Models from Large Language Models (arxiv.org)

A groundbreaking approach to developing large causal models (LCMs) using existing large language models (LLMs) has been introduced with the implementation of a system named DEMOCRITUS. This innovative framework enables the extraction, organization, and visualization of causal relations from diverse domains through targeted queries to LLMs. Unlike traditional causal inference methods that rely on numerical data from experiments, DEMOCRITUS leverages the generative capabilities of high-quality LLMs to formulate causal questions and derive plausible causal statements, ultimately transforming fragmented or conflicting claims into coherent relational causal triples. The significance of this development for the AI/ML community lies in its potential to streamline causal modeling across various fields like archaeology, biology, and economics. By addressing the complexity of synthesizing diverse knowledge into a unified causal model, it opens avenues for interdisciplinary research and practical applications. The DEMOCRITUS system comprises a six-module implementation pipeline, and while it has shown promising results, the authors also highlight current computational bottlenecks and limitations, paving the way for future enhancements that could further expand its capabilities and efficiency in causal inference.