A Novel Approach to Eliminating Hallucinations in Large Language Model-Assisted Causal Discovery

TL;DR

This paper investigates hallucinations in large language models used for causal discovery, proposing retrieval-augmented generation and multi-LLM debate methods to reduce false information and improve reliability.

Contribution

It is the first to survey hallucinations in LLMs for causal discovery and introduces novel techniques including RAG and multi-LLM debate for hallucination mitigation.

Findings

Hallucinations are prevalent in LLMs for causal discovery.

RAG reduces hallucinations when quality data is available.

Multi-LLM debate achieves comparable hallucination reduction.

Abstract

The increasing use of large language models (LLMs) in causal discovery as a substitute for human domain experts highlights the need for optimal model selection. This paper presents the first hallucination survey of popular LLMs for causal discovery. We show that hallucinations exist when using LLMs in causal discovery so the choice of LLM is important. We propose using Retrieval Augmented Generation (RAG) to reduce hallucinations when quality data is available. Additionally, we introduce a novel method employing multiple LLMs with an arbiter in a debate to audit edges in causal graphs, achieving a comparable reduction in hallucinations to RAG.