CausalGraph2LLM: Evaluating LLMs for Causal Queries

2024-10-21

arXiv.org

AI Summary

Large language models (LLMs) exhibit sensitivity to causal graph encoding formats and lack well-defined performance boundaries in causal reasoning. Method: We introduce CausalGraph2LLM—the first large-scale causal graph benchmark comprising over 700,000 dual-granularity queries—and systematically evaluate LLMs on both graph-level and node-level causal tasks. We propose a DAG-based dual-granularity query classification framework and conduct extensive experiments across open- and closed-weight models under diverse graph encoding strategies. Contribution/Results: Our study quantifies, for the first time, substantial encoding sensitivity in mainstream LLMs (e.g., GPT-4, Gemini-1.5), with performance deviations up to 60%. We further identify significant parametric memory biases in interventional and contextual causal inference. These findings establish a reproducible evaluation paradigm for trustworthy causal AI and provide critical diagnostics of key failure modes.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Causality is essential in scientific research, enabling researchers to interpret true relationships between variables. These causal relationships are often represented by causal graphs, which are directed acyclic graphs. With the recent advancements in Large Language Models (LLMs), there is an increasing interest in exploring their capabilities in causal reasoning and their potential use to hypothesize causal graphs. These tasks necessitate the LLMs to encode the causal graph effectively for subsequent downstream tasks. In this paper, we introduce CausalGraph2LLM, a comprehensive benchmark comprising over 700k queries across diverse causal graph settings to evaluate the causal reasoning capabilities of LLMs. We categorize the causal queries into two types: graph-level and node-level queries. We benchmark both open-sourced and propriety models for our study. Our findings reveal that while LLMs show promise in this domain, they are highly sensitive to the encoding used. Even capable models like GPT-4 and Gemini-1.5 exhibit sensitivity to encoding, with deviations of about $60%$. We further demonstrate this sensitivity for downstream causal intervention tasks. Moreover, we observe that LLMs can often display biases when presented with contextual information about a causal graph, potentially stemming from their parametric memory.

Problem

• Evaluating LLMs for causal reasoning
• Assessing encoding sensitivity in causal graphs
• Identifying biases in LLMs' causal interpretations

Innovation

• LLMs for causal reasoning
• 700k diverse causal queries
• Sensitivity to graph encoding