Step-by-Step Causality: Transparent Causal Discovery with Multi-Agent Tree-Query and Adversarial Confidence Estimation

2026-01-15

AI Summary

This work proposes Tree-Query, a novel framework that addresses the limitations of traditional causal discovery methods—susceptibility to error propagation—and current large language model (LLM)-based approaches, which often lack interpretability and confidence calibration. Tree-Query organizes multiple expert LLMs in a tree structure, decomposing causal discovery into interpretable sub-queries concerning backdoor paths, conditional independence, latent confounding, and causal direction. The framework incorporates adversarial confidence estimation to provide theoretical guarantees of asymptotic identifiability and enables the extraction of causal priors even in data-scarce or data-free settings. Experimental results demonstrate that Tree-Query outperforms direct LLM baselines on the Mooij and UCI data-free benchmarks, achieving significant improvements in structural metrics, and successfully identifies confounders while yielding high-confidence causal conclusions in a diet–weight case study.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Causal discovery aims to recover ``what causes what'', but classical constraint-based methods (e.g., PC, FCI) suffer from error propagation, and recent LLM-based causal oracles often behave as opaque, confidence-free black boxes. This paper introduces Tree-Query, a tree-structured, multi-expert LLM framework that reduces pairwise causal discovery to a short sequence of queries about backdoor paths, (in)dependence, latent confounding, and causal direction, yielding interpretable judgments with robustness-aware confidence scores. Theoretical guarantees are provided for asymptotic identifiability of four pairwise relations. On data-free benchmarks derived from Mooij et al. and UCI causal graphs, Tree-Query improves structural metrics over direct LLM baselines, and a diet--weight case study illustrates confounder screening and stable, high-confidence causal conclusions. Tree-Query thus offers a principled way to obtain data-free causal priors from LLMs that can complement downstream data-driven causal discovery. Code is available at https://anonymous.4open.science/r/Repo-9B3E-4F96.

Problem

• causal discovery
• error propagation
• black-box models
• confidence estimation
• interpretability

Innovation

• Tree-Query
• causal discovery
• multi-agent LLM
• adversarial confidence estimation
• data-free causality