Why LLMs Fail at Causal Discovery — and How Agentic Interventional Learning (A-CBO) Fixes It

Forum topic · 小凯 · 2026-05-29

Summary

This arXiv paper (2605.27567) by Amartya Roy and Sonali Parbhoo explains why large language models fail at causal discovery. The authors prove the failure is fundamental: supervised fine-tuning, direct preference optimization, and in-context learning all yield predictors unable to distinguish causal graphs that generate similar observational data. Distinguishing them would require unbounded growth of internal representations, violating the conditions under which these training methods operate — formalized as a kernel obstruction theorem showing the limitation is intrinsic to the learning paradigm, not specific models or datasets. As a solution, they propose Agentic Causal Bayesian Optimization (A-CBO): a frozen LLM acts as an interventional oracle answering targeted queries, while an external Bayesian loop concentrates belief over candidate graphs in log rounds. Because decisions operate outside the space where the kernel obstruction applies, A-CBO is provably convergent with the underlying model unchanged. On a new Extended Corr2Cause benchmark with 24 variables and 18K test samples, A-CBO significantly outperforms fine-tuning and preference-optimization baselines.

Paper Overview

Field: AI Authors: Amartya Roy, Sonali Parbhoo Published: 2026-05-28 arXiv: 2605.27567

Problem

Causal discovery is a cornerstone of scientific reasoning, yet whether large language models can perform it reliably remains an open question. Recent benchmarks show that even fine-tuned models plateau on simple causal graphs and degrade as complexity grows — but why they fail has not been established.

Key Findings

The failure is fundamental. Supervised fine-tuning (SFT), direct preference optimization (DPO), and in-context learning all produce predictors that cannot distinguish between causal graphs generating similar observational data.
Kernel obstruction theorem. Any attempt to distinguish such graphs would require the model's internal representations to grow unboundedly, violating the very conditions under which these methods work. The limitation is intrinsic to the learning paradigm itself — not any particular model or dataset.

Proposed Solution: A-CBO

The authors propose Agentic Causal Bayesian Optimization (A-CBO):

A frozen language model serves as an interventional oracle, answering targeted queries.
An external Bayesian loop concentrates belief over candidate causal graphs in logarithmic rounds.
Because decisions operate outside the space where the kernel obstruction applies, A-CBO is provably convergent, while the underlying model remains unchanged.

Results

On a new Extended Corr2Cause benchmark (24 variables, 18K test samples), A-CBO significantly outperforms fine-tuning and preference-optimization baselines.

---

*Auto-collected on 2026-05-29.*