LLM-Guided Causal Discovery Reveals Fairness Pathways
A new preprint (v2, Jan 7, 2026) by Khadija Zanna proposes a hybrid LLM-guided causal discovery framework that extends breadth-first search with active learning and dynamic composite scoring. The method prioritizes variable-pair queries using mutual information, partial correlation, and LLM confidence and is evaluated on a semi-synthetic UCI Adult benchmark embedding bias pathways and confounders. Results show LLM-guided variants outperform baselines at recovering fairness-relevant structures under noisy conditions.
Key Points
- 1Introduce hybrid LLM-guided causal discovery combining mutual information, partial correlation, and LLM confidence scoring
- 2Demonstrate improved recovery of fairness-critical paths under noise and latent confounding versus statistical baselines
- 3Enable auditors to prioritize variable queries and better detect bias pathways in noisy datasets
Scoring Rationale
Hybrid LLM-guided approach provides stronger recovery under noise; limited by single semi-synthetic benchmark and preprint validation.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems