New GoT-CD method improves causal discovery but highlights fairness audit fragility

Researchers have developed GoT-CD, a new causal discovery method that utilizes a Graph of Thoughts reasoning approach. This method generates multiple candidate graphs in parallel and merges them under a union constraint, ensuring the discovery of valid directed acyclic graphs (DAGs). While GoT-CD demonstrates competitive performance against LLM baselines on several benchmarks, including Asia, Alzheimer's, and COVID-Respiratory datasets, it highlights a critical issue: structural fidelity alone does not guarantee accurate fairness audits. The study reveals that post-hoc path-specific fairness analyses can be misleading when applied to discovered graphs, as some graphs may fail to identify unfair pathways, leading to incorrect conclusions about the model's fairness. AI

IMPACT Highlights the need for more robust methods to ensure fairness audits are reliable when applied to AI models.

RANK_REASON The cluster contains an academic paper detailing a new method for causal discovery. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Alzheimer's disease
• COVID-Respiratory
• directed acyclic graph
• GoT-CD
• Graph of Thoughts
• large language model

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →