GoT-CD: Graph-of-Thoughts Causal Discovery and the Fragility of Post-hoc Path-Specific Fairness Audits

2026-08-03

AI Summary

This work addresses the risk that structural errors in causal discovery can compromise path-specific fairness auditing, particularly when sensitive pathways are misidentified. To mitigate this, the authors propose GoT-CD, a novel method that treats the full candidate edge set as a unified reasoning unit and leverages large language model–driven Graph-of-Thoughts to generate multiple causal graphs in parallel. The approach enforces hard joint constraints to prevent spurious edges and ensures acyclicity through deterministic scoring combined with greedy DAG projection. Experiments demonstrate that GoT-CD consistently produces valid DAGs across five benchmarks, achieving state-of-the-art DAG-valid F1 scores among LLM-based methods on the Asia, Alzheimer’s, and COVID respiratory datasets. Crucially, on the Alzheimer’s data, most existing methods erroneously assess fairness due to missed sensitive paths, revealing that high structural accuracy does not guarantee reliable fairness auditing and underscoring the need for joint evaluation of causal discovery and downstream fairness analysis.

Abstract

Causal discovery recovers directed structure from observational data and is increasingly used in clinical settings to support mechanism reasoning and fairness audits of predictive models. Path-specific counterfactual fairness asks whether a protected attribute influences an outcome through illegitimate pathways, but these estimands are defined relative to a supplied causal graph and therefore inherit whatever errors the discovery step introduces. Discovery methods are routinely scored on aggregate structural metrics that weight all edges equally, and no established evaluation asks whether the specific pathway an audit depends on survives discovery---or what the audit reports when that pathway is missing. Here we show that full-graph Graph-of-Thoughts reasoning yields acyclic discovered graphs that are structurally competitive with large language model (LLM) baselines, yet that structural fidelity alone does not guarantee fairness-faithful audits. We introduce GoT-CD, in which the reasoning unit is a complete candidate edge set: multiple graphs are generated in parallel, scored by a deterministic validity function, and merged under a hard union constraint that forbids invented edges, with greedy projection enforcing a DAG before commitment. GoT-CD returns a valid DAG on all five reported benchmarks and achieves the best DAG-valid F1 score among LLM methods on Asia, Alzheimer's, and COVID-Respiratory datasets. On an Alzheimer's benchmark with known unfair path, a post-hoc path-specific audit shows that five of eight discovered graphs recover no path from the sensitive attribute to the outcome and therefore report a null overall effect while mediated effects persist, necessitating downstream path-specific fairness analysis along with structural discovery.

Problem

• causal discovery
• path-specific fairness
• post-hoc audit
• fairness-faithful
• causal graph

Innovation

• Graph-of-Thoughts
• Causal Discovery
• Path-Specific Fairness
• DAG Learning
• Post-hoc Audit