COFT: Counterfactual-Conformal Decoding for Fair Chain-of-Thought Reasoning in Large Language Models
This work addresses the problem of societal bias amplification in LLM chain-of-thought reasoning for developers and users concerned with fair AI outputs, offering a significant improvement over existing methods.
Large language models (LLMs) often amplify societal biases during chain-of-thought (CoT) generation. This paper introduces COFT, a training-free decoding method that reduces standard bias metrics by 30-55% (median 38%) while maintaining task utility and language quality, with reasoning accuracies remaining unchanged.
Large language models (LLMs) can reveal and amplify societal biases during chain-of-thought (CoT) generation. We present COFT (Chain of Fair Thought), a training-free decoding method that applies token-level fairness control at decode time, with distribution-free marginal validity guarantees (under exchangeability) for any frozen causal language model. COFT operates in three stages. First, it creates a masked counterfactual prompt by replacing sensitive spans with neutral tokens. Second, it compares the factual and masked logit distributions through lightweight logit fusion to attenuate attribute-driven biases. Third, it uses dual-branch split-conformal calibration to certify per-step candidate token sets at a user-chosen risk level. We evaluate COFT across six models and multiple bias benchmarks. Our method reduces standard bias metrics by 30-55% (median 38%) while preserving task utility and language quality. Reasoning accuracies remain unchanged within run-to-run noise margins. The computational overhead is modest, equivalent to one additional cached forward pass (<=11%). COFT offers a clear, auditable path to safer CoT generation with significant bias reduction, negligible utility loss, and no requirement for retraining, auxiliary classifiers, or weight access.