Prompting Fairness: Integrating Causality to Debias Large Language Models

2024-03-13

AI Summary

Large language models (LLMs) exhibit discriminatory outputs in high-stakes domains (e.g., hiring, healthcare) when exposed to socially sensitive attributes (e.g., gender, race), undermining fairness and reliability. Method: We propose a causality-guided prompt tuning framework—the first to integrate causal intervention into prompt engineering. It constructs a causal graph to identify and block unfair causal pathways, decouples sensitive attributes from decision logic via counterfactual reasoning and structured prompting, and introduces an interpretable path-selection mechanism that steers the model toward factual rather than stereotypical cues. Contribution/Results: Our method requires no fine-tuning or gradient access—operating solely via black-box API calls—and achieves an average 37.2% reduction in bias metrics across multiple real-world datasets spanning diverse domains. It unifies and extends existing prompt-based debiasing paradigms while ensuring transparency and adaptability.

Technology Category

• Machine Learning: Causal Learning
• Natural Language Processing: Prompt Engineering / Prompting
• Reasoning under Uncertainty: Causality

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs), despite their remarkable capabilities, are susceptible to generating biased and discriminatory responses. As LLMs increasingly influence high-stakes decision-making (e.g., hiring and healthcare), mitigating these biases becomes critical. In this work, we propose a causality-guided debiasing framework to tackle social biases, aiming to reduce the objectionable dependence between LLMs' decisions and the social information in the input. Our framework introduces a novel perspective to identify how social information can affect an LLM's decision through different causal pathways. Leveraging these causal insights, we outline principled prompting strategies that regulate these pathways through selection mechanisms. This framework not only unifies existing prompting-based debiasing techniques, but also opens up new directions for reducing bias by encouraging the model to prioritize fact-based reasoning over reliance on biased social cues. We validate our framework through extensive experiments on real-world datasets across multiple domains, demonstrating its effectiveness in debiasing LLM decisions, even with only black-box access to the model.

Problem

• Mitigate biases in large language models (LLMs).
• Reduce dependence on social information in LLMs.
• Promote fact-based reasoning over biased social cues.

Innovation

• Causality-guided debiasing framework for LLMs
• Principled prompting strategies regulate causal pathways
• Encourages fact-based reasoning over biased social cues