发现预训练模型的性别偏见会强烈传递到提示适配后的模型中。
Evaluating Gender Bias Transfer between Pre-trained and Prompt-Adapted Language Models
- 通过指代消解任务，研究预训练模型在提示适配下的偏见传播
- 三类模型偏见相关性高达0.94以上，即使调整提示也保持强关联
- 提示长度和刻板印象组合变化下仍稳定传递，适合关注公平性的研究者
大型语言模型（LLMs）正被广泛用于实际决策系统中以实现任务特定化。以往研究发现，预训练掩码语言模型的公平性对微调后模型的公平性影响有限。本文将偏见传递假说（BTH）扩展至基于提示的适配场景，因提示是部署模型的低计算开销方式。实验表明，在指代消解任务中，Mistral、Falcon和Llama等预训练模型的内在偏见与零样本和少样本提示后的偏见高度相关（皮尔逊相关系数rho ≥ 0.94）。即使模型被特别引导表现出公平或偏见行为，偏见传递依然显著（rho ≥ 0.92），且在不同提示长度和刻板印象构成下仍保持强相关（rho ≥ 0.97）。结果强调了在预训练阶段保障公平性的关键作用，尤其当模型后续通过提示适配执行下游任务时。
原文摘要 · Abstract (English)
Large language models (LLMs) are increasingly being adapted to achieve task-specificity for deployment in real-world decision systems. Several previous works have investigated the bias transfer hypothesis (BTH) by studying the effect of the fine-tuning adaptation strategy on model fairness to find that fairness in pre-trained masked language models have limited effect on the fairness of models when adapted using fine-tuning. In this work, we expand the study of BTH to causal models under prompt adaptations, as prompting is an accessible, and compute-efficient way to deploy models in real-world systems. In contrast to previous works, we establish that intrinsic biases in pre-trained Mistral, Falcon and Llama models are strongly correlated (rho >= 0.94) with biases when the same models are zero- and few-shot prompted, using a pronoun co-reference resolution task. Further, we find that bias transfer remains strongly correlated even when LLMs are specifically prompted to exhibit fair or biased behavior (rho >= 0.92), and few-shot length and stereotypical composition are varied (rho >= 0.97). Our findings highlight the importance of ensuring fairness in pre-trained LLMs, especially when they are later used to perform downstream tasks via prompt adaptation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。