分析视觉语言模型推理如何加剧交叉性偏见，揭示其与人类认知的差距。
How Reasoning Influences Intersectional Biases in Vision Language Models
- 通过三种提示风格分析五款开源模型的推理过程
- 32类职业预测中发现偏见推理系统性导致交叉性差异
- 适合关注AI公平性与可解释性的研究者阅读
视觉语言模型（VLMs）在下游任务中应用日益广泛，但其训练数据常包含社会偏见，导致输出中出现偏差。与人类通过上下文和社会线索理解图像不同，VLMs 依赖统计关联进行处理，推理方式常偏离人类思维。通过分析 VLM 的推理过程，可理解偏见如何被延续并影响下游性能。本研究系统分析了五款开源 VLM 在职业预测任务中的社会偏见，基于 FairFace 数据集，覆盖32个职业，并采用三种不同提示风格，获取模型的预测与推理文本。结果表明，偏见推理模式系统性地导致交叉性不平等，强调在部署前需将 VLM 推理对齐人类价值观。
原文摘要 · Abstract (English)
Vision Language Models (VLMs) are increasingly deployed across downstream tasks, yet their training data often encode social biases that surface in outputs. Unlike humans, who interpret images through contextual and social cues, VLMs process them through statistical associations, often leading to reasoning that diverges from human reasoning. By analyzing how a VLM reasons, we can understand how inherent biases are perpetuated and can adversely affect downstream performance. To examine this gap, we systematically analyze social biases in five open-source VLMs for an occupation prediction task, on the FairFace dataset. Across 32 occupations and three different prompting styles, we elicit both predictions and reasoning. Our findings reveal that the biased reasoning patterns systematically underlie intersectional disparities, highlighting the need to align VLM reasoning with human values prior to its downstream deployment.
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。