不可见的影响：通过大型语言模型中的角色工程探究隐性交叉偏见
Invisible Influences: Investigating Implicit Intersectional Biases through Persona Engineering in Large Language Models
AI总结：
本文通过角色工程研究大型语言模型中的隐性交叉偏见，提出BADx指标，揭示角色上下文对偏见动态影响的机制，评估五种LLM在不同角色框架下的表现。
摘要：
大型语言模型（LLMs）在生成人类语言方面表现出色，但往往嵌入并放大隐性交叉偏见，尤其是在角色驱动的上下文中。现有偏见审计依赖于静态的嵌入测试（CEAT、I-WEAT、I-SEAT），用于量化绝对关联强度。我们证明这些方法在捕捉模型采用社会角色时的动态变化时存在局限性。我们通过引入偏见放大差异和可解释性分数（BADx）来填补这一空白：一种新的、可扩展的度量标准，用于衡量角色诱导的偏见放大，并整合局部可解释性见解。BADx包括三个组成部分——差异偏见分数（BAD，基于CEAT、I-WEAT、I-SEAT）、角色敏感性指数（PSI）和波动性（标准差），并辅以LIME分析以强调可解释性。本研究分为两个不同任务进行。任务1建立静态偏见基准，任务2应用六个角色框架（边缘化和结构上占优）来测量BADx、PSI和波动性。该研究在五种最先进的LLM（GPT-4o、DeepSeek-R1、LLaMA-4、Claude 4.0 Sonnet和Gemma-3n E4B）上进行。结果显示角色上下文显著调节偏见。GPT-4o表现出高敏感性和波动性；DeepSeek-R1抑制偏见但波动性不规则；LLaMA-4保持低波动性和稳定的偏见特征，放大有限；Claude 4.0 Sonnet实现平衡调节；Gemma-3n E4B达到最低波动性，偏见放大适度。BADx比静态方法表现更好，揭示了静态方法忽视的上下文敏感偏见。我们的统一方法提供了一种系统的方法，以检测五种流行LLM中的动态隐性交叉偏见。
英文摘要：
Large Language Models (LLMs) excel at human-like language generation but often embed and amplify implicit, intersectional biases, especially under persona-driven contexts. Existing bias audits rely on static, embedding-based tests (CEAT, I-WEAT, I-SEAT) that quantify absolute association strengths. We show that they have limitations in capturing dynamic shifts when models adopt social roles. We address this gap by introducing the Bias Amplification Differential and Explainability Score (BADx): a novel, scalable metric that measures persona-induced bias amplification and integrates local explainability insights. BADx comprises three components - differential bias scores (BAD, based on CEAT, I-WEAT, I-SEAT),Persona Sensitivity Index (PSI), and Volatility (Standard Deviation), augmented by LIME-based analysis for emphasizing explainability. This study is divided and performed as two different tasks. Task 1 establishes static bias baselines, and Task 2 applies six persona frames (marginalized and structurally advantaged) to measure BADx, PSI, and volatility. This is studied across five state-of-the-art LLMs (GPT-4o, DeepSeek-R1, LLaMA-4, Claude 4.0 Sonnet and Gemma-3n E4B). Results show persona context significantly modulates bias. GPT-4o exhibits high sensitivity and volatility; DeepSeek-R1 suppresses bias but with erratic volatility; LLaMA-4 maintains low volatility and a stable bias profile with limited amplification; Claude 4.0 Sonnet achieves balanced modulation; and Gemma-3n E4B attains the lowest volatility with moderate amplification. BADx performs better than static methods by revealing context-sensitive biases overlooked in static methods. Our unified method offers a systematic way to detect dynamic implicit intersectional bias in five popular LLMs.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
