反事实审计：GPT、Claude、Gemini 等大模型在疼痛管理中未检出种族与性别偏见
Do Large Language Models Show Racial and Sex Bias in Pain Management? A Counterfactual Audit of GPT, Claude, Gemini, Llama, DeepSeek and Qwen
17 个主流模型、1386 组配对比较的疼痛管理偏见审计，结论是没测出种族性别差异，但模型自身配置的波动大得惊人。做医疗 AI 评测或关心公平性审计方法的人，这篇的方法论细节很值得抠。
一项发布于 arXiv 的反事实审计研究测试了 17 个大语言模型在疼痛管理中的种族与性别偏见，共 24 种配置，涵盖 OpenAI 的 GPT-4.1 mini 与 gpt-oss、Anthropic 的 Claude Haiku 4.5、Google 的 Gemini 2.5 Flash、Meta 的 Llama、DeepSeek 和 Qwen。Study 1 用 4 个开源模型评估 6 个急性疼痛病例，覆盖 4 个种族与 2 个性别的所有组合，与同种子下的白人男性参照组配对比较。
在单提示、JSON 约束任务下，Study 1 没有任何差异在校正后存活，Study 2 的 1,386 组比较全部未通过错误发现率校正；所有疼痛评分差异在 0-10 量表上等价于 1 分以内的零差异，238 组 Q-Pain 临床病例比较中 237 组差异不超过 0.25 分。阿片类药物决策的置信区间较宽（Study 1 不能排除约 20 个百分点的差异），因此只能说明未检出差异，而非证明无差异。模型之间的差异远大于患者身份之间的差异：同一参照患者获得高剂量阿片类药物的比例在不同配置下从 6% 到 93% 不等。
研究建议临床 AI 公平性审计应做到每个种族使用多个姓名、按随机种子配对比较、报告噪声底限，并对每个实际部署配置单独测试。对医疗 AI 开发者与医院采购方而言，这项工作提供了可复用的审计方法框架，也提示部署配置本身（服务软件、精度、推理开关）可能是比患者身份更重要的变异来源。
Do Large Language Models Show Racial and Sex Bias in Pain Management? A Counterfactual Audit of GPT, Claude, Gemini, Llama, DeepSeek and Qwen
Racial and ethnic minority patients receive less pain treatment than non-Hispanic White patients, and large language models (LLMs) used as medical AI could reproduce that bias. Prior audits tested a few proprietary models, stated race explicitly and rarely measured the models' own response instability. In exploratory, non-preregistered analyses, Study 1 had four open-weight LLMs assess six acute-pain vignettes in every combination of four races and two sexes, with identity stated or cued only by a name (two names per group), comparing each identity with a White male reference on the same case and sampling seed. In Study 2, 24 configurations of 17 models, including OpenAI's GPT-4.1 mini and gpt-oss, Anthropic's Claude Haiku 4.5, Google's Gemini 2.5 Flash, Meta's Llama, DeepSeek and Qwen, with reasoning on and off and at two precisions, answered the same vignettes and the 50 clinician-written Q-Pain cases. In this single-prompt, JSON-constrained task, no difference survived correction in Study 1, none of 1,386 comparisons survived the false discovery rate in Study 2, and every pain-severity difference was equivalent to zero within one point on a 0-10 scale (within a quarter point for 237 of 238 Q-Pain comparisons). Opioid decisions were bounded less precisely (Study 1 could not exclude differences of about 20 percentage points), so they show no detected difference, not proof of none. The few differences surviving correction within a model favoured Black patients. Models differed far more from one another and across their own settings than between patients (the reference patient received the high opioid dose in 6% to 93% of responses), although serving software, provider and prompt format also differed. Fairness audits of clinical AI should use several names per group, pair comparisons on the seed, report the noise floor and test each deployed configuration.
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include GPT-4.1, Claude, Gemini, 医疗AI.
Primary source verification: published directly by arXiv: DeepSeek with first-party authority. Cross-source corroboration: 11 independent sources have covered this development (including 宝玉, TestingCatalog, Rowan Cheung), confirming industry-wide signal strength.
