原创核心摘要
本研究是目前规模最大的真实商用招聘算法实地调研，样本覆盖 340 万求职者、400 万份求职简历，横跨 150 家企业、11 个行业、1700 个招聘岗位，全部简历由同一家第三方 AI 招聘系统统一筛选。研究参照美国 EEOC 五分之四歧视判定法则，证实 AI 招聘工具存在明确种族筛选偏差：26% 黑人应聘者、15% 亚裔应聘者在部分岗位遭遇算法歧视，若消除偏见，两类群体可新增 4 万次面试晋级机会。
研究指出统计陷阱：全岗位数据均值会抹平单岗位歧视，算法常会偏好黑人应聘仓储岗、排斥金融岗，汇总平均后歧视数据被隐藏。行业多家企业共用同一款筛选算法催生「算法单一化弊病」，人工招聘时代企业决策相互独立，而统一算法下 10% 投递四份简历的求职者全部惨遭全拒，形成系统性就业封锁。
AI 招聘工具兼具全民普及、影响重大、运行黑箱三大特征，严重左右劳动者就业走向。课题组呼吁强化算法招聘独立学术调研，为相关就业监管立法提供实证依据。
原文节选
The research tracks 3.4 million applicants with 4 million applications across 150 employers. Following EEOC’s four-fifths rule, 26% of Black applicants and 15% of Asian applicants face discriminatory screening on certain roles; removing bias would let another 40,000 applications advance.
Aggregated overall results conceal position-level racial bias. Algorithmic monoculture leads to systemic rejection: 10% of applicants with four applications get rejected everywhere, a phenomenon unseen under independent human hiring decisions.
AI hiring tools are widely deployed, high-stakes and opaque, highlighting the need for independent academic research to support evidence-based industry regulation.
版权声明：内容观点节选引用自斯坦福大学以人为本人工智能研究所 (Stanford HAI)，附原文链接：https://hai.stanford.edu/news/ai-hiring-tools-can-yield-racial-bias-and-systemic-rejection