美国人力资源软件公司Workday因其AI求职筛选工具涉嫌基于年龄、残疾和种族等因素实施歧视而面临诉讼1。这起案件凸显了一个更广泛的问题：AI系统中的偏见并非源于算法本身的技术缺陷，而是贯穿数据收集、标注和模型训练全过程的人为决策链条1。
研究表明，AI在招聘、教育和公共服务领域正在复制并放大既有的歧视模式1。2025年的研究显示，GPT-4和Microsoft Copilot在模拟招聘场景中明显偏好男性候选人，这种性别偏见在高级职位的筛选中尤为突出1。生成型AI还被发现倾向于创建年轻、较瘦、肤色较浅的工程师形象1。同年另有研究发现，近半数已报告的AI事件与多样性或包容性问题相关1。
要建立真正具有包容性的AI生态系统，仅依靠技术手段是不够的1。解决方案需要跨学科的知识协作、受影响群体的实质参与，以及对权力结构的深入审视，覆盖从设计到部署的整个AI生命周期中的人为决策改革1。
Human Resources software company Workday is facing lawsuits alleging that its AI-powered recruitment screening tool discriminates based on age, disability, and race.1 Research has demonstrated that AI systems across hiring, education, and public services replicate and amplify existing biases related to gender, race, and age—problems that originate not from faulty code but from human choices made during data collection, annotation, and model training.1
Recent studies illustrate the scope of these issues. A 2025 analysis found that GPT-4 and Microsoft Copilot favor male candidates in simulated recruitment scenarios, with the preference most pronounced for senior positions.1 Similarly, AI-generated images of engineers tend to depict younger, slimmer individuals with lighter skin tones.1 Another 2025 study determined that nearly half of reported AI incidents involve diversity or inclusion concerns.1
Addressing AI bias requires more than technical fixes.1 Creating an inclusive AI ecosystem demands interdisciplinary expertise, meaningful participation from affected communities, and critical examination of underlying power structures throughout the entire lifecycle—from initial data collection and labeling through system design to deployment.1
评论
还没有评论，欢迎留下第一条。