Data-Driven Discrimination at Work
Data-Driven Discrimination at Work
复制标题
工作中数据驱动的歧视
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
P. Kim
中科院分区：
文献类型：
--
作者：
P. Kim
数据革命正在改变工作场所。但是，具有“中性”数据的人类决策者不是中立的，算法可以区分法律世界。平等。数据分析的风险从根本上重新考虑了反歧视学说。在决策算法会产生偏见的结果时，它们似乎与熟悉的差异案例相似，但该学说证明是不合适的不同的上下文，不同的影响学说无法解决算法可以引入偏见并造成危害的方式。标题VII，不同的待遇和不同影响的唯一公认的歧视形式。沿着种族或其他受保护的类别。关于如何将反歧视规范应用于算法的结论截然不同，这表明标题VII的以责任模型的可能性和限制既是”。
A data revolution is transforming the workplace. Employers are increasingly relying on algorithms to decide who gets interviewed, hired or promoted. Proponents of the new data science claim that automated decision systems can make better decisions faster, and are also fairer, because they replace biased human decision-makers with “neutral” data. However, data are not neutral and algorithms can discriminate. The legal world has not yet grappled with these challenges to workplace equality. The risks posed by data analytics call for fundamentally rethinking anti-discrimination doctrine. When decision-making algorithms produce biased outcomes, they may seem to resemble familiar disparate impact cases, but that doctrine turns out to be a poor fit. Developed in a different context, disparate impact doctrine fails to address the ways in which algorithms can introduce bias and cause harm. This Article argues instead for a plausible, revisionist interpretation of Title VII, in which disparate treatment and disparate impact are not the only recognized forms of discrimination. A close reading of the text suggests that Title VII also prohibits classification bias — namely, the use of classification schemes that have the effect of exacerbating inequality or disadvantage along the lines of race or other protected category. This description matches well the concerns raised by workplace analytics. Framing the problem in terms of classification bias leads to some quite different conclusions about how the anti-discrimination norm should be applied to algorithms, suggesting both the possibilities and limits of Title VII’s liability focused model.