Like Human, Like Algorithm: Responses to Algorithmic Discrimination Among Individuals From Protected Classes
研究发现，当算法被认为像人类一样进行社会分类时，受保护群体成员（如少数族裔）会对其产生更负面的反应，这一效应在非代表性训练数据、代理分类规则等特征下均成立。
Abstract Algorithms, commonly used in business practice, often discriminate against members of protected classes (e.g. racial minorities). Previous research findings suggest that individuals, including those from protected classes, under some circumstances, may not respond negatively to discriminatory algorithms. Other evidence suggests the opposite. Given the conflicting evidence, there is an opportunity to understand how and when protected class members respond to businesses that employ algorithms when these algorithms make predictions or decisions resulting in discrimination. Drawing on an empirical package comprising one secondary data study and four experiments, our research demonstrates that when algorithms are perceived to engage in human‐like social categorization, they elicit more negative responses from members of protected classes. This effect is observed across various algorithm features, including nonrepresentative training data, proxy classification rules and non‐statistical classification rules. The research's findings extend the literature on algorithmic discrimination and business ethics, providing suggestions to mitigate algorithmic discrimination and improve societal well‐being.