人工智能时代算法性别歧视的类型界分与公平治理

The Classification and Fair Governance of Algorithmic Gender Discrimination in the Age of Artificial Intelligence

在人工智能时代,算法、大数据已经渗透到日常生活的方方面面,智能化正在急速改变我们的工作和生活方式。尽管算法使生活便捷高效,但算法作为一种新型社会权力,给女性的数字化生存带来广泛而系统的性别歧视风险。算法决策下的性别歧视包括镜像同构、耦合互动、关联交叉三种类型,相较于传统歧视行为具有高度隐蔽性、系统性和反复性特征,造成了放大性别歧视和结构性锁定女性不利地位的后果。算法性别歧视风险存在于算法全生命周期的各个环节。技术性偏见和社会性偏见的耦合交织,反歧视规范体系的应对失效以及软法体系缺位形成的“治理真空”使得歧视风险向实害结果转化。治理算法性别歧视,应以算法公平理念为指引、以算法生命周期为节点有效开展歧视风险管理,结构性扩容反歧视规范体系并培育公平向善的算法运行生态。

Algorithms,as a new type of social power,bring widespread and systematic risks of discrimination to women’s digital life.Gender discrimination under algorithmic decision-making mainly focuses on three types:mirror isomorphic discrimination,coupling interaction discrimination,associative and intersectional discrimination.Compared with off-line discriminatory behaviors,algorithmic gender discrimination is highly invisible,systematic,and repetitive,resulting in the amplified discrimination and locking women’s disadvantaged social position structurally.The risk of algorithmic gender discrimination may arise from all aspects of the algorithm’s entire life cycle,in which technical biases and social biases are intertwined.The ineffectiveness of non-discrimination law and the “governance vacuum” of the soft law system have transformed the risk of discrimination into the result of actual harm.Centering around the principle of algorithmic fairness,formulating risk management within the whole life cycle of algorithms,revising the anti-discrimination law respectively and cultivating an ecosystem for algorithm operation.