学界有关算法歧视规制的研究,欠缺算法歧视的认定这一逻辑前提,导致反算法歧视的规制范围过宽,算法应用与算法正义之间的矛盾被过分夸大,有必要明确算法歧视的认定规则。算法歧视以横向权力关系的存在为必要前提,不存在横向权力关系则算法歧视不成立。当算法控制者以用户固有的、难以改变的特性对其进行分类,并单方决定适用于用户的算法规则,用户只能被动接受决策结果而无退出算法系统之自由时,横向权力关系形成。在算法歧视的具体认定上,算法歧视以区分决策为行为要件,以用户遭受差别不利对待为结果要件。算法控制者若能证明差别对待具有合理性,则不构成算法歧视。搜索算法、人机对话算法对社会歧视的单纯呈现,以及社会经济发展、算法应用普及不平衡导致的算法红利分配失衡,均因欠缺区分决策要件,不构成算法歧视;“大数据杀熟”、评分算法所造成的差别对待并非必然构成歧视,须进行个案分析。
Determination of Algorithmic Discrimination
Ning Yuan (Wuhan University)
Abstract Existing research on anti-algorithmic discrimination lacks the logical premise of discrimination determination, which leads to the generalization of algorithmic discrimination and needs to be corrected. The contradiction between algorithmic application and algorithmic justice has been over-emphasized, which requires a clear determination rule for algorithmic discrimination. Horizontal power relations are a necessary field for algorithmic discrimination: algorithmic discrimination only exists when there is horizontal power. The horizontal power relationship is formed when the algorithm controller classifies users with locked distinction, unilaterally decides the algorithm rules applicable to users, and users can only passively accept the decision results without the freedom to exit the algorithm system. In terms of specific elements, algorithmic dis‐ crimination is based on the act of differentiation, and on the result of differentiation and disadvantage. Algorithmic controllers are exempt from anti-discrimination regulation by justifying differential treatment. The pure presentation of social discrimination in search algorithm and human-machine dialog algorithm as well as algorithm dividend distribution imbalance caused by social-economic development and imbalanced application of algorithm does no lead to algorithmic discrimination. Differential treatment caused by profiteer on big data and scoring algorithm does not necessarily constitute discrimination and should be analyzed case by case. Key words algorithmic control; algorithmic discrimination; horizontal power; distinguishing decision; differential adverse consequence; profiteer on big data; scoring algorithm
■ 作者简介 宁 园,法学博士,武汉大学法学院特聘副研究员;湖北 武汉 430072。