算法决策系统公平性的事实逻辑与法律审视  

Factual Logic and Legal Examination of Fairness in Algorithmic Decision Systems

作　　者：张璐 ZHANG Lu(Guangming School of Journalism and Communication,China)

机构地区：[1]中国政法大学光明新闻传播学院,北京102249

出　　处：《大连理工大学学报（社会科学版）》2024年第6期86-95,共10页Journal of Dalian University of Technology(Social Sciences)

基　　金：教育部人文社会科学研究青年项目“生成式人工智能背景下网络内容治理的平台权责边界研究”(24YJC860033)。

摘　　要：算法决策系统的公平性成为治理算法中的关键问题,事实层面的公平性测量指标能否转化以及如何转化为法律标准,需要从法律逻辑和规范体系进行分析。数据性偏差、技术性偏差和社会性偏见是算法偏差的典型类型,反分类公平、群体公平、个人公平成为自动化决策公平性测量的重要指标。算法决策系统通过分析技术系统地实现“不可变性”,对个体的自主性带来影响,并在歧视意图、参照对象、劣势影响等方面存在认定困境。法律意义上公平的算法决策系统,具有价值判断、全过程公平、免责例外等特征,基于情境性的前置逻辑,应实现算法使用的必要性、算法设计的合规性和决策结果的可信度的三重审视。University of Political Science and Law,Beijing 102249,China The fairness of algorithmic decision systems has become a key issue in algorithm governance.Whether and how fairness measurement indicators on the factual level can be transformed into legal standards needs to be analyzed from the perspective of legal logic and normative systems.Data bias,technical bias,and social bias are typical types of algorithmic bias,while anti-classification fairness,group fairness,and individual fairness have become important indicators for measuring the fairness of automated decision-making.Algorithmic decision systems achieve‘immutability’through technical system analysis,affecting individual autonomy and presenting dilemmas in areas such as discrimination intent,reference objects,and adverse impacts.In the legal sense,a fair algorithmic decision system is characterized by value judgment,process fairness,and liability exceptions.Based on the contextual pre-logics,it should realize a threefold examination of the necessity of algorithm use,the compliance of algorithm design,and the credibility of decision results.

关 键 词：算法治理 算法歧视 算法影响评估 差别待遇 

分 类 号：D913[政治法律—民商法学]