作者：王莹
作者介绍：中国人民大学法学院未来法治研究院、刑事法律科学研究中心副教授，法学博士
出版日期：2022年6月9日
内容提要 算法决策的主体多元性、开放性、不透明性、自主性等技术特征给算法侵害归责带来挑战，有必要在检验与拓展传统侵权法与刑法责任框架基础上，结合算法技术特征构建一个层级式的算法责任框架。对于产生具体侵害结果的算法损害归责而言，分别设置因果支配型的责任与非因果支配型的义务型责任。前者是指针对具有理解与控制技术可能性的算法适用过错责任与严格责任，后者是指针对不具有理解与控制技术可能性的算法适用义务型责任，为算法设计者与应用者设置避免算法侵害风险现实化的义务或对用户、社会公众法益的保护义务及相应责任。对算法妨害而言，因无具体权利侵害结果因而不存在结果归责问题，仅需对该算法妨害行为加以禁止，设置相应行政违法行为，或设置类似侵权法的公共妨害甚或刑法上的抽象危险犯进行追责。
关键词 算法侵害 算法妨害 算法损害 义务型责任
Abstract: The technical characteristics of algorithmic decision-making, such as subject pluralism, openness, opacity, and autonomy, pose challenges to the attribution of algorithmic harm. It is necessary to construct a hierarchical algorithmic liability framework based on the testing and expansion of traditional tort law and criminal law liability framework, combined with the technical characteristics of the algorithm. For the attribution of the algorithmic harm that causes concrete harm outcome, causal dominant liability and non-causal dominant duty liability or duty-violating liability should be set up respectively. The former refers to the application of fault liability and strict liability to the algorithm which is explainable and controllable according to its technological characteristics, and the latter refers to the application of duty liability to the algorithm which is unexplainable and uncontrollable according to its technological characteristics, which sets the duty to avoid the realization of algorithm harm risks or the duty to protect the legal interests of users and the public and corresponding liabilities for algorithm designers and deployers. For the algorithmic nuisance, because there is no specific result of right infringement, there is no problem of result attribution, and it is only necessary to prohibit the algorithm nuisance, set up corresponding administrative delicts, or set up the public nuisance similar to tort law or even abstract dangerous offense in criminal law to be held accountable.