算法决策的结构化问责机制:欧盟经验与中国借鉴

Structuralized Accountability in Algorithmic Decision Making: The EU Experience and its Implications in China

算法决策由于不透明和潜在歧视等风险而引发法律问责难题,法学界逐渐形成对算法决策综合治理的趋同共识,但既有的问责理论未能充分实现有机协作的结构化问责目标。算法决策的结构化问责是以算法正义为纽带,实现纵横维度上的问责要素联动的问责模式。欧盟算法结构化问责在横向维度上要求满足信息层面的透明度、论辩层面的可解释性、达成价值层面的算法正义框架共识并且可能面临一定的法律后果,在纵向维度上要求实现结构化的规制理论,并且实现行政机构与私人机构之间的协同监管。深入分析欧盟算法决策的结构化问责机制,对充实我国算法决策的问责框架具有启发意义。建构我国算法决策的结构化问责机制,需要在横向维度上明确透明度要求的构成要素和层级,区分信息层面的透明度要求和论辩层面的可解释性要求,建构形成算法正义共识的法学方法论并融合既有的社会正义理论,在纵向维度上需要完善结构化的监管体系,并实现行政机构与私人机构之间的协同监管。

Algorithmic decision making has caused accountability problems due to risks such as opacity and potential discrimination. In this regard, legal scholars have gradually reached a consensus on comprehensive regulations on algorithmic decision-making, but the existing accountability theories fail to achieve the goal of structuralized accountability, which requires organic combination of different regulation methods. The structuralized algorithmic accountability is an accountability model that takes algorithmic justice as a link and realizes the linkage of accountability elements in the vertical and horizontal dimensions.In the horizontal dimension, the EU’s algorithmic structuralized accountability requires transparency in the information hierarchy, explainability in the argumentation hierarchy, consensus on the algorithmic justice framework in the value hierarchy and may face certain legal consequences. In the vertical dimension, it requires regulation theories to realize structuration and co-regulation between administrative and private agencies. The in-depth analysis of EU’s structuralized accountability in algorithmic decision making is instructive to enrich our algorithmic accountability framework. To construct our mechanism for algorithmic structuralized accountability, in the horizontal dimension, it is necessary to clarify the components and hierarchies of transparency requirements, distinguish the transparency requirements in the information hierarchy and the explainability requirements in the argumentation hierarchy, construct a legal methodology to build a consensus on algorithmic justice and integrate existing social justice theories;in the vertical dimension, it is necessary to improve the structuralized regulation system and realize the co-regulation between administrative and private agencies.