摘要
基于大数据与机器学习的第三代算法推动政府部门自动化行政变革.算法行政放大算法常规风险的同时,也凸显出行政领域的特殊风险.为规范算法行政活动,欧美地区实施了第一波算法行政问责政策.每一政策都反映为一种或多种问责机制,以运行机制和问责主体为标准从中析出科技伦理问责、算法影响评估、透明度机制、行政程序法问责、审计/监管检查、独立问责机构、禁令/暂停措施、政府采购等八种问责机制,分析机制运行假设和实施中反馈的问题.结合行政问责理论对欧美算法行政问责政策进行总体评估,发现现有政策在信息提供、质询强度和有效激励等方面存在不足.政府数字化转型背景下,我国算法行政问责制度需要体系化展开,结合我国现状和欧美经验教训,提出优化算法行政问责制度体系的建议.
Abstract
Third-generation algorithms based on big data and machine learning drive automated administrative change in government departments.Algorithmic administration amplifies the conventional risks of algorithms while also highlighting the special risks in the administrative field.To regulate algorithmic administrative activities,the first wave of algorithmic administrative accountability policies have been implemented in Europe and the United States.Each policy is reflected in one or more accountability mechanisms,from which eight accountability mechanisms are analyzed based on the criteria of operation mechanism and accountability subject,such as scientific and technological ethics accountability,algorithmic impact assessment,transparency mechanism,administrative procedural law accountability,auditing/regulatory inspection,independent accountability body,ban/suspension measures,government procurement,etc.The operation assumptions of the mechanisms and the feedback problems in the implementation are analyzed.An overall assessment of the algorithmic administrative accountability policies in Europe and the United States in combination with administrative accountability theory reveals that the existing policies are deficient in terms of information provision,questioning intensity and effective incentives.Under the background of government digital transformation,China's algorithmic administrative accountability system needs to be systematically unfolded.Combining China's current situation and the lessons learned from Europe and the United States,we put forward suggestions for optimizing the algorithmic administrative accountability system.Firstly is to improve the level of algorithmic administrative information provision,and secondly is to strengthen the questioning intensity of algorithmic administrative accountability,and thirdly is to improve the multifaceted incentives and constraints framework.
NETL
NSTL
万方数据