自动化行政中算法的法律控制  

Legal Control of Algorithms in Automated Administration

作　　者：王宾 Wang Bin

机构地区：[1]北京大学法学院

出　　处：《财经法学》2023年第1期61-75,共15页Law and Economy

摘　　要：自动化行政中的算法可分为转译型算法和自我学习型算法。算法的运用面临合法性危机:处于私主体地位的算法设计师将法律语言转译成机器语言时会嵌入自身的判断,带来改写法律的风险;算法决策有时在事实上超出法律授权范围,且缺乏畅通的救济机制。算法的合法性控制方式应与算法类型适配。针对转译型算法,需结合算法的性质以及技术特点,从转译主体、所译法律的明确性、转译过程的透明度等方面施以控制。针对自我学习型算法,首先应当确立“民主—科学”的合法性框架,其次应当从建立算法信任的角度,围绕保障公众主体地位和算法科学性对算法进行控制。Algorithm in automated administration can be divided into translation algorithm and self-learning algorithm, and the use of algorithm is faced with legitimacy crisis. Algorithm designers in a private position embed their own judgments in translating legalese into machine language, creating the risk of rewriting the law. In addition, algorithmic decision-making sometimes exceeds the scope of legal authority in fact, and lacks the smooth relief mechanism. The legality control method of the algorithm should be adapted to the algorithm type. It is necessary to control the translation subject, the clarity of the translated law and the transparency of the translation process in combination with the nature and technical characteristics of the translation algorithm. For self-learning algorithm, we should first establish a“democracy-science”legitimacy framework, and the algorithm should be controlled from the perspective of building trust in the algorithms by guaranteeing the status of the public and the scientific nature of algorithms.

关 键 词：自动化行政 形式合法性 行政民主 行政科学 算法信任 

分 类 号：D922.1[政治法律—宪法学与行政法学]