机器学习算法的法律规制  

Legal Regulation for Machine Learning Algorithms

作　　者：崔聪聪[1] 许智鑫 CUI Congcong;XU Zhixin(Institute of Internet Governance and Law,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学法律系,北京100876

出　　处：《上海交通大学学报（哲学社会科学版）》2020年第2期35-47,共13页Journal of Shanghai Jiao tong University(Philosophy and Social Sciences)

基　　金：国家社会科学基金重大项目"国家网络空间安全法律保障机制研究"(13&ZD181)。

摘　　要：机器学习算法通常具有缺乏控制和不可预期的技术特点,在数据的分析预测过程中不可避免地产生不公平、歧视性的决策结果。通过梳理人工智能和机器学习的技术特点和实际应用,针对机器学习算法存在的安全风险,文章在分析已有的社会问题和法律困境的基础上,提出相适应的法律治理思路。机器学习算法的规制方案应以技术内容为核心,结合无效原则和算法影响评估制度,采用基于数据、代码和算法系统的不同路径。面对机器学习不透明性的特点,算法系统的监管者应当确立合理有效的监管方法,制定动态的算法影响评估制度,有效防范和降低安全风险,实现对机器学习算法应用最大限度的安全保障。Since technical characteristics of machine learning algorithms are often uncontrollable and unpredictable, unreasonable and discriminatory decision-making results are inevitably produced in the process of data analysis and prediction. By summarizing the technical characteristics and practical applications of artificial intelligence and machine learning, aiming at the security risks of machine learning algorithms, and based on the analysis of existing social problems and legal dilemmas, this paper proposes appropriate legal governance ideas. The regulation scheme of machine learning algorithms should focus on technical content, combine invalidation principles and algorithmic impact assessment, and adopt different schemes based on data, code, and algorithm systems. In response to the opacity of machine learning, the regulator of algorithm systems should formulate reasonable and effective supervision methods, develop a dynamic algorithmic impact assessment system, effectively prevent and reduce security risks, and achieve maximum guarantees for the application of machine learning algorithms.

关 键 词：机器学习算法 不透明性 无效原则 算法影响评估制度 

分 类 号：D910[政治法律—法学] TP181[自动化与计算机技术—控制理论与控制工程]