《电子技术应用》
您所在的位置:首页 > 其他 > 设计应用 > 算法影响评估制度的实践路径分析
算法影响评估制度的实践路径分析
网络安全与数据治理
王宇晓1,2,吴少卿2,张静怡1,2
1.中国信息通信研究院; 2.移动应用创新与治理技术工业和信息化部重点实验室
摘要: 算法影响评估制度作为国内外算法治理的基础性制度,是当前阶段算法治理的主要方案,但在应用实施过程中仍存在评估对象泛化、评估方法透明性不足的问题,需进一步细化。聚焦分析算法影响评估制度在应用实施过程中应关注的评估对象、评估内容和评估方法,提出在实际评估过程中应根据初步风险对算法评估对象分类分级,加强对以机器学习为基础的复杂算法的评估,重点关注算法的应用环节,重点评估算法输入的数据与特征、输出的目标与结果、干预与保障措施以及个人权利响应。
中图分类号:TN919;D922.16文献标识码:ADOI:10.19358/j.issn.2097-1788.2024.10.009
引用格式:王宇晓,吴少卿,张静怡.算法影响评估制度的实践路径分析[J].网络安全与数据治理,2024,43(10):57-61.
Practical solutions of algorithm impact assessment system
Wang Yuxiao1,2,Wu Shaoqing2, Zhang Jingyi1,2
1.China Academy of Information and Communications Technology;2.Key Laboratory of Mobile Application Innovation and Governance Technology, MIIT
Abstract: As the basic system proposed in the first stage of algorithm governance, the algorithm impact assessment system is the main solution for algorithm governance at the current stage. However, during the application and implementation process, there are still problems of generalization of assessment objects and insufficient transparency of assessment methods, which need to be further elaborated. This article focuses on analyzing the objects, assessment content and assessment methods that should be assessed during the application and implementation of the adjustment impact assessment system. During the actual assessment process, it is proposed to initially classify and grade the adjustment assessment objects based on risks, and strengthen the use of machine learning-based algorithms. Evaluation focuses on the application stage of the algorithm, focusing on the input data and feature complexity of the algorithm, the output goals and results, intervention and safeguard measures, and individual rights responses.
Key words : algorithmic impact assessment;algorithmic risk;algorithmic governance;classification and grading

引言

随着人工智能技术的迅速发展,算法广泛应用于社会生活的方方面面,与此同时,算法带来的安全风险引起了广泛关注[1]。面对日益攀升的算法治理压力,算法影响评估制度(Algorithm Impact Assessment)作为重要的治理方法被提上中外立法议程,成为当下算法治理实践中瞩目的焦点[2]。算法影响评估制度是指依据系统制定的衡量标准对算法自动化决策系统的系统设计、应用流程和数据使用等内容进行全面评估,以明确该系统的风险等级和影响范围的一种算法治理实践[3]。本文主要剖析在算法实际应用过程中,算法影响评估制度应重点关注的评估对象与范围、评估内容中的要点和细化方法,以及对推动算法影响评估制度落地提出相关建议。


本文详细内容请下载:

https://www.chinaaet.com/resource/share/2000006197


作者信息:

王宇晓1,2,吴少卿2,张静怡1,2

(1.中国信息通信研究院,北京100010;

2.移动应用创新与治理技术工业和信息化部重点实验室,北京100010)


Magazine.Subscription.jpg

此内容为AET网站原创,未经授权禁止转载。