摘要
《生成式人工智能服务管理暂行办法》以信息内容安全为规范重点,相关条款可分为绩效标准、设计标准、内部管理型标准。绩效标准设定了“提供者、使用者不得生成特定信息”等结果义务,不匹配生成结果的不可控性,合规成本畸高;设计标准明定提供者应采取的技术、行为,但部分设计标准却针对“发现违法行为、信息”这一或然事件缺乏针对性,内部管理型标准要求提供者自主选取提高信息质量的措施,可匹配生成式人工智能的不可控性、快速迭代性、大数据特性,但对有效执行提出较高要求。可通过转换标准类型、构建执行环境、以规制学习促进标准体系成熟等方法,提高生成式人工智能服务管理的有效性。
The Interim Measures for the Administration of Generative Artificial Intelligence Services focus on the safety of information content,and the relevant provisions can be categorized into performance standards,design standards,and internal management standards.Performance standards set obligations such as“providers and users shall not generate specific information,”highlighting the uncontrollability of mismatched generative results and the extremely high compliance costs.Design standards establish specific behavioral obligations,defining the technologies and behaviors that providers should adopt,with the premise of clear regulation targets.However,some design standards in the method target the contingent event of“discovering illegal behaviors or information”.Internal management standards set abstract behavioral obligations,requiring providers to autonomously take measures to improve information quality,matching the uncontrollability,rapid iteration,and big data characteristics of generative artificial intelligence,but relying on effective implementation.By transforming standard types based on applicable scenarios,constructing execution environments,and promoting the maturity of standard systems through regulatory learning,the conformity of standards can be improved.
作者
张学府
Zhang Xuefu(School of Political Science and Public Administration,China University of Political Science and Law,Beijing 100088)
出处
《中国行政管理》
北大核心
2024年第4期120-128,共9页
Chinese Public Administration
基金
国家社会科学基金重点项目“通用行政法典编纂问题研究”(编号:22AFX007)
中国政法大学科研创新项目“行政法法典化研究”(编号:21ZFD82001)。
关键词
生成式人工智能
绩效标准
设计标准
内部管理型标准
规制标准
generative artificial intelligence
performance standards
design standards
internal management standards
regulatory standards
