生成式人工智能风险治理元规则研究  

On the Meta-rules for Risk Governance of Generative Artificial Intelligence

作　　者：商建刚 Shang Jiangang

机构地区：[1]上海政法学院经济法学院

出　　处：《东方法学》2023年第3期4-17,共14页Oriental Law

基　　金：国家社会科学基金重大项目“构建人类命运共同体国际法治创新研究”(项目批准号:18ZDA153);上海市教育委员会人文社科重大项目“创新驱动发展战略下知识产权公共领域问题研究”(项目批准号:2019-01-07—00-07-E00077)阶段性研究成果。

摘　　要：生成式人工智能由算法与数据驱动,属于深度合成技术,在国家安全层面存在引发暴力冲突、意识形态渗透风险,在社会层面存在数据垄断、道德操纵风险,在私权利层面存在增加犯罪、挖掘用户隐私等风险。然而,生成式人工智能的责任承担机制不明、现有制度分散、不成法律体系;理论界尚未针对生成式人工智能治理进行专门研究,治理范式缺失。构建生成式人工智能风险治理体系,应构建生成式人工智能风险治理的元规则。从人工智能的治理原则视角切入,通过梳理不同国家、地区、组织、企业所构建的人工智能治理原则,提炼其共性,是生成式人工智能风险治理元规则形成的可行进路。生成式人工智能风险治理元规则由协同共治、提升透明度、保障数据质量以及伦理先行四个方面组成。Generative AI is driven by algorithms and data,and belongs to deep synthesis technology.It may bring threats of violent conflicts and ideological penetration with respect to national security,risks of data monopoly and moral manipulation in the society and risks of crimes and mining users'privacy in the regard of private rights.However,the responsibility mechanism of generative artificial intelligence is unclear,the existing system is scattered and is not a unified legal system;theoretical studies have not been conducted specifically for the governance of generative AI,and the governance paradigm is absent.In order to construct a generative artificial intelligence risk governance system,the meta-rules on the risk governance of generative AI should be proposed.From the perspective of AI governance principles,the AI governance principles constructed by different countries,regions,organizations and enterprises have been sorted out,and the common ground has been found and thus a feasible set of meta-rules for generative AI application risk governance has been proposed.The meta-rules on the risk governance for generative AI applications consist of four aspects:collaborative governance,enhancing transparency,ensuring data quality and priority on ethics.

关 键 词：ChatGPT 生成式人工智能 数据风险 元规则 算法 大型语言模型 

分 类 号：DF0[政治法律—法学理论]