摘要
以ChatGPT、Sora、文心一言等为代表的生成式AI的问世,在极大提高人们生产力与创造力的同时,亦引发了诸多数据风险。基于生成式AI的技术原理,可将这些风险归纳为“三阶段风险”。在准备阶段,“用户供给”与“机器爬取”将引发数据非法获取风险;在运行阶段,“数据偏见”与“数据泄露”将导致数据安全风险;在输出阶段,则可能被非法分子利用诱发不良内容生成风险。对此,传统回应型治理与集中式治理面临着适用困境,难以满足复杂多变的治理需求。而敏捷治理凭借其适应性、多元性与包容性的特质,为应对生成式AI数据风险治理提供了有效方案。在敏捷治理的引导下,应构建数据风险全生命周期防控体系,完善数据安全协同治理机制。推进技术融通型法治,以有效纾解“三阶段风险”,赋能我国数字经济发展。
The emergence of generative AI,represented by ChatGPT,Sora and ERNIE Bot,has greatly improved people’s productivity and creativity,but also triggered many data risks.Based on the technical principles of Generative AI,these risks can be classified as“three-stage risks”.In the preparation stage,“user supply”and“machine crawling”will trigger the risk of illegal data acquisition during the operational phase,“data bias”and“data leakage”will lead to data security risks;in the output stage,it may be exploited by illegal elements to induce the risk of generating harmful content.In this regard,traditional responsive governance and centralized governance face application dilemmas,making it difficult to meet complex and ever-changing governance needs.Agile governance,with its adaptability,diversity,and inclusiveness,provides an effective solution for addressing Generative AI data risk governance.Under the guidance of agile governance,a data risk full-lifecycle prevention and control system should be established,and the collaborative governance mechanism for data security should be improved.It is necessary to promote the rule of law based on technological integration to effectively alleviate the“three-stage risks”and empower the development of China’s digital economy.
出处
《上海政法学院学报（法治论丛）》
2024年第6期84-100,共17页
Journal of Shanghai University of Political Science & Law(The Rule of Law Forum)
基金
2022年度国家社会科学基金项目“中国民法自主性知识体系的构建研究”的阶段性研究成果,项目编号:22VRC175。
作者简介
郑煌杰,中南大学法学院博士研究生。
