摘要
通用人工智能模型具有涌现能力与通用能力,构成人工智能研发应用的关键技术底座,也引发专门的规范需求.我国目前对大模型研发应用活动的法律调整,或侧重于结果问责,或过于强调算法模型备案,未必适应通用大模型的技术特点与发展属性.考虑到大模型并不等同于人工智能系统,对大模型的规范不应责之过苛,应主要通过信息披露制度实现风险预防,具体包括要求大模型提供者编写并提供技术文档、制定并公布著作权保护政策以及发布已公开个人数据处理报告.此外,法秩序还应通过行为准则、监管沙箱与真实测试规则等治理工具建立容错机制,激励通用人工智能模型的技术迭代与产业推广.通用人工智能模型具有系统性风险的,应额外借助科技系统的科技伦理机制(如伦理委员会)与法律系统的风险管理制度(如红队测试等),实现安全与发展之间的动态平衡.
General-purpose artificial intelligence models exhibit emergent intelligence and general capa-bilities,forming the foundational technological base for the development and application of artificial intel-ligence(AI),while giving rise to specific regulatory demands.Currently,China's legal regulation of large model development and application activities tends to focus either on outcome accountability or to o-veremphasize algorithm model filing.This approach may not align with the technical characteristics and developmental nature of large models.Given that a general-purpose AI model is not equal to an AI sys-tem,regulation of large models should not be overly stringent.Instead,risk prevention should be a-chieved through an information disclosure system that requires large model providers to create and provide technical documentation,formulate and publish copyright protection policies,and release reports on the processing of publicly available personal data.Furthermore,the legal order should establish a fault-toler-ant mechanism through behavioral guidelines,a regulatory sandbox and real-world testing rules,so as to incentivize technological iteration and industrial adoption of large models.Where general-purpose AI models pose systemic risks,the legal order should additionally leverage the scientific and ethical mecha-nisms of the technological system(such as ethics committees)and the risk management institutions of the legal system(such as red-teaming tests,etc.),in order to achieve a dynamic balance between security and development.