大模型时代生成式人工智能治理的逻辑更新与回应路径——以《生成式人工智能服务管理暂行办法》为视角
Generative AI Governance in the Era of Large Models—Logical Update and Response Path from the Perspective of The Administrative Measures for Generative Artificial Intelligence Services
秦瑞标1
作者信息
- 1. 上海财经大学法学院,上海 200433
- 折叠
摘要
以大规模多种类训练参数、高效复杂算法及强大算力支持为基础的大模型已成为生成式人工智能的主要发展方向.相较于传统人工智能,大模型具备涌现性和通用性的技术特性,催生了"大模型+具体应用"的研发模式和"模型即服务"的服务业态.《生成式人工智能服务管理暂行办法》以风险分类分级、平台主体责任和算法透明度为内核的治理策略根植于传统人工智能时代,暂未能与大模型时代下的技术突破和产业变革充分适配.面向大模型的生成式人工智能治理应当在适应本土产业发展需求的基础上逐步探索基于既有制度框架的优化方案,通过以分层治理和敏捷治理弥补风险定级难题和僵化缺憾,以延伸扩展治理主体补全产业生态责任链条,以伦理治理补足模型透明度和可解释性不足的短板,进而实现治理逻辑的更新和治理路径的优化.
Abstract
Large model based on large-scale and multi-type training data,efficient and complex algorithms,and strong computing power support has become the main development direction of generative artificial intelligence.Compared with traditional artificial intelligence,the emergence and versatility of large models have given birth to the R&D model of"large models plus specific applications"and the service format of"model as a service".The Administrative Measures for Generative Artificial Intelligence Services which take a risk-based approach,platform governance approach,and algorithm transparency approach as the core governance strategies,are rooted in the traditional artificial intelligence era,and fail to adapt to technological breakthroughs and industrial changes in the era of large models.The generative AI governance adapted to large models should based on the existing institutional framework based on the development needs of local industries.The governance logic and Path should updated By using hierarchical governance and agile governance to make up for the difficulties and inflexibility of the risk category,by extending and expanding the governance role to complete the industrial ecological responsibility chain,and by ethical governance to make up for the lack of transparency and explainability of the model.
关键词
大模型/生成式人工智能/模型即服务/ChatGPT/《生成式人工智能服务管理暂行办法》Key words
large model/generative AI/model as a service/ChatGPT/The Administrative Measures for Generative Artificial Intelligence Services引用本文复制引用
出版年
2024
NETL
NSTL
万方数据