生成式人工智能训练中直播数据利用的著作权规制研究

Research on Copyright Regulation of the Use of Live Streaming Data in Generative Artificial Intelligence Training

摘要: 生成式人工智能的训练需要海量的数据，而网络直播数据作为其中的重要组成部分，便逐渐成为了训练重要的数据来源之一。此类数据通常同时具备视听作品等多重作品属性，而人工智能训练过程中不可避免的数据复制与处理行为，现实中极有可能落入著作权专有权的控制范围，从而引发侵权风险。现行合理使用制度因依赖个案裁量、对市场替代效应敏感且难以适用于大规模商业化利用，无法为AI训练提供稳定合规基础。在此背景下，引入法定许可制度被认为是一种更具制度可行性的解决路径，其允许AI开发者在支付合理报酬的前提下使用数据，并通过集体管理组织整合权利主体、设计差异化版税分配机制，以期在降低交易成本、避免“反公地悲剧”的同时，保障权利人的获酬权，实现技术创新与知识产权保护之间的平衡。

Abstract: The training of generative artificial intelligence requires massive amounts of data, among which online live streaming data has gradually become an important data source. Such data often possesses multiple layers of copyright-protected attributes, including those of audiovisual works. During the process of AI training, the unavoidable acts of data copying and processing are highly likely to fall within the scope of exclusive rights under copyright law, thereby giving rise to significant risks of infringement. The existing fair use doctrine, due to its reliance on case-by-case judicial discretion, sensitivity to market substitution effects, and limited adaptability to large-scale commercial data utilization, fails to provide a stable and predictable legal basis for AI training activities. Against this backdrop, the introduction of a statutory licensing regime has been proposed as a more institutionally viable solution. Under this framework, AI developers are permitted to use copyrighted data upon payment of reasonable remuneration, while collective management organizations are employed to consolidate fragmented rights and design differentiated royalty distribution mechanisms. Such an approach aims to reduce transaction costs, avoid the “anticommons tragedy”, safeguard the remuneration rights of copyright holders, and ultimately achieve a balance between technological innovation and intellectual property protection.