大语言模型发展的版权掣肘及其治理——以美国版权集体诉讼的实践为视角

The Copyright Constraint and Governance of the Development of Large Language Model——From the Practice of Class Copyright Action in the United States

生成式人工智能大语言模型的训练和应用在美国引起大规模版权集体诉讼,版权问题成为大语言模型发展的首要障碍。在风险类型上,大语言模型版权集体诉讼引发了伴生性的法律风险、经济风险以及创新风险。以公开听证凝聚社会共识,以能动司法缓冲版权矛盾,以行业自治消弭版权风险,是美国应对大语言模型版权集体诉讼的三种重要策略,体现了创新优先的治理思路。我国大语言模型正处于研发和应用的关键阶段,目前已面临类似风险。美国经验作为“前车之鉴”,对我国大语言模型版权治理具有重要的参照意义。我国的版权治理对策在宏观层面应当坚持“创新优先”的价值导向,在中观层面应当采取“多元协同”的治理模式,在微观层面则应当引入“人工智能训练合理使用”类型。

The training and application of generative artificial intelligence large language models have caused large-scale copyright collective lawsuits in the United States and other regions,and copyright issues have become the primary obstacle to the development of large language models.In terms of risk types,the copyright class action lawsuit of the large language model has triggered accompanying legal risks,economic risks,and innovation risks.The three important measures taken by the United States to respond to collective lawsuits over copyright in large language models are to consolidate social consensus through open hearings,buffer copyright conflicts through active judiciary,and eliminate copyright risks through industry autonomy,reflecting the innovative governance approach of prioritizing innovation.The big language model in our country is currently in a critical stage of research and application,and is facing similar copyright risks.The experience of the United States serves as a valuable reference for copyright governance in China’s large language models.Our country’s copyright governance measures should adhere to the value orientation of“innovation first”at the macro level,adopt a governance model of“diverse collaboration”at the meso level,and introduce the type of“reasonable use of artificial intelligence training”at the micro level.