生成式人工智能训练使用作品的许可复合机制研究

Research on Multi-type Licensing Mechanism of Utilizing Copyrighted Works for Generative AI Training

导出

摘要合理使用制度无法完全规范生成式人工智能训练使用作品行为,需正视许可制度对增进作品效用的价值。在作品是模型训练成功的基础资源、使用多元化作品可避免大模型崩溃以及作者对作品的贡献等因素的考量下,确有通过许可制度补偿作者表达成本的必要性。引入“卡一梅框架”可解决财产规则与作品需求高效性、责任规则与权利自治性之间的冲突,通过建立作品许可复合机制实现作品许可在成本与效率上的平衡。具体而言,责任规则下经法释义学解释的法定许可,可成为生成式人工智能训练使用作品的合法性基础。明确法定许可的临时安排性质和引入多元化许可费标准,可满足现阶段行业对使用作品的刚性需求。伴随生成式人工智能信息披露义务的法定化,许可机制遵循财产规则还需借助著作权集体管理组织。该路径既可以降低传统许可机制在实现收益上的成本,也可以解决因互联网全球性特征导致的单一法定许可无法覆盖全场景许可市场的难题。 The fair use doctrine proves inadequate in comprehensively regulating the utilization of copyrighted works for generative AI training,and it is necessary to recognize the value of licensing system in enhancing work utility.Given that copyrighted works constitute foundational resources for model training,that diversified work usage prevents model collapse,and that authors'creative contributions warrant compensation,it becomes imperative to establish licensing mechanisms to offset creators'expressive costs.The introduction of the C&M Framework resolves conflicts between property rules and operational efficiency,as well as between liability rules and rights autonomy,enabling balanced cost-efficiency optimization through composite licensing mechanisms.Specifically,a statutory licensing system interpreted through legal hermeneutics under liability rules may serve as the normative foundation for generative AI training.Clarifying the provisional nature of statutory licensing arrangements and implementing a multi-tiered royalty framework would address the industry's demands for copyrighted works utilization under current circumstances.As statutory disclosure obligations for generative AI materialize and licensing mechanisms evolve toward property rules,copyright collective management organizations could effectively reduce transaction costs in royalty distribution while addressing the global licensing challenges inherent in internet-based applications that surpass the jurisdictional limitations of unitary statutory licenses.

作者倪朱亮 Ni Zhuliang

机构地区西南政法大学民商法学院

出处《法律科学(西北政法大学学报)》北大核心 2025年第4期90-103,共14页 Science of Law:Journal of Northwest University of Political Science and Law

基金国家社科基金一般项目(23BFX081)“数字时代互联网平台算法推荐的著作权问题研究” 西南政法大学知识产权保护协同创新中心2022年度开放课题(22CICIPP002)“数字经济时代互联网平台不正当竞争行为规制研究”。

关键词生成式人工智能许可复合机制卡一梅框架法定许可著作权集体管理组织 generative AI multi-type licensing mechanism C&M Framework statutory permission copyright collective management organization

分类号 D923.4 [政治法律—民商法学]

作者简介倪朱亮,西南政法大学民商法学院副教授。