生成式人工智能机器学习中的著作权风险及其化解路径

Copyright Risk in Generative Artificial Intelligence Machine Learning and Its Solution Path

作为一种基于机器学习的人工智能技术,生成式人工智能通过大规模数据集的学习训练来生成新的内容。生成式人工智能的快速迭代发展与机器学习的强数据依赖,在带来“创作”便利的同时,也给传统著作权理论带来了挑战。尤其是机器学习中的大规模素材训练所面临的著作权风险,亟待理论上的剖析厘清、司法实践上的准则确立。笔者从生成式人工智能的技术原理出发,通过梳理其素材来源与使用方式,分析机器学习素材训练中的典型侵权类型。并从理论基础、国内司法、欧盟立法等方面分析合理使用制度相对于法定许可制度的可行性。最终提出将生成式人工智能机器学习中的版权作品使用行为纳入合理使用制度范畴,既是当今科技发展浪潮中促进人工智能产业发展的最优解,也是维持“个人利益”与“公共利益”之间平衡的理性考量。

As an artificial intelligence technology based on machine learning,generative artificial intelligence(generative AI)generates new contents through the learning and training of large-scale data sets.The rapid iterative development of generative AI and the strong data dependence of machine learning,not only brings convenience to“creation”,but also brings challenges to the traditional copyright theory.In particular,the copyright risk faced by large-scale material training in machine learning needs to be clarified theoretically and established in judicial practice.Starting from the technical principle of generative AI,the author analyzes the typical infringement types in machine learning material training by sorting out the source and use of its material.It also analyzes the feasibility of the fair use system compared with the statutory licensing system from the aspects of theoretical basis,domestic justice and EU legislation.Finally,it is proposed that the use of copyright works in generative AI machine learning should be included in the category of fair use,which is not only the best solution to promote the development of artificial intelligence industry in the current wave of scientific and technological development,but also a rational consideration to maintain the balance between“personal interests”and“public interests”.Based on the technical principle of generative AI,this paper analyzes the typical types of infringement of copyright in machine learning material training by sorting out the source and use of material.The exemption analysis of non-expressive use and transformative use provides theoretical support for the copyright exception of machine learning material training.Then,on the basis of analyzing the progress of the relevant legislation and the evolution of the rules of the European Union,it puts forward and analyzes the feasible ways to solve the copyright risk.