摘要
生成式人工智能大模型语料训练是否构成侵权,有几种不同的解释论。通过对产业界、创作者、其他利益相关者等各方诉求分析,对比不同解决路径的优缺点,合理使用规则是解决该问题的最优路径。合理使用规则的适应性,不仅取决于其制度价值本身,还在于传统“四要素”和“三步检验法”分析框架已经发生的变化,即“转换性使用”在合理使用制度中具有了独立的地位。新技术时代,“变革性使用”又进一步超越了传统合理使用理论中对行为人主观意图的考量,转而关注使用行为在客观上是否推动了技术发展与社会进步。司法层面,合理使用路径适用于大模型语料训练具有比较优势,在当前法律框架下也具有可行性。诚然,也存在典型的不适用合理使用、应当被认定为侵权行为的情形。
There are several different interpretive theories on whether the use of training data in generative AI models training constitutes infringement to the original author.By analyzing the demands of various parties,including the industry,creators,and other stakeholders,and comparing the advantages and disadvantages of different solutions,the fair use rule is identified as the optimal path to address this issue.The adaptability of the fair use rule depends not only on its institutional value but also on the changes that have occurred in the traditional analysis frameworks of the“four factors”and the“three-step test,”notably the independent status that“transformative use”has gained within the fair use system.In the era of new technologies,“trans-innovative use”further transcends the traditional fair use theory's consideration of the actors'subjective intent and instead focuses on whether the use objectively promotes technological leaps and social progress.Legally,the fair use path has a comparative advantage in applying to large model corpus training and is feasible under the current legal framework.Admittedly,there are also typical situations where fair use does not apply and should be deemed infringing behavior.
出处
《中国版权》
2024年第6期5-26,共22页
China Copyright
