人工智能深度学习阶段的规制路径

The Regulatory Path of the Deep Learning Stage of Artificial Intelligence

摘要: 人工智能在生成内容之前需要海量作品的输入作为前提，可能侵犯到著作权人的专有权利，目前我国对此问题尚无立法规定。该问题亟待解决并应分类型讨论机器利用作品是否会侵犯著作权，对于非表达型的机器学习不会侵犯到著作权人的专有权利，而表达型机器学习就有可能侵犯著作权。对于表达型机器学习的规制路径，主流观点有合理使用和法定许可制度两种路径，合理使用制度会让人工智能生成内容产业与著作权人陷入两难的状态，因此建议用折中的法定许可制度兼顾二者利益。

Abstract: Artificial intelligence requires the input of many works before generating content, which may infringe the exclusive rights of copyright owners, and there is no legislation on this issue in China at present. The problem needs to be solved urgently and whether the use of works by machines will infringe copyright should be discussed by type. For non-expressive machine learning, it will not infringe the exclusive rights of copyright owners, while expressive machine learning may infringe copyright. The mainstream view on the regulation path of expressive machine learning is that there are two paths: fair use and statutory license system. The fair use system will put the AI-generated content industry and copyright owners in a dilemma, so it is suggested that a compromise statutory license system should be used to balance the interests of both.