AI训练素材的“使用”是否构成著作权法意义上的使用？

PART.01：闭门缘起 理振闭门会
以ChatGPT为代表的新一代AI产品自问世以来，全世界真实地感受到智能和个性化的AI时代已经到来，《钢铁侠》中的贾维斯（J.A.R.V.I.S.）似乎在不久的将来会取代现在的“小爱同学”。在被新一代AI能力惊艳的同时，各国也看到了现行法律框架在AI监管上的空白。自ChatGPT发布以来，世界主要国家和地区相继发布了AI监管方面的立法计划、并密集出台了多个法律法规对AI行业进行管理。认为自己被AI产品侵权的法律主体也对AI产品的发布公司提起诉讼、尝试用现行法律制度解决AI侵权的困扰。
关注到这个快速发展的行业动向，刚满月不久的理振律师事务所于2023年9月1日举办了第一次闭门讨论会，理振合伙人李振武律师主持了会议，包含理振团队在内的16名法律业内人士和AI技术专家参加了本次讨论会。
On Friday, Li Zhen hosted its inaugural professional discussion since its establishment in July, drawing over ten senior legal professionals and AI enthusiasts to explore the legal challenges associated with the rapidly growing AI industry. The discussion began with Mr. Li Zhen Wu, the founder of Li Zhen, addressing recent AI-related conflicts worldwide and China's new AI legislation.
PART.02：核心观点 理振闭门会
应当把AI的素材使用分为训练和输出两个阶段。
训练阶段对素材的使用是否需要版权授权，
要看AI的使用是否属于版权上的法定“使用”行为。
在会议的开始，AI行业的技术人员熊先生和AI技术爱好者张研博先生首先介绍了AI发展史、模型分类和训练逻辑，为后面的法律角度讨论提供了依据。
接着，上海广播电视台版权中心的姚岚秋老师首先发言，分享了他对AI训练素材授权问题的思考。
姚老师分析道，“应当把AI的素材使用分为训练和输出两个阶段。训练阶段对素材的使用是否需要版权授权，要看AI的使用是否属于版权上的法定“使用”行为，如果非法定使用行为，即使在生活中被认为是使用了他人作品，也不存在版权授权的问题。假如AI训练素材的使用方式是类似人类阅读和学习的行为，难以构成法定的版权上的“使用”，因而无需授权；如果AI生成训练使得素材产生了新的复制件，则构成版权上的“复制”使用，接下来才要考虑该“复制”使用是否构成合理使用或者应当取得授权的问题。在输出阶段，必须要辨明的是AI生成物对于原训练使用的素材所借鉴的是思想还是表达？如果只是通过训练借鉴素材的理念、风格，没有直接在输出阶段呈现原素材的文字、线条、色彩、旋律、画面等具体表达，也不存在事先版权授权的问题。如果AI输出的生成物与原素材在表达上存在实质性相似，则可能涉及需要复制或改编或信息网络传播的授权问题。”
Mr. Xiong, an AI expert, and Mr. Zhang Yan Bo, an AI enthusiast, then provided insights into AI's history and methodologies. Mr. Li Zhen Wu raised the question of whether AI usage could qualify as "FAIR USE" within copyright laws in both AI training stage and output stage.
Subsequently, Mr. Yao Lan Qiu of Shanghai Media Group shared his perspective on defining AI usage, proposing two methods for determining whether it qualifies as fair use. The first approach distinguishes between AI usage during training and AI output. Mr. Yao analyzed that during the AI training stage, where AI learns from source materials similarly to a human, the usage could indeed be categorized as fair use under copyright laws. However, in the AI generation stage, the incorporation of external materials becomes the pivotal factor in defining fair use. Mr. Xiong and Mr. Zhang added that AI typically gets collect original materials and stores them in the cloud rather than on its servers. Mr. Yao then concluded that, in the AI generation stage, AI product owners should seek authorization before employing external materials, as "copying" constitutes one of the usages necessitating prior approval under copyright laws.
Mr. Yao's second method proposes viewing the AI training stage as a precursor to the AI generation stage, given that the ultimate purpose of AI products is content generation. This implies that AI owners should obtain authorization from the rightful owners of materials before usage. Mr. Yao also raised the question of precisely what AI derives from the materials it is fed: style or logic?
PART.03：其他观点 理振闭门会
北京盈科（上海）律师事务所合伙人、全国盈科文体传媒成员王慧娟律师对姚老师的目的论观点表示认同，并补充举例AI在音乐行业的应用，AI公司使用AI把热门歌曲进行拆解分析后得到的流行元素组合成的新歌曲成功打入流行歌曲排行榜。这种AI应用很难通过人耳或者其他方式发现与其他已发表作品的实质相似性，能经对歌曲创作者的就业市场和收入造成冲击。基于此，王律师提出AI在著作权应受到相应监管的观点。
Ms. Wang Hui Juan, a partner of Yingke Law Firm, concurred with Mr. Yao's
perspective regarding the determination of AI material usage based on the purpose of the AI product, in compliance with copyright laws. She shared examples from the music industry where entertainment companies employ AI to deconstruct hit songs into constituent elements to identify favored components. Subsequently, these favored elements are utilized to create new songs, often resulting in the popularity of these regenerated songs, thereby displacing creative artists from the industry. She emphasized the significance of valuing human creativity, advocating stringent regulations for the AI industry due to its potential for adverse consequences.
Following Ms. Wang's remarks, Mr. Yao delved into potential methods for supervising the AI industry, suggesting that copyright laws and potentially antitrust laws could play pivotal roles.
华东政法大学的李易凡老师同意姚老师和王律师的观点，并介绍了美国在AI监管方面的相关案例和趋势。郑律师补充了韩国的AI监管现状。
Ms. Li Yi Fan of East China University of Political Science and Law echoed her agreement with many of the opinions expressed and offered insights into AI industry supervision practices in the United States. Another scholar from East China University of Political Science and Law discussed the necessity of institutions capable of authorizing a large number of patents simultaneously and introduced AI supervision practices in Japan. Another lawyer contributed further insights into AI supervision practices in Korea.
姚老师补充道：“抛开版权问题，仅就AI对素材训练和输出之间的关系上看，训练阶段是输出阶段的必要前期准备，那么
训练阶段的学习行为应当与AI产品的研发目的相结合，两者共同利用了他人素材对外提供服务，包括商业服务。
从利益平衡的角度讲，应当对AI在各个阶段使用他人素材的行为予以监管，要求其给予必要的补偿。”
Mr. Yao voiced concerns that the waiting for People’s Congress to pass laws to restrict the AI industry might take years. He called upon relevant government departments to address this issue expeditiously.
理振的何美婕律师指出严格的AI监管会阻碍技术的发展。熊先生从AI从业者角度总结，AI是工具，使用AI侵害他人合法权益、压缩他人就业机会、生存空间的只会是AI使用者。
As the discussion continued regarding constraints on AI, Ms. He Mei Jie of Li Zhen
highlighted the potential drawbacks of overly stringent restrictions, which could impede China’s competitiveness in the AI era. Mr. Xiong agreed and added his perspective, contending that it is humans to be blamed as AI is a tool used by humans.
拥有五年药品专利工作经验的沈慧芝律师提出”AI在版权领域的属性研究，还应当考虑AI对知识产权其他细分领域的影响”，介绍了目前医药公司使用AI帮助药品研发的情况，并由此引出“AI不适宜作为专利法意义上的“人”来看待，因而也不适宜作为专利的发明人和权利人”的观点。
Ms. Shen, a pattern lawyer with ten years of experience in the pharmaceutical industry, suggested that patent laws could serve as a means to restrict the AI industry. She also provided examples of pharmaceutical companies using AI to produce medicines and then concluded that AI cannot assume the role of a patent holder.
来自复旦大学的硕士在读学生孙红同学提出“AI产品的研发和素材应该获得授权后再使用，著作权法上合理使用制度中的“人”难以经解释包含AI，但基于鼓励技术创新的考量，商业上授权协议的设计，是否必须以收费为对价及收费时间的安排，不妨探索更加灵活的安排。”
某互联网公司的涉外法务江女士和其他参会人员亦参加了讨论、分享了自己的观点。
Ms. Sun Hong, a postgraduate of East China University of Political Science and Law, emphasized that AI might can be considered the copyright holder under copyright laws as the laws don’t ban a program from being a copyright holder. Ms. Jiang, a postgraduate of Arizona State University and other lawyers of Li Zhen also shared their opinions.
PART.04：闭门结束 理振闭门会
经过近三个小时的热烈讨论，李振武律师用自己的参会感受结束了会议。
会后，参会人在理振办公室拍照留念，感到意犹未尽的参会人继续在理振办公室进行交流。
After nearly three hours of stimulating discussion, Mr. Li concluded the session with a concise summary and expressed gratitude to all participants for their valuable insights into the AI industry.
（本文英文非翻译件，系独立撰写）