人工智能训练的版权困境及其出路:模块化许可机制探析  

The Copyright Dilemma of AI Training and Its Solution:An Exploration of the Modular Licensing Mechanism

作　　者：孙靖洲 Sun Jingzhou

机构地区：[1]北京大学法学院 [2]慕尼黑大学

出　　处：《知识产权》2024年第11期94-111,共18页Intellectual Property

摘　　要：创作者对人工智能利用其作品进行训练的抵制,缘于利益分配机制付之阙如。在调整因新技术带来的作品使用形式变化所引发的新的社会关系时,既要确保创作者能公平地参与到由创作带来的收益分配中,维护其劳动尊严和劳动收入,又要防止版权人通过杠杆优势制约技术发展。可考虑从知识产权制度为解决市场失灵而创设的四种特殊许可模式中汲取经验,为人工智能训练建立一套整体协调但内部区隔的模块化授权许可机制:大型人工智能企业应尽最大努力获取授权,主动建立版权许可机制,与版权人分享收益;中小企业同时面临被大型内容平台拒绝许可与缔约成本高的双重困境,应要求大型内容平台作出以公平、合理、无歧视方式进行授权的声明,并发挥其中介组织的优势,保障个体创作者的合法利益、企业获得充足训练语料,同时防止掌握海量内容数据的大型内容平台封锁人工智能产业。Creators resist AI training on their works due to the lack of a fair benefit-sharing mechanism.To address new social dynamics from technological changes,it is vital to ensure creators share in the profits fairly,while preventing copyright holders from stifling innovation.A modular licensing system can draw from four special licensing models in intellectual property law addressing market failures.Large AI companies should make efforts to obtain licenses,set up copyright licensing systems,and share profits with creators.For SMEs facing both refusal of licenses by large content platforms and high contracting costs,platforms must offer fair,non-discriminatory licensing.Intermediary organizations could help safeguard individual creators'rights,ensure companies have sufficient training data,and prevent large platforms from monopolizing the AI industry.

关 键 词：生成式人工智能 合理使用 法定许可 AI投喂 训练数据可携带权 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] D913[自动化与计算机技术—控制科学与工程]