美国人工智能模型训练合理使用认定的成案经验研究  

The American Experience of Measuring Fair Use in AI Model Training

作　　者：熊琦[1,2] 陈子懿[1,2] Xiong Qi;Chen Ziyi(Law School,Huazhong University of Science and Technology,Wuhan 430073,China;Research Center for Judicial Protection of Intellectual Property,Huazhong University of Science and Technology,Wuhan 430073,China)

机构地区：[1]华中科技大学法学院,武汉430073 [2]华中科技大学知识产权司法保护理论研究基地,武汉430073

出　　处：《科技与法律（中英文）》2024年第6期11-23,共13页Science Technology and Law（Chinese-English Version）

摘　　要：人工智能模型依赖于对大量作品的复制分析,从而导致传统版权业者与人工智能训练需求之间的紧张关系。目前模型训练合理使用争议的原因,在于著作权人对人工智能模型训练方式认识不足以及对技术驱动下新兴市场收益预期未能达成。在美国的司法实践中,合理使用已被广泛适用于从广播时代到互联网时代的诸多使用行为,在历史上具有典型意义的索尼案、谷歌数字图书案和坎贝尔案中给利用新技术的新兴产业拓展了发展空间。尽管美国法院在人工智能模型训练的合理使用问题上仍在继续要求各方补充证据,但其合理使用条款解释的丰富历史经验已经提供了诸多可供参考的答案,对我国调整版权产业与人工智能产业的关系具有参考意义。AI models rely on the replication and analysis of many works,which has led to conflicts between copyright holders and the demands of AI training.The reason for the current controversy lies in the lack of understanding of artificial intelligence model training and the failure of copyright owners to achieve profits in emerging markets driven by technology.In U.S.judicial practice,the fair use doctrine has been applied in areas such as software reverse engineer⁃ing and full-text digitization.This experience of interpreting new technologies in a way that allows room for develop⁃ment reflects a judicial tolerance for innovation.Although the United States has not yet come to a definitive conclusion on the controversy over AI model training,its historical experience suggests a trend towards adjudicating it as fair use.Such experience can help us balance the relationship between technological innovation and copyright protection,and reconcile the relationship between old and new industries.

关 键 词：人工智能 模型训练 合理使用 非表达替代 

分 类 号：D923[政治法律—民商法学]