摘要
生成式人工智能(GAI)预训练容易引发版权法律困境,根源在于海量训练数据需求与著作权专有属性之间的结构性矛盾。现行《著作权法》列举的合理使用情形难以包括GAI预训练行为,而法定许可虽具理论自洽性,却因海量作品识别困难、付酬机制缺失等现实障碍难以落地。未来制度构建应以合理使用为优先选择,在立法完善方面可考虑以下三条路径:其一,借鉴欧盟及日本的立法经验,直接在《著作权法》第24条中增设以GAI数据训练为典型场景的合理使用条款;其二,若对现行《著作权法》的合理使用制度进行结构性改造,可以引入美国版权法中的“四要素”判断标准,建立针对非表达性使用的合理使用认定机制;其三,若选择保持《著作权法》不变,则可以第24条中的“兜底条款”为抓手,通过行政法规对GAI预训练的合理使用情形作出特别规定。路径二与路径三为未来向法定许可规制模式的转型预留了制度空间。
The pre-training of Generative Artificial Intelligence(GAI)is prone to triggering copyright legal dilemmas,rooted in the structural contradiction between the demand for massive training data and the exclusive nature of copyright.The current Copyright Law’s enumerated circumstances of fair use hardly encompass GAI pre-training activities.While statutory licensing theoretically offers coherence,it faces practical obstacles such as the difficulty in identifying vast numbers of works and the absence of a remuneration mechanism,making it hard to implement.Future institutional development should prioritize fair use as the preferred approach.Legislative improvements could consider the following three pathways.First,drawing on the legislative experiences of the European Union and Japan,a fair use clause specifically targeting GAI data training could be directly added to Article 24 of the Copyright Law.Second,if a structural reform of the current Copyright Law’s fair use system is pursued,the“four-factor”test from U.S.copyright law could be introduced to establish a fair use identification mechanism for non-expressive uses.Third,if the Copyright Law remains unchanged,the“catch-all clause”in Article 24 could be leveraged to enact special provisions on fair use for GAI pre-training through administrative regulations.The second and third pathways reserve institutional space for a future transition to a statutory licensing regulatory model.
作者
王琦
甘红
Wang Qi;Gan Hong(Anhui Vocational College of Police Officers,Hefei,230031,Anhui)
出处
《湖南警察学院学报》
2025年第6期102-111,共10页
Journal of Hunan Police Academy
基金
2024年度安徽省科研计划编制项目“人工智能生成内容版权问题研究”(2024AH052366)。
