人工智能版权训练数据来源合法性的规范构造
The Normative Construction of the Legality of Copyrighted AI Training Data Sources
摘要
人工智能模型开发者在使用他人作品作为训练数据进行模型训练时,应基于尊重在先权益原则,承担通过合法来源获取版权训练数据的注意义务.在鼓励AI开发者与版权人达成合作以激励更多训练数据高质量供给的规范目标下,版权训练数据来源合法性具有重要的实践意义,是版权合理使用"三步检验法"中"不得不合理地损害著作权人的合法权益"要件的重要判断因素.因此,AI开发者应在合理的技术成本范围内,积极承担版权训练数据获取的手段披露义务、模型训练目的的说明义务和数据溯源的辅助标识义务等三项基本义务,由此更好地提升版权训练数据的来源合法性.
When Artificial Intelligence(AI)developers utilize others' works as training data,they shall abide by the principle of respecting prior rights and assume the duty of care to obtain copyrighted training data from legitimate sources.Against the regulatory goal of encouraging cooperation between AI developers and copyright holders to stimulate the supply of more high-quality training data,the legality of the sources of copyrighted training data carries significant practical importance.This requirement serves as a key determining factor for the criterion"not unreasonably prejudice the legitimate interests of the rights holder"within the"three-step test"of the fair use doctrine.Accordingly,within reasonable technical cost limits,AI developers shall perform three fundamental obligations:disclosing the acquisition methods of copyrighted training data,clarifying the purposes of model training,and attaching auxiliary marks for data traceability.These practices can effectively improve the legitimacy of copyrighted training data sources.
年,卷(期):
2026, 41(4)
分类号:
D923.4
机标分类号:
D923
D22
G434
在线出版日期:
2026年8月12日
基金项目:
22JD820030:教育部人文社会科学研究项目,