生成式人工智能預(yù)訓(xùn)練中著作權(quán)合理使用研究

打開(kāi)文本圖片集

中圖分類號(hào)：D923.4 文獻(xiàn)標(biāo)志碼：A 文章編號(hào):1003-5168(2025)14-0117-05

DOI:10.19968/j.cnki.hnkj.1003-5168.2025.14.023

Research on Fair Use of Copyright in Generative AI Pre-training

HOU Xianjie (Law School/Intellectual Property School, Zhongyuan University of Technology, Zhengzhou 45ooo7,China)

Abstract: [Purposes] The pre-training process of generative artificial intelligence involves large-scale utilization of copyrighted works. Under China's current Copyright Law and related legal framework,entities engaged in such utilization face prohibitive transaction costs and potential copyright infringement risks.This paper aims to explore feasible solutions that align with the developmental requirements of China's generative AI industry while addressing these legal challenges.[Methods] Through an examination of technological principles underlying generative AI pre-training phase,this study deconstructs data processing workflows and systematically categorizes potential infringement types in data transcoding, tagging, organization， and aggregation phases using an input-side infringement analysis framework.A comparative legal analysis evaluates the operational eficacy of three regulatory approaches: licensing agreements,statutory licensing,and fair use provisions.[Findings] The findings demonstrate that conventional licensing agreements and statutory licensing mechanisms inevitably incur substantial transaction and administrative costs.While fair use provisions struggle to provide effective copyright infringement defenses for generative AI trainers,the open-ended provisions under Article 24 of China's current Copyright Law preserve implementation flexibility through its residual clause.[Conclusions] This study recommends China capitalize on the ongoing amendments to the Implementing Regulations of the Copyright Law to establish a "Fair Use for Generative AIPre-training" clause,explicitly defining its subject qualifications,protected objects,permissible purposes,and behavioral criteria. Keywords: generative artificial intelligence; data training; copyright; fair use

0 引言

近年來(lái)，生成式人工智能（GenerativeArtificialIntelligence，GAI)發(fā)展迅猛，其強(qiáng)大的學(xué)習(xí)能力、交互能力和創(chuàng)造能力，給人類社會(huì)生產(chǎn)和生活方式帶來(lái)了巨大改變，也標(biāo)志著人工智能時(shí)代的到來(lái)。（剩余7332字）