An Exploration of Training Data Compliance from a Copyright Law Perspective

Abstract

AI technology, including generative artificial intelligence such as data training, has become a key force in developing new quality productivity. Currently, the copyright controversy in large model training has attracted much attention both at home and abroad. Starting from the legality of data source at the data input end and the compliance of data use at the output end, the copyright risk of the big model training process can be comprehensively deconstructed. Diversified dispute resolution mechanisms such as the copyright exception system, the safe harbor principle and the collective management system have faced many challenges in practice, especially the traditional authorization and licensing model has shown obvious limitations in the face of the huge volume of data and the large number of right subjects. We analyze the trend of extra-territorial regulation and explore the feasible paths for solving data training copyright disputes in China, i.e., guided by the principle of balancing interests, promoting the construction of a consultation mechanism between the right subjects and AI enterprises; implementing a data labeling system; drawing on the open source licensing in the field of computers; and broadening the boundary of fair use.