CHINA Trends and Developments Contributed by: Qiang Ma, Jingtian & Gongcheng
mined that it constituted the plaintiff’s work. From the Court’s reasoning, it is evident that the Court effectively regarded Midjourney, a generative AI tool, as analogous to typical drawing tools and therefore merely a tool for human creative use. It is worth noting that these two courts are pri - mary people’s courts and do not necessarily rep - resent broader judicial practice in China. There are currently no cases from higher-level courts. Several scholars have raised concerns about the court’s rulings. They first argue that only humans can be considered authors and AI-generated content is clearly not a human-created work. Their specific reasons are as follows. • According to the idea-expression dichotomy, only the expression itself is protected under copyright law. When using generative AI, it is not the user but the AI that substantially determines the expressive elements that make up a piece of work. • Different types of artistic works have distinct forms of expression and cannot be directly transformed. So a person cannot create a visual artwork simply by inputting text. Training AI with copyrighted works In 2024, four illustrators filed a lawsuit against the developers of the generative AI tool, Trik. They claimed that the developers used their copyrighted works without their consent during the training of Trik, infringing upon their rights to reproduction, adaptation and other rights that should be held by the copyright owners. This case is the first in China to discuss whether using copyrighted works to train AI constitutes copyright infringement. It is being heard by the Beijing Internet Court and is still pending.
According to the Copyright Law currently in force, it is difficult to determine that using (espe - cially reproducing) others’ copyrighted works during training constitutes fair use. However, there have been some precedents where the courts do not strictly follow the relevant fair use provisions in their judgments. Some scholars hope to amend the legislation to define this behaviour as fair use, while others argue that a balance should be struck between the interests of authors and generative AI devel - opers, advocating for statutory licensing and other solutions. Liability for infringing AI-generated outputs Unlike the previous issue in the input phase, the concern here is the potential infringement in the output phase. When prompted appropriately, these AI models may produce outputs that are substantially similar to a particular copyrighted work. This was the exact situation in the Ultra - man case. In that case, the plaintiff was the rights holder of the Ultraman works and the defendant was the provider (but not the developer) of the genera - tive AI tool, Tab. The plaintiff discovered that by entering Ultraman into the defendant’s Tab web - site, they could generate images identical to the plaintiff’s Ultraman works. When they entered “Ultraman with long hair”, the generated image retained the distinctive features of Ultraman with long hair added to the head. The plaintiff then sued the provider of the generative AI tool, Tab. The Guangzhou Internet Court ultimately ruled in 2024 that the provider failed to fulfill the obli - gations outlined in the Interim Measures for the Administration of Generative Artificial Intelli - gence Services (the “Interim Measures”), issued in 2023 by seven departments. As a result, the
90
CHAMBERS.COM
Powered by FlippingBook