context: The rapid rise of AIGC (AI generated content) tools has raised concerns over ownership. Can AI developers or users claim copyright for generated content? Can AI developers freely use data from the internet to train models? In a landmark lawsuit in November 2023, the Beijing Internet Court ruled that AI-generated images may be copyrightable. Since future lawsuits do not have to follow this judgement, the debate continues.
A seminar on property rights governance for AIGC (AI-generated content) was held on 26 January 2024 in Beijing chaired by Zhang Linghan 张凌寒 China University of Political Science and Law Professor Institute of Data Law professor.
On the legality of using training data from the internet without consent of the original creators
- Zhang Linghan stressed
- the PRC is still playing catch-up with leading companies in the US, because domestic developers struggle to access high-quality training data for large language models; only around 4 percent of typical corpuses are in Chinese
- further restrictions on training data would only aggravate this
- policymakers need to think about ways original content creators can be reimbursed
- the PRC is still playing catch-up with leading companies in the US, because domestic developers struggle to access high-quality training data for large language models; only around 4 percent of typical corpuses are in Chinese
- Zhang Xinbao 张新宝 Renmin University Law School professor argued that
- copyright was developed for the ‘knowledge economy’, but is outdated as the PRC transitions to the ‘digital economy’
- open use should be the norm
- compensation should depend on the scenario, user and model type, rather than rigid copyright protection
- Liu Xiaochun 刘晓春 Chinese Academy of Social Sciences University Law School professor explained that AIGC usually does not generate direct copies of training data, thus not leading to direct copyright infringement; nevertheless, compensation mechanisms for original content creators need to be explored
On the copyrightability of AIGC
- Cui Guobin 崔国斌 Tsinghua University Center for Intellectual Property director argued for a case-by-case approach
- if users only write one simple prompt, copyright cannot be claimed but if they write detailed prompts with direct correspondence to the output, copyright can be claimed
- Cheng Xiao 程啸 Tsinghua University Law School professor stressed the need for unity of rights and liabilities
- if users or AI developers can claim copyright, they also need to bear liability for potential copyright infringement if the work resembles work from the training data
- Zhang Ping 张平 Beijing University Law School professor stressed that the main question is not whether copyright can be claimed for AIGC, but by whom: AI developers or users. This depends on the level of automation and contractual agreements
- Liu Xiaochun stressed that while the prompts written by users are copyrightable, the final output may not
- further reflection is needed on the degree of correspondence between user input and AI output, a new variable for copyright lawyers