Author

Li Mingxuan, Lecturer, School of Interdisciplinary Studies, Renmin University of China; Research Fellow, Big Data and Responsible Artificial Intelligence for National Governance, Renmin University of China

Abstract

Due to the memorization phenomenon in generative artificial intelligence( AI), such systems may produce content identical or similar to existing copyrighted works, thereby posing risks of copyright infringement. In practice, generative AI service providers are the primary targets of infringement claims brought by copyright holders. Given the diversity of generative AI service business models, it is necessary to categorize service providers and determine their liability based on the extent of their control over the generated content. Specific measures to cease infringement may include model destruction, retraining, unlearning, reinforcement learning, input filtering, and output filtering. The scope of liability should be limited to the reasonable measures that a service provider can take. Determination of liability for damages should take into account whether the service provider was at fault. In assessing fault, both the duty of care prescribed by existing laws, regulations and industry standards, as well as other relevant factors, should be comprehensively considered.

Keywords: generative artificial intelligence ; service provider；copyright infringement; ceasation of infringement ; compensation for damages