이미지 생성형 AI의 학습데이터 이용과 원저작권자의 이해관계조정 - 저작물의 공정이용과 외부효과를 중심으로 -

Data is a key resource for the innovations in artificial intelligence(AI). AI companies argue that they should be allowed to train their AI systems on copyrighted works under the text and data mining exception or applying fair use doctrine. Recently, the emergence of generative AI systems raises questions about various legal issues such as whether AI generated outputs should be protected by the Copyright Act, who owns the copyright, and whether it would be possible to grant legal personality to AI. In particular, the image-generative AI systems like Midjourney or Stable Diffusion use large quantities of existing works for training without any legal permissions and generate outputs similar to the existing works. Therefore, copyright owners are raising issues about their works being used for AI system training without any compensation. The innovations of AI technology or the growth of tech companies affect the public interest, making it seemingly appropriate to apply fair use doctrine for training AI with copyrighted works. However, copyrights are granted under the Constitution to authors of creative works, and the author’s economic right is a type of property rights. Therefore, the balance of legal interests must be evaluated in cases where the copyright limitations or exceptions may undermine the author’s legitimate interest. This paper will explore the externalities and reasons for copyright limitations or exceptions to identify any missing factors that could balance the public interest and the private or personal interests when applying fair use. We hope that this paper will help to find a way to align the public interest in the innovations of AI technology with the welfare of individuals.