The development of "generative AI" technologies has significantly stimulated academic research in recent years, especially in the area of legal implications. In particular, copyright issues have been discussed more than other intellectual property rights, largely due to the widespread use of pre-existing works to train these advanced AI models.
The primary copyright issues related to generative AI include whether an AI-generated output can be copyrighted, what rights can be granted if copyrightability is denied, who the rights holder might be in such circumstances, and issues of copyright infringement and liability. This article delves into these three issues across separate chapters.
First, in terms of authorship, Korean copyright law defines a "copyrighted work" as "a creative expression of human thought or feeling. This implies that authorship is not extended to non-human creators.
Second, it was confirmed that this position is consistent with international law and case law. Furthermore, if authorship is denied, the discussion explored whether AI-generated output should remain in the public domain or be protected by rights other than copyright. However, if any rights are not granted to AI-generated outputs, it would be challenging to address the unauthorized use of these outputs in the secondary market. Moreover, the absence of rights could reduce the incentive to create and use such outputs. From this perspective, I argue that AI-generated outputs should be afforded legal protection through intellectual property rights. Accordingly, I have suggested that it is appropriate to grant neighboring rights. This is consistent with the legislative intent behind neighboring rights, which is to protect the act of making works available to the public. Moreover, because neighboring rights are less exclusive than copyright and have a limited scope, they are proposed as the appropriate rights to grant. This approach could facilitate the use of AI-generated works.
The third issue concerns copyright infringement and liability with respect to AI-generated outputs. This article has identified two primary types of copyright infringement in generative AI: 'input-phase infringement,' which occurs during the process of feeding and training an AI model, and 'output-phase infringement,' which occurs during the creation and secondary use of outputs that bear substantial similarity to the original work. A key question is who is liable for infringement at each phase, so this article has also examined this issue.