Copyright Thickness, Thinness, and a Mannion Test for Images Produced by Generative Artificial Intelligence Applications
Molly Stech
DOI: https://doi.org/10.2139/ssrn.4704079
2024-01-01
SSRN Electronic Journal
Abstract:Two main questions challenge the nexus of copyright and artificial intelligence (AI): 1) Whether AI can be an author; and 2) Whether AI ingestion of copyrighted material amounts to infringement. As more and more thoughtful commentary circulates about these issues, the more interesting and unsettled questions are, at first blush, regarding whether copyrighted content can or should be ingestible without rightholder consent, compensation, or credit by large language models or other AI applications, and the extent to which such scenarios amount to infringement. In considering the question whether an AI application can or should be considered an author for purposes of copyright law, there seems to be growing international consensus – correct, in my view – that AI can be used as a tool, but that a human author must have ideated a copyrighted work and that the resultant creative work is the outcome of his or her own intellect. Despite some international convergence on this issue, however, it is worth reviewing the backdrop of the “author question,” and uncovering some of the more vexing practicalities regarding how creatively autonomous a person must be to receive a copyright. Are fifty binary choices enough to confer authorship? Are 624 prompts enough? Similar to other areas of copyright, such as the idea-expression dichotomy, or the notorious unpredictability of the U.S. fair use doctrine, there are almost no bright lines to be drawn. I submit that this grayness is inherent to copyright law and that litigation will define the contours of these questions, just as they have done for other vital copyright questions. This is appropriate because the technology will continue to evolve, as will peoples’ command of the technology and the extent to which the tools themselves sharpen and become more able to reflect peoples’ creative intent. An “author question” that is not raised as frequently is the one about attribution. I submit that the moral right of attribution for human authors is more important than it has ever been, despite attribution’s uneven international implementation, and it deserves pointed attention in debates on these issues. As for other areas of nexus between generative AI and copyright, existing law is largely adequate to properly deny copyright protection to the outputs of generative AI. Lastly, I suggest that the existing Mannion test for identifying originality in photographs is applicable to images and other outputs produced by generative AI, although it can be put in a softer focus to better orient the test around the real question at play. The trouble with the Mannion test for these purposes, if any, is not in its requirements but in highlighting the proper perspective on how generative AI usurps the essential step of creativity in humans that copyright is meant to protect.