On March 16, 2023, the U.S. Copyright Office published policy guidance on the registration of works containing materials generated by artificial intelligence (AI). The Copyright Office has been receiving copyright applications for AI technology, as author and co-author within the name of the work and in the acknowledgments. The guidance clarifies how copyright applications should address AI content (in hopes of more uniformity).
The office repeats, reiterates, restates, and all but yells from the mountaintops: This art better be human-sourced.
Human Authorship and U.S. Copyright
Copyright protection in the U.S. is only available for works that are the product of human creativity. More specifically, the works must have been authored by a human.
This “human authorship” rule sounds like science fiction, but it pre-exists AI. It has come up in the context of works created by machines, serial processes, and other atypical methods of creation.
Much of the guidance is dedicated to the Copyright Office setting out “The Human Authorship Requirement” in the context of AI generated content. This seems odd until you notice that the guidance is directed to “Works Containing Material Generated by Artificial Intelligence.” By the end, it is abundantly clear that the Copyright Office will:
- Not register a work generated by AI
- Require all AI generated content in a work to be excluded from the registration
But what does “work generated by AI” mean?
Defining Work Generated By AI
For the Copyright Office, it means the expressive elements of the work were determined by AI instead of a human author.
Consider prompt based AI programs like Midjourney (images) and ChatGPT (text content). Users type in a prompt of what they want the AI to generate, and then the AI generates that content with varying degrees of success.
Who is determining the expressive element of that content? User or AI?
For the most part, the AI. No matter how particular or technical a user’s written prompt may be, it does not change that the expressive element was performed by the AI. The user is like a patron giving instructions to a commissioned artist to generate work. There is little debate that the expressive element of the work comes from the artist, not the patron.
And so the guidance directs all applicants who use AI generated content in their work to:
- Use the Standard Application
- Identify themselves as authors
- Use the “Author Created” field to describe their human authorship contribution
- Where AI generated content constitutes more than de minimis work – applicants must identify that content as “Material Excluded”
This doesn’t just apply to copyright applications going forward. The guidance states that:
- Pending applications should be updated with information regarding AI generated content
- Existing registrations should be updated through a supplementary registration that pinpoints and excludes AI generated content.
The Future of IP in AI
AI is rapidly advancing, and there are two scenarios on the horizon likely to alter or at least challenge the default of excluding AI generated content from copyright registrations.
The first scenario is real time control and interaction with the AI output. Most publicly accessible AI programs work with a written prompt and a processing time in which the AI generates the requested material. After generating, the user has limited ability to modify the content.
It is only a matter of time until users can direct the AI to make immediate changes to specific parts of the generated content and/or modify that content in real time. At that point, the AI program begins to resemble real time editing software like Photoshop, Ableton, or Microsoft Word. As more available control is exerted by the user, the assumption shifts away from AI responsibility for the expressive elements of the work.
The second scenario is a lot trickier. AI is trained by being fed content to be used for reference, i.e. photographs, paintings, songs, poems, scripts. With sufficient reference points, the AI is able to not only generate pieces, but to replicate different styles. There are even artists feeding their own works to AI in order to generate content in their own style.
If a painter trains an AI on her style so that it matches her style exactly, is that work authored by the AI or the painter? Where do the expressive elements originate? If the AI and painter have the exact same approach to expressive elements, could the painter claim to have authored the generated content? I don’t know, but I can’t wait to find out.
Note: This article was authored for Crystal Broughan’s blog by a human guest author, Logan K. McEwen, and no AI was used in the generation of same.