Can Smart Software Get Copyright? Wrong?

September 15, 2023

It is official: copyrights are for humans, not machines. JD Supra brings us up to date on AI and official copyright guidelines in, “Using AI to Create a Work – Copyright Protection and Infringement.” The basic principle goes both ways. Creators cannot copyright AI-generated material unless they have manipulated it enough to render it a creative work. On the other hand, it is a violation to publish AI-generated content that resembles a copyright-protected work. As for feeding algorithms a diet of human-made media, that is not officially against the rules. Yet. We learn:

“To obtain copyright protection for a work containing AI-generated material, the work must have sufficient human input, such as sufficient modification of the AI output or the human selection or arrangement of the AI content. However, copyright protection would be limited to those ‘human-made’ elements. Past, pending, and future copyright applications need to identify explicitly the human element and disclaim the AI-created content if it is more than minor. For existing registrations, a supplementary registration may be necessary. Works created using AI are subject to the same copyright infringement analysis applicable to any work. The issue with using AI to create works is that the sources of the original works may not be identified, so an infringement analysis cannot be conducted until the cease-and-desist letter is received. No court has yet adopted the theory that merely using an AI database means the resulting work is automatically an infringing derivative work if it is not substantially similar to the protectable elements in the copyrighted work.”

The article cites the Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190 (March 16, 2023). It notes those guidelines were informed by a decision handed down in February, Zarya v Dawn, which involved a comic book with AI-generated content. the Copyright Office sliced and diced elements, specifying:

“… The selection and arrangement of the images and the text were the result of human authorship and thus copyrightable, but the AI-generated images resulting from human prompts were not. The prompts ‘influenced,’ but did not ‘dictate,’ the resulting image, so the applicant was not the ‘mastermind’ and therefore not the author of the images. Further, the applicant’s edits to the images were too minor to be deemed copyrightable.”

Ah, the fine art of splitting hairs. As for training databases packed with protected content, the article points to pending lawsuits by artists against Stability AI, MidJourney, and Deviant Art. We are told those cases may be dismissed on technical grounds, but are advised to watch for similar cases in the future. Stay tuned.

Cynthia Murrell, September 15, 2023


Comments are closed.

  • Archives

  • Recent Posts

  • Meta