Can Works Created Using AI Be Protected? A perspective from a machine learning researcher

By Serge Belongie[*]Professor of Computer Science, University of Copenhagen.

From the perspective of an academic in the area of Machine Learning, the simple answer to the question whether works created using AI can be protected is “no”. More specifically, my claim is that the work in isolation cannot be protected, but if one opens the scope to include a record of the creative process, then the answer moves closer to yes.

The reason for my “no” response for the work in isolation comes down to the cat-and-mouse nature of steganographic techniques. State-of-the-art digital watermark protections developed in 2023 can be easily bypassed with algorithms developed in 2024[1]See for example https://arxiv.org/abs/2310.00076., and all indications are that this pattern—one that stretches back to antiquity—will continue for the foreseeable future.

If we look behind the scenes of the artist’s process, however, we may encounter evidence amenable to protection. In the context of present-day Generative AI systems for domains such as creative writing, image synthesis, or music composition, such evidence may comprise a record of the artist’s prompts and, if applicable, source material from which derivative works were obtained. This observation does not point directly to a solution—that, of course, requires one to follow a formal legal review process—but it suggests that normalizing the practice of “showing one’s work” will be instrumental to the prospect of AI artists protecting their work.