Images from neural networks cannot be protected by copyright, US rules Automatic translate

This decision would effectively remove protection for, for example, Jason Allen’s work "Théâtre D’opéra Spatial" ("Spatial Theatre"), created using a neural network.

According to new guidance from the U.S. Copyright Office, images generated by "artificial intelligence" based on a text query cannot be copyrighted, even if the artist uses long, targeted prompts or creates multiple iterations of the work before being satisfied with the final result.

The decision would effectively strip protections from works like Jason Allen’s "Space Opera," which shocked the art world by winning first place in the digital category at the Colorado State Fair.

The Copyright Office made its decision in a new report on the copyrightability of neural creations created in 2023, asking the public for nearly 10,000 comments. Artists, performers, and composers were invited to weigh in. The federal government, noting that the Copyright Office has been dealing with computer-based authorship issues since 1965, concluded that copyrightability could be determined in the context of existing laws that provide limited protection for such works.

“Neither the use of AI as an assistive tool nor the inclusion of AI-generated content in a larger copyrighted work affects the copyright protection of the work as a whole,” the report says. “However, the capabilities of emerging generative technologies raise difficult questions about the nature and scope of human authorship.”

Federal officials noted that human-AI collaboration in the creation of art can take many forms, and that much of the Copyright Office’s rules will be determined on a case-by-case basis.

“If AI simply assists the author in the creative process, its use does not change the copyright of the result. On the other hand, if the content is entirely generated by AI, it cannot be protected by copyright,” they say. “In between these boundaries, there may be various forms and combinations of human input into the creation of AI output.”

But they did find that “prompts alone do not provide sufficient human control to make neural network users the authors of the output.” Unlike human-to-human artistic commissions, such as when an artist directs studio assistants to create a painting, they found that a human using AI does not have the same oversight.

However, this may not always be the case, and the issue may need to be revisited in the future. The Copyright Office has acknowledged that in the future, “theoretically,” generative systems could allow artists much more control over the process, to the point that using the technology would be more “mechanical,” like an artist’s tool.

“Evidence from modern generative neural networks suggests that this is currently not the case,” the report says. “Prompts (i.e., text queries or suggestions) do not appear to sufficiently identify meaningful elements or control how the system transforms them into the final output.”

To prove its point, the Copyright Office described how Google’s Gemini system created an image of a cat smoking a pipe and reading a newspaper that ignored some of the prompt’s instructions while adding elements such as an "alien human hand."

But the interesting aspect of the report is that federal authorities determined that the text prompts themselves, if creative enough, could be copyrighted, even though most commentators agreed that entering simple prompts was not enough to claim ownership of AI-generated results.