A February 21, 2023, decision by the U.S. Copyright Office has given all creators and companies reason to think twice before jumping on the generative AI bandwagon.
Over the past six months, the world has been fascinated by the explosion of “generative” artificial intelligence programs — in other words, AI programs that generate content in response to human requests. The most popular of these applications include ChatGPT — the surprisingly conversational chatbot that Microsoft recently incorporated into its Bing search engine — and image-creating software such as DALL-E 2 and Midjourney, which turn a few words of input into visual artwork of remarkable quality.
But it was an artist’s use of the latter program — Midjourney — that prompted a recent ruling from the U.S. Copyright Office that calls into question the utility of AI-generated images for anyone concerned about protecting their intellectual property. Visual artist Kristina Kashtanova sought, and initially received, a copyright registration for her graphic novel Zarya of the Dawn. But when the office learned that the imagery in the book was generated by Midjourney, it decided to revoke the registration. Well-established, and recently re-affirmed, law establishes that copyright only exists to protect the original, creative expression of human authors — not of animals or machines.
Kashtanova appealed the revocation, arguing that Midjourney had merely been an “assistive tool” in the realization of her own creative vision. And the office did ultimately conclude that the end product contained some of her original expression. For example, she wrote the text balloons spoken by the characters, and she decided how to arrange the images in the book. Copyright law recognizes these as sufficiently creative to merit protection, so the office decided to maintain Kashtanova’s registration as to these elements.
But the imagery itself was not copyrightable human expression, in the office’s view, and will be removed from the scope of the registration. Kashtanova argued that, because she chose the “prompt” words she supplied to Midjourney, she was the legal author of the resulting image, and the AI was merely a tool she used to create it.
After a detailed review and explanation of how Midjourney, and other generative AI programs work, the office disagreed. “Generation involves Midjourney starting with a field of visual noise, like television static, used as a starting point to generate the initial image grids and then using an algorithm to refine that static into human-recognizable images” by drawing from the program’s training data, it wrote. “The initial prompt by a user generates four different images based on Midjourney’s training data. While additional prompts applied to one of these initial images can influence the subsequent images, the process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time.” The fact that Kashtanova used Photoshop to touch up a few of the images did not change this result, because the alterations were so minor as to not add any new, original expression. As a result, the images were “works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author,” and are therefore ineligible for copyright protection.
The takeaway from this decision for any creator or business that intends to own copyright in the text or imagery that it uses should be clear. Relying on AI-generated content could leave your copyright portfolio looking like a donut: with some valuable content around the edges but a giant hole in the center. Worse yet, if your content is not protected by copyright, that means it is public domain, and free for anyone to copy and use. That may have little impact if the work in question is a few lines of ad copy or a stock image to accompany an article. It becomes a much greater problem if the AI-generated content becomes the centerpiece of a commercial product, like Casanova’s book. In that case, you could end up investing great sums of capital, time and goodwill into work product you do not own and cannot prevent others from copying.
And that’s not the only potential copyright problem with generative AI. Recall that AI programs apply algorithms to their training data in order to create output. Yet much of that data is itself copyrighted content from other creators that has been scraped off the internet. That is why there are several lawsuits currently pending alleging that Midjourney and other AI programs are committing copyright infringement on a massive scale. In the worst-case scenario, those who use these programs could end up with content they don’t own and that infringes someone else’s rights.
Obviously, these are emerging and rapidly developing areas of law. Other courts and agencies could reach different results under different circumstances, and the technology will always continue to adapt. In the meantime, however, creators and businesses should resist jumping on bandwagons, and should instead think deliberatively about the creative tools they use.
