Using ChatGPTo, I spent close to two hours refining prompts to generate the sunset image shown. I’m proud of the final picture. A bright orange cloud slightly obstructs the large surreal moon. There is just enough wind to put a hint of white caps on the waves. Silhouetted against a bright yellow-orangeish sky, seagulls look for fish they can poach near the surface. To add to its surreal appeal, I placed small islands floating in the sky. I didn’t tell ChatGPTo where to put them and it ended up placing them on the left.

Since I put a lot of effort into fine tuning this AI-generated picture to my taste, should I not be credited with at least a modicum of credit for creativity? Just as a painter uses oil paint and canvas, I was using ChatGPTo as a tool. Should not this be enough for the US Copyright office to grant me a copyright?

The answer is no. And there is good reason.

First, I lied. The sunset picture was generated by a single prompt to ChatGPTo. My prompt was:

“Draw a surreal picture of an ocean sunset.”

There was no iterative revision of the sunset image. Outside of the prompt, I supplied no additional instruction for generation of the image. All US copyrights and patents must be assigned to humans. By law, therefore, the sunset image cannot by copyrighted. From my single prompt, the image was generated in toto by AI.

The Théâtre D’opéra Spatial controversy

A similar copyright question is a problem for Jason Allen who is appealing the US Copyright Office’s decision to deny registration for his award-winning AI-generated work “Théâtre D’opéra Spatial.” (Click here to see Allen’s creation.) Allen says, using Midjourney AI software, he fine tuned the image for 100 hours using over 600 prompts.

There is no doubt that this commitment warrants a copyright. Allen used Midjourney as a tool and not as the source of his creativity. He is probably an honorable man but the precedent set by awarding him a copyright has ramifications. His claim needs evidence. Without evidence, I could present my sunset picture, lie, and get a copyright.

The only solution I can see to this AI copyright problem is to present meta-data to verify that numerous highly detailed prompts were used to create an image. Presenting the intermediate images would be helpful.

Fuzzy boundaries

But how many prompts is sufficient to establish creativity? The boundary is fuzzy. This is a problem for lawmakers because fuzzy boundaries are everywhere. The amount of cocaine possessed (indicating an intent to distribute) is an example of a fuzzy boundary. In Texas, it’s 4 grams, in Ohio 5 grams, and in Alabama 8 grams.

Fuzzy boundaries can be made law by using fuzzy words. For example “copyrights for art generated by AI can be granted if there is ample evidence AI was used more as a tool in the creation of the work rather than as a source of creativity.”

I believe that Jason Allen should be granted copyright for “Théâtre D’opéra Spatial.” However, doing so would set a risky precedent. AI is here to stay, and artists will continue to use it in their creative processes for years to come. Lawmakers must modernize copyright laws to ensure that artists can use AI as a tool and protect ownership of their art.

You may also wish to read: Generative AI is creating a copyright crisis for artists. How does an artist assert copyright when her image was only one of many used to create a new image? How does she make a living if she can’t? AI companies could be required to record images’ origin and compensate copyright holders. But as Robert J. Marks notes, end users might then have to pay.