A couple of issues ago I wrote an editorial about artificial intelligence, but at the time I said “but not what you’d expect” (or words to that effect). Well, now that we’re doing an issue whose subtitle is “Not What Things Seem,” I’ll be contrarian and do the expected.

Actually, this issue was brought up a couple of times over the last few months, once obliquely (someone asked online whether a piece of art was AI-based because it scored highly on an AI-based AI detector app) and once directly (before signing a contract with us). And I think now’s the time to address it as best I can. And I say “as best I can” because it’s not as easy as just saying, “Nothing any sort of AI has ever touched can be accepted.”

One big problem is the definition of AI. Every company wants to jump on the bandwagon, so anything that uses some kind of algorithm that could in any way be using some sort of “training” to get better at its job is going to be called AI. So this means anything from ChatGPT and its ilk to a grammar checker like Grammarly that makes phrasing suggestions, or anything from DALL-E to Photoshop tools that help remove backgrounds from photos.

So in practice, there’s a continuum of AI use. At one extreme, you have a writer who gets their entire story out of ChatGPT and submits that without even looking at it, or an artist who uses DALL-E to generate an image and submits in similar fashion. These we would not accept; in fact, we couldn’t accept them. The US copyright office has ruled that such things aren’t copyrightable, and as such the “author” and “artist” in this scenario have no rights in the work and cannot sell Penumbric any rights. Anybody signing a contract with us in this case would immediately be in breach of contract. (How one figures this out is another kettle of artificial fish; AI-detector programs are super fallible at present.)

But at the other end of the continuum, say there’s a writer who creates a story and then uses an AI-powered grammar checker to correct grammar and get a few phrasing suggestions for their work. Or, likewise, a digital artist who uses an AI-powered processing program to remove the background from a piece they have created, and then inserts another background (which they’ve also created themselves). Maybe they also use “AI” to recolor portions of their own work. These would, to my mind, be no different to someone using tools (grammar checkers and design programs) to help them change or augment their own work; these would be obviously copyrightable and would be acceptable to us.

The difficulty is somewhere in that vast swath in the middle of these extremes—and this is a difficulty for the copyright office as well. There’s some undefined amount of author/artist input that tips the balance between author-created and AI-created, between copyrightability and not. And even programs that attempt to make this judgment are fallible (and AI-powered themselves).

Our contracts currently don't have anything specific about AI usage, partly because we're still using roughly the same contract form as we did in 2020, partly because I don't know how to put all that "grey zone" stuff into legalese, and partly because there's technically no need (as the purely AI-generated stuff can't be copyrighted and rights cannot be bought). However, going forward we’re going to put in a clause about copyrightability, with generative AI explicitly called out. And we’ll continue looking at this issue going forward.

Jeff Georgeson
Managing Editor
Penumbric