Discover more from Aaron Ross Powell
No, Midjourney Is Not Stealing Your Work
The argument that AI image generators steal from artists would mean that all artists are thieves.
Most of the arguments against the use of AI like ChatGPT and Midjourney to create prose and images, especially for commercial use, don’t make a lot of sense. And this is understandable, because once you clear away the inconsistent claims and unexamined assertions, you’re left with two main reasons for opposition: AI will economically undercut artists and writers, and AI is creepy.
We can’t dismiss either of these out of hand. It’s perfectly reasonable for artists and writers to worry that generative AI means it’ll be harder for them to earn a living. And feelings of unease about computer programs that drift ever closer to producing human-style output inevitably arouse feelings of, if nothing else, the uncanny.
But that doesn’t mean we should let bad arguments off the hook. That opponents of the commercial use of ChatGPT and, especially, Midjourney, Stable Diffusion, etc., avoid stating their real concerns is a problem. It’s one thing to say, “I don’t think you should use Midjourney to make art for your commercial project because I think you should pay me instead.” It’s quite another to concoct transparently bad arguments as cover for that simple economic interest.
One of the least convincing is that generative AI amounts to IP theft–or, in the words of the Author’s Guild, “plagiarism.”
Text and image generators work by first getting trained on a corpus of content, which in practice means going through the internet and looking at stuff. This could be the whole of Wikipedia or the depths of DeviantArt. The information then gets processed into a neural network, with assigned tags (for subject, style, and so on), and then becomes the basis of what the software generates when you ask it for an image of Donald Trump getting arrested or the answers to the bar exam.
In other words, AI image generators have looked at a lot of images, have a sense of the appearance of different things (objects, styles, tones, and so on), and then use that knowledge to make new images based on user prompts. Occasionally, this will result in perfect or near perfect copies of other art, but you have to put some effort into getting them to do that, and doing that isn’t what they’re predominantly used for. Instead, they’re used to generate new art in the style of old, or mash together styles, or create scenes that don’t exist elsewhere.
In other words, they do exactly what human artists do. Every artist has influences, and no artist is free of them. Every artist gets a sense of what different objects look like by either seeing them in the real world or seeing them in the art of others. Every artist makes images that are “original” in the sense that they aren’t perfect copies (unless it’s tracing, or a forgery), but aren’t “original” in the sense that they aren’t in large part based on what’s come before. The images and books artists and authors learn and draw inspiration from don’t get processed into a computer’s large language model, but they do become information in the artist’s brain. (Or they get saved to cloud storage for future browsing, or copied into personal mediums for practice.)
Thus the only real difference between the way human artists go about using what’s come before and the way computers do it is that, in the case of computers, it’s a computer doing it. And maybe that matters, somehow, but it’s not clear how it could be relevant to distinguishing why human artists are “inspired” by others, but computers doing the same thing are “stealing.”
One common approach to making the case that they in fact are distinct is to claim that Midjourney isn’t coming up with anything new, while human artists inject a non-synthesizable creativity into their creations. Thus the computer, because it isn’t creating anything new, is merely copying, if even in a loose sense. And copying, in the world of intellectual property, is stealing.
But this argument doesn’t work, either. Occasionally an artist comes along who creates something we declare genuinely new, but even that has influences, even that draws upon what’s come before–and, besides, this level of novelty is exceedingly rare. For most artists, what they’re doing, let’s be honest, looks an awful lot like what plenty of other artists are doing, as well. How much actual, novel creativity is there in the countless Tolkien pastiches filling shelves in the bookstore fantasy section? How much is that new police procedural a freestanding work of uniquely human creativity, and how much is it just the same thing we’ve seen countless times before? When an artist gets hired to do yet another anime inspired cover for a tabletop roleplaying game, or when he draws another picture in the house style, it’s difficult to say it’s meaningfully distinct from Midjourney doing the same. Except, again, that one is a human and one is a computer.
And maybe there actually is an important difference between the two. Maybe humans have a metaphysical essence that imbues their output with soul in a way a computer can never match. But if so, that means the real concern is about that essential nature, and not with a papered on claim to IP theft.