AI Art, or generated art, is a problem, yes, but not for the reasons people think. It’s a problem because nobody was prepared for how rapidly it would impact our world of creatives. Nobody was ready for how hard it would shake the box.

For better or worse, AI as a production tool is here to stay. Whether it’s writing, art, architecture, film  or music, AI is being used in practically every field of creative human endeavor.  The role it has had in our daily lives so far has been incalculable already, but now the tools are so good that they can generate content that can stand — if you stand back and squint — on its own.

It’s like the admonishment about recognizing the fae: be wary of travellers on the road.  Count the eyes, count the teeth, count the fingers, and whatever you do, don’t invite them in.

Except that they’re already here, already doing work that no human would want.  Would a human do a detailed painting to illustrate a goofy joke for fourteen cents?  Of course not.

Alternatively, would an AI art tool be able to carry the illustration of an entire graphic novel without some serious, iron-fisted control from a human artist that anybody would want to read?  You might get away with that trick a few times, but in the long run, no.  If a human didn’t make it, why should we care what the story has to say?  There’s nothing of the stuff of life in it, and that’s what we find compelling: our common experience, of which the AI cannot and can never share.

The Argument

People claim that it steals artwork from the original artists (it doesn’t, only making generalizations made from the observance of the artwork of humans, just as a human artist would do) or that it takes jobs away from humans (TOR Books is in some hot water over a book cover they commissioned that used stock library art that turned out later to have been AI generated).

If I paint in the style of Van Gogh (warm saturated earthy colors, impasto, impressionistic, with emphasis on the arcs and swirls that flow between in the negative spaces), am I stealing from his work? No reasonable person would claim this. Now, what If I use an AI Art generator like Midjourney to do the same thing? It’s a shortcut, yes, but stealing, or cheating somehow? To me, it just appears to be a really sophisticated tool, and one in its rocky infancy.

It is, however, a new process whose potential as an art tool is understood by very few, and whose operation is understood by even fewer. It is my observation that the alarum being raised is similar to that raised about the rise in popularity of synthesizers as early as the mid-1950’s. Everyone was sure that the synthesizer would put a lot of professional musicians out of work. Of course, that did not happen. It’s true that synthesizers were used in place of an ensemble of real musicians, but in a lot of those situations there just wouldn’t have been money to pay humans.  Instead, music became possible where the alternative would have been silence, canned music taken from something else, or somebody trying to make do with a single guitar or a piano and a set of bongo drums. Synthesizers simply became one more tool in the toolbox.  AI Art is just one more step past CGI, and nobody these days is claiming that isn’t art.

A healthy debate is already in full swing. Already facing some backlash from artists, Artstation is allowing artists to opt out of having the artwork they submit to the Artstation web site used to feed AI art generators, and there is an on-going protest there among artists who think Artstation shouldn’t be allowing people to sell AI-generated images there. The site was awash with anti-AI posts protesting the original policy, with the illustrator Alexander Nanitchkov, creator of the No AI logo, proclaiming AI generated work to be “soulless stealing”. The counter-argument to this is, of course, if a human looks at a body of work and says, “Yeah, I think I can paint in that style”, and then does it, is that stealing? Few would argue that studying and replicating somebody else’s art style is theft, because original artwork isn’t being simply copied. Yet, when a machine does it instead of a human, somehow it’s supposed to be different.

How It Works in Very Basic Terms

AI-generated art doesn’t just copy bits and paste them together. Instead, the artificial intelligence is taught about art by looking at a huge number of images, adding noise to each one until it becomes unrecognizeable for what it was, and then taking notes on exactly what made that image recognizable as being a certain thing. The process is repeated on a very large number of similar subjects so that the AI can tell, in general, what makes that subject look like what it is. This may include photographs, or the work of human artists, but always in great quantity, usually tens of thousands or more.

To generate a new image, the process is reversed. It begins with a noise field, and then everything that doesn’t look like the requested subject is slowly repaired. It’s further skewed by another instruction layer that adds other elements to the scene as described, to create a new and unique image. The resulting image may contain varying percentages of a given individual’s artwork, but it’s never a straight up copy.

Opting Out

If you’re an artist and you want to know whether your work has been used to train AI art synthesis systems, the service HaveIBeenTrained.com may work for you (we haven’t tried it). Artstation has also introduced a metatag you can put on your work, NoAI, which is meant to tell AI art generators not to use your work to train their diffusion models. Presumably, this doesn’t work on stuff that’s already been accessed that way, and makes no promise that any of the services that use Artstation as a resource will honor the tag. There’s a setting in your preferences on ArtStation that will make the addition of the NoAI metatag the default.

Artstation has also changed its terms of service to prohibit the use of materials carrying that tag, but again, there’s no way known to make that retroactive because of the way diffusion modeling works. Once your artwork is used for training, the information is essentially atomized and becomes part of the greater whole, with no way to reverse the process. There is a lot of criticism of Artstation at the moment, and they’re making strides toward resolving those complaints, but to be honest, there likely isn’t more they can do other than what they’re already doing.

The topic is further blurred by the fact that what the AI generators are doing, humans have been doing themselves for centuries. Artists of every stripe have copied the work of others while learning their craft, Van Gogh included. It would be very easy to point to this and assert that the complaints seem to be that a person directing a machine is doing it rather than a person directing a paintbrush.

It Needs What It’ll Never Have On Its Own: A Little Heart

Synthesized art suffers the same problem that synthesized music does: it lacks heart. A performance in either medium created solely by algorithm lacks the human touch, the emotional connection that makes that creative product worth consuming. Without it, it’s just an attractive but ultimately soulless effort. It can save time producing creative content, but without the guidance of an actual artist it will produce only the facade of meaning without ever actually touching it. As a result, AI created art is actually pretty easy to spot when you see it.

I predict that there will be a great deal of arguing back and forth about AI art, but in the end, people will pause long enough to realize that AI art can’t reasonably compare to the creativity of a skilled human artist, and we’ll all get on with our lives. There is precedent; this is pretty much how the arguments against synthesizers went.  After a while people realized that the synthesizer was just another tool, and in the wrong hands it could produce flavorless pap just like any other tool in any other medium—or in the other direction, allow the art form to be taken to new places previously inaccessible.

The argument that AI generators are stealing the style of artists might be true in some sense, but even that means different things to different people. It’s highly subjective. That not withstanding, one cannot copyright an artistic style, any more than one can copyright a color palette, or a chord sequence, or the alphabet. You cannot sue the manufacturer of a typewriter because it uses the same letters used in your most recent book.

You can certainly stop somebody from selling a picture of Mickey Mouse, but that happens post creation, not at the tool level where the AI is. Beyond that, we do not even have fully formed concepts for what the AI art generators are doing in terms of content rights, let alone laws that apply.

AI-generated art is like a chainsaw: it can do a lot of damage very fast, and it seems dangerous to be around. Without its human guiding what it does, though, it’s just another tool to be tamed. And, after a while, we’ll get used to the idea that there are such things as chainsaws in the world that do useful things.

-30-

Update

The U.S. Copyright Office had recently revoked the copyright application of Kristina Kashtanova’s registration for his graphic novel, produced using the AI art generator Midjourney. It turns out that the revokation was simply a system glitch in the relatively new Copyright Public Records System.

The USCO had previously registered the work in September 2022. However, a month later, and following significant press attention, the Office issued a notice that appeared to indicate that the registration would be canceled. The copyright is still in progress; Kashtanova has responded to their inquiry by providing evidence that the Midjourney tool could not have produced the graphic novel on its own without significant input and control from a human author (herself).