Like most writers (I think) I have way more ideas than I can handle, meaning that promising concepts often languish for years while I work on other things. You always think there will be time to go back and develop the good ones, but that might not be true. The sudden explosion in art generated by artificial intelligence makes me think a future I dreamed up years ago is here now, for better or worse.
The idea I had was that a group of computer scientists had built a super computer that, when fed with enough examples of a particular kind of art, could then produce the “perfect” example of whatever it had been fed. The computer, at the time the story would have started, had already written a best-selling mystery novel and created some impressive visual art, too. For its next project it would take on music, leading the cranky main character to show the world that music was a human endeavor that machines could never match. I actually started this story a couple times, but it never got very far.
If I were to try and finish it now it would become more historical fiction that sci-fi, as AI art is really getting its moment in the sun. I’d seen some people on Twitter playing around with various programs where you give it a text prompt and it produces a picture. Some of them were kind of neat, others were horrifically creepy in an unintentional way. It all seemed like a lark until somebody took one to a state fair.
A guy named Jason Allen submitted an artwork called “Théâtre D’opéra Spatial,” which, from the presentation of it in this New York Times article, looks like something from a sci-fi or fantasy cover from the 1970s or 1980s. That’s no criticism – it’s pretty cool and I could see it prompting some interesting stories.
Allen submitted the work in the Colorado State Fair’s art competition, specifically in a category for “artistic practice that uses digital technology as part of the creative or presentation process.” The “digital technology” in this case is an AI program/website called Midjourney. It and similar setups work this way:
Apps like DALL-E 2 and Midjourney are built by scraping millions of images from the open web, then teaching algorithms to recognize patterns and relationships in those images and generate new ones in the same style. That means that artists who upload their works to the internet may be unwittingly helping to train their algorithmic competitors.
In other words, a user like Allen feeds text inputs into Midjourney, which then comes up with an image attempting to match the prompt. The process itself sounds like it could be excruciating, tweaking terms one word at a time to see what impact it might have on the final process (as someone who does that with legal search terms every day, I sympathize).
Pretty nifty, huh?
All of this would have had the air of a neat experiment or side story, had Allen and his work not won the category (and the $300 that went with it). For what it’s worth, I think Allen is clearly right that he didn’t break any rules. They apparently didn’t define “digital technology” to exclude AI or to only include tools wielded by humans on art they created, so good on him for exploiting the loophole.
But that isn’t the real question. The issue that has people upset about Allen’s win is whether his AI-generated painting is really “art” at all and what it means for human artists. The objections seem to break down along two lines.
The first is, well, that machines just can’t make art or, rather, “art” made by machines lacks something that human-made art does. As someone who at least tries to make art, of the written and aural variety, I’d like to think there’s something to this. On the other hand, as someone who makes electronic music specifically, I’m well aware of the history of musicians (and others) panicking at the onset of some new technological advance. British session musicians in the 1960s were initially super pissed at the Mellotron, one of my favorite instruments, because it was going to put them out of work. It didn’t, of course, because Mellotron strings or flutes or voices never sound quite real, which is part of its charm. Sequencers and drum machines, too, got similar hate for not being “real,” which ignores the fact that what makes them interesting is that they don’t sound “real” in the first place.
But in those examples there’s still a human being pushing the buttons, doing the work. But isn’t that what Allen did? He apparently tried numerous strings of search terms before getting images he liked, tweaking them as he went along. Is that so different than finding a synth patch you like, then using the tools of the synth to shape and sculpt it into something unique and personal? It’s not as if Allen sat down, had a conversation with the AI about what he was looking for, and the AI went away and created. All is doing was responding to prompts Allen gave it which is, for practical purposes, all any artistic medium does.
The second objection feels more immediately justified, but doesn’t quite hold up under close scrutiny. As I said earlier, these AI image generators work by “scraping” millions of images already on the internet and using them as grist for the images the AI generates. This, to some, sounds a little like machines stealing the work of real human artists. Admittedly, I’d be pissed off if I found out some AI writing thing had scraped some of my stories as fuel for its work. But is that gut reaction reasonable?
After all, what are artists if not meat machines that absorb as influences the works of those who came before them? There’s even a well-known saying to the effect that “good artists copy, great artists steal,” although who said it is a mystery (I heard it from Zappa first, I think). Granted, a computer algorithm can absorb a lot more data than a human brain can in a lifetime, but does that make a difference? We’re hashing that out in the legal realm with regards to the Fourth Amendment and whether computers doing what police would never have the time to do themselves raises questions about the constitutionality of searches. Is the difference between Allen feeding the AI prompts and sitting down with a human artist and saying “here’s the kind of stuff I like, can you do one in that style?” one of kind or degree?
Which is where, I think, humans have it all over machines. As slick as any AI is, all it can do is what people have programmed it to do. A machine has yet to wake up one day and decide “I’m going to paint a flower.” I’m not sure they ever will, although I’m not sure I’d bet against it. There’s something to be said for humanity for having the creative urge in the first place.
Ultimately, this issue is one more in the long line of humans worrying about becoming obsolete. Machines and computers do more and more of our jobs. They’re getting to be a bigger part of law enforcement. And now they’re coming for the arts. It was probably inevitable. It’s worth shifting, then, to wondering not whether AI can make art, but whether it can appreciate it. Show a group of humans the same painting or movie or play for them the same song and you’re likely to have numerous reactions to it. Maybe it’s the reaction, not the creation, that is indelibly human?