While I was out in Utah at Dragonsteel Nexus, I had a chance to watch Brandon Sanderson deliver his keynote address on the opening night. His topic? Can AI Be Art? Here’s the video of the event, and I think I set it up to start at the keynote. (If it doesn’t work, just go to 32:58.) It’s a great discussion of the topic, and as a longtime writer and a frequent user of AI these days, I have many thoughts, to say the least.
Brandon’s point seems to boil down to the argument that people using AI skip much of the creative process, and that it’s that act of creation that makes art unique. And I really don’t dispute that at all. However, I believe the act of using AI often gets reduced to a very simple equation. “Someone puts in a prompt, hits enter, and then takes whatever AI gave them and presents it as the finished product.” And while that’s certainly one way to approach AI, I don’t think it’s a very good way to do it. (Though I’d agree that via that approach, nothing really human is created, in much the same way that if I sit down at a player piano and hit Play, I am far from a pianist.)
I use AI in many different ways, both professionally and creatively. I use it to brainstorm. I use it to give me feedback on my thoughts, I use it to draft. I use it to edit. And as I’ve used it, sometimes I’ve leaned more heavily on it, and sometimes I’ve used it only very lightly (or not at all). But at each step along the way, it’s always me directing it. And Brandon has made this argument before: that someone who uses AI to create a drawing isn’t really an artist; they’re an art director at best. But to me, this brings up an even more fundamental question: what is creation?
If I have an idea in my head, and I work to get that idea out in a way that you can experience it as well, isn’t that creating something? The thing didn’t exist beforehand. Now it does. The manner in which the thing came to be seems to me to be largely irrelevant, if what I was really doing was taking that concept and turning it into something someone else can see and experience.
Don’t get me wrong: if I have no idea of my own, and I turn to AI to just give me something, then the act of creativity is pretty low. But let’s look at the art of photography. A human takes a machine and uses it to capture an image of something that’s already existing. They can do it poorly or effectively, but I think most people agree there’s an art to it, especially at the higher levels. It’s different with a photocopier, of course. You still take a machine and use it to capture an image of something that’s already existing, but you do it in a way that is (typically) identical to what was already there. Unless you’re making stylistic decisions, most people would likely agree that’s not art.
There’s a spectrum there. It’s not an either/or situation with cameras. You can have something that’s not art, and something that very much is. The difference is how much of you you put into it. (Note: I think Brandon and I agree on this. I’m not saying he’s wrong when he says “We are the art.” I’m saying we can still be the art through the medium of AI, if it’s approached correctly, and since the field is so new, we’re still learning how to approach it.)
Concrete example: my brother-in-law is a copyright lawyer. He represented the person who tried to copyright a written work (graphic novel, I believe) that had been generated by AI. The one that ended up with the ruling that AI works can’t be granted copyright. That person gave up, but my brother-in-law still thought there was something more to be explored there. Knowing I’m an author, he came to me, wondering if we might try an experiment. Could I write a story completely with AI? And if I did, could I be given copyright for it? This was a great intersection between two things I’m passionate about, so I agreed.
But I wanted to do it my way.
We took my book Vodnik and (with the help of my niece, who’s also a part of this) worked on training a version of Claude to write in my style in that book. (Essentially, we fed in scene by scene and gave it prompts that would have resulted in that scene, though I’m oversimplifying it.) I then used that version of Claude to write a short story sequel to Vodnik. I did it scene by scene, giving the AI prompts about what I wanted it to do. What was supposed to happen. What the mood was supposed to be. And I did that until the story was finished.
It wasn’t particularly good. This is already a year or more ago, and AI wasn’t nearly as good. But even then, it still wasn’t something I’d want my name on. So I did what I always do with my writing. I got to editing. The goal was to totally use AI to do it, though. So I read through the whole manuscript and made notes for what I wanted changed. Different actions. Different descriptions. Different pacing. And then I took those notes and worked on the story scene by scene with AI, copy and pasting in the original, and then giving it directions on how to change it. It was a lot of work, but I got it done. And then I did that once more, because I still wasn’t happy with it. Finally, I made a pass where I started giving it specific directions of wording tweaks to make. But the entire output was all me putting in prompts to Claude and getting text out of it, and pasting it into a final document.
You can read it here. Is it my best work? No. And telling you all the process of how it was made will no doubt cause it to be much more heavily scrutinized. That’s okay. It was a side project and a first attempt. if I did it again today, I’d do a better job, I’m sure. AI is better, and I’m much better at using it.
The story is 100% AI generated, but it’s also a work of art by me. I learned a few things in the process. For one thing, I hated working like that. Forcing myself to only use AI was obnoxious, but a fun challenge. I would have much preferred to directly dive into the draft after that very first rough draft and make the changes myself. But it also still felt like a creative effort on my part. We’ve since applied for copyright, and it’s working its way through the system. No clue where it’s going to end up.
But what this amounts to is that I feel like I have some experience to speak from in this realm. I don’t think Brandon’s 100% wrong, but he’s also not 100% right. There will come a time when people just use AI to churn out text, and that text will be “good.” It also won’t be art. But I think people will continue to seek out artists and works written by humans, because art connects us. We want that feeling of experiencing the same thing the artist first came up with in their head. I think AI will be one more tool for the artist to use, poorly or well. I think it’s a thorny mess, yes, but also not one I’m willing to just dismiss.