A sunny, chilly Tuesday morning in Portland, Oregon on Tuesday, 31 January, 2023.
I like science fiction and I like things that people imagine might be in space. That means things like the art of Chris Foss, Gerard O’Neill’s cylinders, the through-line to Mitchell Stuart’s concept work on the Elysium station and so on. Which is why I end up following concept artists, both human and non-human (for the latter, see This Terran Trade Authority Does Not Exist).
A thing I’ve been paying attention to is how people choose to incorporate (or not) AI/ML tools into their work, so it caught my attention when I saw concept artist Paul Chadeisson1 experimenting with AI tools.
(There’s a good chance, if you’re reading this, that you’ve seen some of the result of Chadeisson’s work. Recent examples include Amazon Prime Video’s adaptation of William Gibson’s The Peripheral, and Apple TV+’s adaptation of Isaac Asimov’s Foundation.)
Back last year, Chadeisson posted some explorations with AI tools, here’s some links:
There is part of me that reads the notes to each of these posts and can’t help placing them in the context of videogame story: notes from an artist who in the trope experiments with a forbidden method and, well, bad things probably happen.
From what I can tell from the comments (for what it’s worth), Chadeisson’s not using the free version of midjourney, not that it matters (“but it don’t make difference with the result), with some of the explorations he’s posted he’s not training them on his own work (“just spamming the new version button”). He’s also well aware of the issues around resulting copyright, but in a reply points out that he’s focusing more on generating new ideas, “instead of adding some image or artists name in the prompts”.
Something I really like about Chadeisson’s approach is one of his replies to someone pointing out their experience that generated results are “too far off from what I’ve asked for”. Chadeisson says in response that “it’s better to be open for surprise than wanting an exact specific thing.”
This, this is something that’s struck me in an analogy about AI-generated art (I suppose it can also be transposed to AI-generated writing, too). The way Chadeisson talks about this is in the vein of getting inspiration quickly, and not wanting an exact specific thing. It’s a little like, I don’t know, going for a walk and seeing something interesting that catches your eye and prompts you to try something new or another direction. Now, it isn’t going for a walk because that is a different thing, so let’s be clear there. But what it is, is a sort of directed-random exploration of a space, which is exactly how these systems work in the first place. It just happens to be that the fidelity is high that you may find exactly what you’re looking for on the walk, but that’s an example of approach.
The text generation example would be, I don’t know, going up to the Librarian at the Library of Babel, asking for a specific text, and then getting a map and directions to a book that contains, as much as can be gleaned as possible, the text that you’re asking for.
I feel like a point has consistently been made that “if you’re good at something that AI is doing now, then you probably don’t have anything to worry about”, but the more specific version of that is:
Now, they might have a shit-ass boring brief where they just want you to make something like something else but, you know, with your name on it for bragging rights, in which case if you had a choice you might choose not to work with those people.
But otherwise, I think that’s what “good” means, and it means having the reassurance and the ability to build up the audience.
I do not think it is as much a job as “prompt engineering” as, like I’ve said before, in editing and knowing what to pick. For many uses of AI tools, “knowing what to pick” isn’t that much of a big deal because what you pick doesn’t matter. It does not matter, unfortunately, or people treat it like it doesn’t matter whether you pick performance review report option A, B, or A^10239^.
It might matter more in a film, or a story book, perhaps? And now we’re back to curating and we’re back to, I don’t know, the debate as to whether your qualia-sense is so good that it can pick out the right thing from a selection of things versus someone who, I don’t know, doesn’t have taste and is impressed by pretty picture which means AI has emotion!. Sorry. I digress in a mean way.
Chadeisson is an example of an accomplished, sought-after professional artist whose job involves, well, translating briefs (“prompts”) into evocative imagery that can then be used in production work. That certainly sounds like a job that could have the bottom drop out of it. One of his commenters asks “When’s your film coming out?”, to which his answer is “When there’s good AI scriptwriting software!” and, well, there’s a point there, even if it’s an obvious one.
One way of thinking this through is that Chadeisson’s work might disappear if:
If all that happens, would Chadeisson’s work disappear? If the market for such creative work is saturated by quick(er), easy (easier) to create films and tv shows, will audiences still choose enough “premium”, human-influenced work?
There is something here – a thought that I have to reluctantly leave for later, because this episode has gone on for far too long already – along these lines:
You know, if I were Meta or whomever, I’d be interested in developing the infinite gallery, a never-ending, high-res 3D environment of AI-generated work, where virtual navigation is part of the prompt refinement. That’d be a fun lark for a week or so.
Phew. A longer one again.
How are you doing?
P.S. This entire newsletter episode was generated in a 15 minute process with a large language model transformer additionally trained on the entire corpus of these newsletter episodes.
P.P.S. No it wasn’t.
P.P.P.S. Or was it?
P.P.P.P.S Oh go on, reply and let me know what you think :)
Things That Caught My Attention is a free newsletter, and if you like it and find it useful, please consider becoming a paid supporter, at pay-what-you want.
Do you have an expense account? Money starting to burn a hole toward the end of the year? Let your boss pay, at $25/month, or $270/year, $35/month, or $380/year, or $50/month, or $500/year.
Paid supporters get a free copy of Things That Caught My Attention, Volume 1, collecting the best essays from the first 50 episodes, and free subscribers get a 20% discount.
Paul Chadeisson’s linktree and Paul Chadeisson on Artstation2. ↩
Artstation, a portfolio site for artists, is itself a subject of protest against how it plans to use AI art, see: Artists Are Revolting Against AI Art on ArtStation, Chloe Xiang, Motherboard/Vice, 14 December, 2022, and Some Of My Favorite Images From ArtStation’s AI Protests, Luke Plunkett, Kotaku, 20 December, 2022 ↩