I found Tim Boucher via ooh.directory’s Recently Updated feed, which I maybe need to check more often. I wonder if it’s possible to put it in a digest format? I don’t want all that in my RSS reader. It’s interesting that the site estimates update frequency and lets you filter by that, but not (say) category? Like it might be cool to be able to follow all recently updated blogs within Writing (of which mine remains, inexplicably, one of 10, alongside Austin Kleon).
… anyway, I’m skeptical of the project Boucher outlines in this thing, but so is he. Which is what I’m interested in. “I honestly thought the results would be more ‘out there’ but something about Midjourney still has a way of leveling off the highs and lows, and making the outputs always look, well, Midjourneyish.” I first noticed this listening to an episode of Mark Leslie Lefebvre’s podcast (I’m not hunting up the link, but it would have been in the first half of 2022, plus or minus) that he created in an AI version of his own voice — AI audio sticks to the thick part of the distribution, creating a product that’s very recognizably the source’s timbre but flattening out the variations. There’s a sameness to AI art that I’m starting to recognize as well. Not so much with the photorealistic stuff, but much of the more “arty” looking product (which is the kind of thing I run into more often on LinkedIn, blogs, &c) has a sort of half-3D look about it that starts to be distinctive; the best way I can describe it at the moment is it feels like a compromise between 3D rendering and the sorts of ways you’d render depth with a brush or pen. It does make you wonder if there’s some sort of psychometrics on this — show people 10 human-produced images and 10 images produced by AI in the same style, see if viewers can detect it?
Boucher’s “I feel constrained by AI” is also an interesting short post:
… at first it seemed like these tools gave access to unlimited spaces, and the only edges were my own imagination.
But after a year of heavy lifting, I know for a fact that this is untrue. The edges are many, the spaces in fact quite small, the further you go into some of them.
Tim Boucher, “I feel constrained by AI”
… which sounds like a vindication of the skeptics, but then:
Constraints can be both breeding grounds for innovative explorations, and also a place of great frustration, where you constantly have to miss on the grand vision, and settle for many of the smaller ones instead, piled up high as you can make it.
Tim Boucher, “I feel constrained by AI”
… which is, of course, one of the oldest ideas in art, and continues to motivate (or at least rationalize) the use of restrictive artistic forms today.
It also maybe brings back the basic point of any exercise like Boucher’s AI Lore Books, which is to tap the deepest wells of human creativity. Trying to put your thoughts into a villanelle or render a scene in pen and ink doesn’t buy you creativity for free; the frustration Boucher mentions after the semicolon is the engine of the innovative exploration he mentions before it. If you can’t do it the easy way, how do you do it?
Which is the opposite of how almost everyone who thinks about AI-assisted art seems to be thinking about it. Skeptics and doomers are worried about generative glurge replacing the fruits of human labor. Boosters tend to emphasize how AI can make things easier, de-tyrannizing the blank page or joggling the author’s mind out of a local minimum. It’s not often you hear the limitations of AI acknowledged in the same way as the limitations of any finite artistic tool.
None of this gets to the ethics of how these things are created and maintained, and I’m not going to try to force it to go there. But it is of course worth keeping front of mind. In that context, it’s interesting to think about Boucher’s title for this particular piece of work. The proliferation of AI text and art on the very Internet that’s used, with little curation, to train the next generation of algorithms is famously courting “model collapse,” where contamination of the training data with AI-generated output causes the new model’s output to grow more and more outlandish (from the abstract, note the telling phrase: “tails of the original content distribution disappear.” Hugging the thick part of the distribution!). It’s doubtful the factory will be permitted to dissolve any time soon; but the danger does seem to be there.
Currently reading: SOLANIN, by Inio Asano. Acquired at Second Story Books on a recent visit to DC; recommended en passant during the Mangasplaining episode on GOODNIGHT PUNPUN.
If you’re enjoying my writing, you can get some of my short fiction on your e-reader for the low, low cost of $0. Remembered Air is a collection of six poems and short stories not available anywhere else. Download it here.