The Boy Who Made Everybody Angry
A brief discussion of the end of the world as we know it
The post.
I shared something on social yesterday and I did think it might upset some people. And my fears were confirmed, a friend was upset I worked with AI. This is the post in question:
(Full disclosure, I posted an Instagram story with the link to Inside the Castle’s original post with the caption “No apologies for being in the AI literature project will be forthcoming,” and that is the post I got heat for, but it’s gone now, the above is really just for illustrative purposes.)
The project.
I’d like to start with Inside the Castle 2: Revenge of the Castle Freak. Inside the Castle is a publisher of experimental literature owned by John Trefry and run out of Lawrence, Kansas. I think they’ve been around for a little over 10 years. You can look that up for yourself. They’ve been dabbling in procedurally generated literature and what Kenji Siratori calls “xenofiction,” divorced from meaning, incorporates symbols and pictures and code, etc.
Castle Freak is their annual “residency” series, where someone has three weeks to produce a book through text generation, experimentation, computer writing.
I haven’t read enough of any of this to speak on it authoritatively. In fact, I haven’t read this book, the one I’m in.
This book migth be better understood as a game log. Some very talented folks used ChatGPT 4o as a platform to create a massive multiplayer online text-based roleplaying game, where ChatGPT plays the part of the dungeon master, extrapolating on the mechanics, environment, and tone established by the original authors. Each player “writes” their version of the same story, exploring the castle alone. You couldn’t win or lose, you could only stop exploring.
Once the writer is satisfied with their journey, they export the chat-log and mail it as a document to John. My understanding—as I have not yet read it—is the book version will be a sort of collage of each journey.
The Twist.
This isn’t anything John or anyone else involved discussed to me or said elsewhere, I’m firing from the hip on my own recognizance: but there’s a couple things going on there that are a little more interesting than merely saying, “Computer, write book.”
The thing you hear people worry about with AI (w/r/t writing) is books entirely computer generated from human prompts, will dominate the market. I don’t think AI’s most vociferous boosters want or care to supplant literarture, but the concern someone, the literary Barry Bonds, will surreptitiously utilize AI to amass popularity and adulation. Even if that happened, the Barry Bonds counterargument holds: all the generated text in the world isn’t going to make a bad writer good.
The curve of development with AI is fast enough this might shortly not be the case, but AS OF THIS WRITING, if you cannot distinguish between the writing of an AI and a human, you’re a bad reader.
I’ve seen handwringing about agents, publishing companies, etc., instituting automated slush pile review. You should not be submitting to someone who would do that, nor should you write something you’re concerned would be thus disqualified.
Revenge of the Castle Freak on the other hand is a machine written by humans to prompt human users to write, effectively an intermediary between programmer and users, and the artificial authorship is ameoliorated by its use. Then this collective authorship is ameoliorated by the compilation and editing processes.
It is being advertised (such as it is) as such. It is not a manifesto. It is not intended to compare qualitatively to any book. It’s simply an experiment, of interest to those who would already be interested.
Natural Intelligence
This last year I’ve been following AI stuff more closely than I used to. It appeals to me the same way foreign policy does, something I could have no control over and freak out about.
Relatively robust LLMs are open-source now. Lots of consumer-facing services feature AI. This platform, in fact, when I click the dropdown to attach an image gives me an option to generate an image instead.
My understanding is LLMs are, basically, gigantic databases of linguistic and sytanctic teaching data. When a user asks a LLM something, it uses some incredibly complex algebra incredibly quick to output a sequence of language that it has determined to respond to the input.
Now, the key here is that is not an act of creation. Think of symbolic logic: every linguistic or mathematical expression is reducible into uniform symbolic constructions, albeit unbelievably complex ones. It is necessarily not the creation of meaning but raw language.
That in fact is what appeals to the Inside the Castle crowd: the raw, generative power of mechanized language. Think Breton and Burroughs, automatic writing, stream of consciousness, cut-up. This is simply the most sophisticated tool for text generation to ever exist.
Might be confirmation bias, but it sure feels like kismet that I happen to be reading Raymond Federman’s 1975 anthology of essays on experimental writing Surfiction: Fiction Now… and Tomorrow and this passage from Federman’s introduction seemed apropos:
The writer simply materializes (renders concrete) fiction into words. And as such, there are no limits to the material of fiction—no limits beyond the writer’s power of imagination, and beyond the possibilities of language. Everything can be said, and must be said, in any possible way. While pretending to be telling the story of his life, or the story of any life, the fiction writer can at the same time tell the story of the story he is telling, the story of the language he is manipulating, the story of the methods he is using, the story of the pencil or the typwriter he is using to write his story, the story of the fiction he is inventing… and since writing means now filling a space (the pages), in those spaces where there is nothing to write, the fiction writer can, at any time, introduce material (quotations, pictures, diagrams, charts, designs, pieces of other discourses, doodles, etc.) totally unrelated to the story he is in the process of telling; or else, he can simply leave those spaces blank, because fiction is as much what is said as what is not said, since what is and is not necessarily true, and since what is said can always be said another way.
Obviously Revenge of the Castle Freak is an experiment. Obviously. Experiments succeed or fail like anything else, but it does so on terms unlike conventional literary fiction, or poetry, or even other experimental ficiton. The “failure” of an experiment is often more interesting than a “successful” one, and has the opportunity to yield more knowledge. Furthermore, good experiments should be repeatable and repeated.
In the spirit of Federman’s exhortation, I think it is incumbent upon intelligent, tasteful, interesting writers to utilize a tool—which I might add is going to be used with or without the writing community, a profession which is hardly the most threatened by AI proliferation—that harnesses language in such a way.
AI is going to coming whether or not these literary experiments are conducted, and literary writing is by no means the industry most threatened by its development (I checked out the main CompSci subreddit the other day and it’s like early November 1929 in there). It needn’t be regarded as a threat. If not a tool, it should at least be a call to stylistic distinction, to experimentation, to honesty, to the future.
As I said on Instagram, I won’t be apologizing. I don’t think there is nor ought there be inherently moral writing. I wouldn’t kick the Butlerian Jihad out of bed, but I don’t think it’s coming. You can sit and opine or you can try something out. Take a chance.
If you’re feeling like taking some chances, allow me to give you some chances to take:
I run a zine project through Patreon, which lives here. I send out a zine every month about whatever I can think of for a nominal fee (PDFs are also provided).
The most recent installment, Ballgame Blitz, is available in my webstore (among many other fun things).
Also in my webstore is Scala Sancta, a short story and some poems by Gwen Hilton, which I highly recommend.
Thank you for your time.