Imagine I am lent a book by a friend. Let’s say it’s a work of fiction. It looks interesting enough. I start it. It is, indeed, interesting enough. About twenty pages in, my friend tells me that it was written by an AI.
Here is the thing I am stuck on. I just cannot imagine sustaining interest past this. Maybe I would read one book of this kind, insofar as I am curious about what a computer is capable of — but certainly not as a work of literary art.1 Maybe, if the plot were “interesting” enough, I’d skip to the end and see what happened. But in the same way that I would not waste my time applying for a library card at Borges’s Library of Babel2, I would not waste my time reading this book.
Every time I am presented with an extended piece of AI-generated writing, my eyes glaze over. So here is my brief question about AI-generated writing: Who the hell is going to spend their time willingly reading this shit? I literally do not get it.
I am sure there are people who will. I don’t doubt their existence. I just can’t imagine it. This is a personal problem, maybe, but it’s one I keep coming up against.
The more ambient a cultural object is, the more I get it. Would I listen to background noise while I work, generated by an artificial intelligence, according to parameters I choose? I mean, sure, man, but who cares. I wouldn’t pay extra for it. Half the shit on Spotify is generic “lo-fi hip hop beats to study to” nonsense anyway. As long as there are retail spaces and people doing their homework in coffee shops, what we used to call elevator music will have a place in society. Maybe there are people who make this now who will lose their livelihoods, and they definitely shouldn’t, but media whose use-value is primarily subconsciously influencing people to buy more consumer goods or increase their productivity isn’t exactly what I’m talking about.
If there is something useful that can emerge from this condescending morass, it is a refinement of our answer to the question of what a work of art is and can do. I suppose what I’m learning about myself is that I am committed to a kind of higher-order intentional fallacy — aesthetic objects that do not express consciousness in some capacity just aren’t interesting to me. The work is not entirely autonomous, because it is expressive. Form is a means of communication, and a computer cannot communicate: it regurgitates. And while puking serves all sorts of necessary somatic functions, I’m not trying to fucking eat it.
The counterpoint, or whatever, is that there will probably be an intermediary consciousness prompting and sculpting the work. But a good book is fractally expressive, expressive in every sentence. Every word choice is a word that has been chosen. It’s easy to (literally) skim over this, but I think it’s important. Give me two identical texts and the one you tell me was written by a person would be categorically more interesting to me.
On certain assumptions, this might be irrational, but assumptions that are, strictly speaking, irrational undergird everything we do. The idea that the text should be encountered free of context3 is just as made-up as saying you have to know the biography of a writer to understand it. Or, like, I think living creatures are intrinsically dignified, and deserve to be treated as such.4 Why? Because the world that follows from this is the world I most want to see, and which my experience most supports belief in. But that’s not an argument, and it’s not an argument in the same way as what I’m saying here. I don’t mean to newslettersplain the idea of first principles to you guys, but it’s just, like, agh!! What the hell!!
Imagine flipping the terms. I can write all I want, but the only entities which will “read” what I write are the large language models harvesting data from my text. The terms are the same — one person, one AI — but they’re in the other position in the equation. It seems manifestly, glaringly obvious that, if this were true, I would quit writing anything but the occasional diary entry and spend a lot more time trying to think up Dungeons and Dragons campaigns for my friends. I see no real reason to believe this scenario is any more or less meaningful than the converse.
Look. I don’t doubt people are already passing off AI-generated writing as human-made, and I don’t doubt they will continue to. I don’t doubt that there will be controversy and perplexity and consternation to come. This question, which is about the material conditions of writing and publishing, is ultimately the more relevant one, probably. But passing off AI-generated writing as human-made is lying, and doing something because I am lied to is a different problem. Assuming, as I cannot but would love to, that people aren’t lying, I think my point stands. The world is saturated with interesting things. I probably won’t ever find the (human-written) book that would be my actual favorite in the world.5 All of which is just to say who on earth has time for this shit.
Before its portentous and tragic downfall, that AI-generated Seinfeld parody was extremely funny, but ten more wouldn’t be funny, because the joke was the uncanniness vis-à-vis the source material, which is a) obviously directly predicated on the existence of the (human-made) source material and b) exactly one joke, which does not much more than a short-lived Twitch stream make.
If you haven’t been to the website that simulates the library (and Borges’s story is here if you want to read it; it’s cool), it’s useful, both because it’s rad on its own terms and because it emphatically makes the point clear that, practically speaking, everything you’d ever read in that wretched place would be total nonsense.
I actually think New Criticism gets sort of a bad rap — I.A. Richards in particular was working towards something actually remarkably egalitarian, a highly portable pedagogy designed to teach readers of poetry to make aesthetic judgements on their own terms, before weirdo Lost Cause freaks at Vanderbilt or whatever took it up later. But whatever.
My copy of Groundwork of the Metaphysics of Morals is lost to the all-devouring sands of time but I’m pretty sure even Kant says that the only reason people will decide to act on the categorical imperative is because they feel it’s good to do. You can convince them to feel this way but there’s still an irrational impulse under there. I’m sure there’s more about feelings and shit in the three critiques but I don’t think I’m going to read those in the near or probably even the far future so do whatever the hell you want with this undergraduate-ass point, I just wanted to put it here. (Maybe I should read them? Sound off in the comments. Lol)
Honestly I think the best book in the world is probably a file on some random person’s laptop in Jakarta or rural Romania or Guam or Bamako or something named like “manuscript FINAL ACTUAL FINAL Really the last one.docx”. But I probably won’t find my favorite published book, either.
AI "creative" writing is better than pages of babel and pressing the next suggested word on your phone's keyboard over and over, but not by a lot. I remember going to an exhibit a couple of years ago that had a receipt printer spitting out lines of AI generated poetry and I really hated it. I would have rather watched a CRT playing static. At least then I wouldn't be distracted by the pointless exercise of trying to chase after some sort of meaning in the words and could instead step back and contemplate the piece more holistically. Maybe that's what I'm doing now. Sorry artist, I finally get it.
Beyond the necessity of human intent, part of this deficiency is due to the biggest baddest model, GPT-4, having been RLHF'd to suppress any personality, and the models that don't have this limitation are running on much smaller training runs. Open-source language models a couple of years down the line will likely be more idiosyncratic. That's not the same thing as expressing consciousness, but I do think it's part of why the current state of the art (so to speak) is so especially soulless.
Dwarf Fortress + language models will be coming (https://arxiv.org/abs/2304.03442) and I can imagine people getting really rapt by that. A little further down the line, "computer: re-render this movie but where the bad guy wins in the end" ways of interacting with media.
I would write more about this continuum between bespoke & authentic hand-crafted Art, Kid A, generative/chaotic systems art, and machine learning art, but I need to do other stuff.
I enjoyed this piece!
I meant to write "circulation desk" not "reception desk" but I can't edit these so I"m putting it here. Okay see ya