A question for published, income-earning writers:
As I watch the proliferation of writing books on Amazon advising how you can write fast and get rich quick writing books with ChatGPT, I'm wondering which genres you think will be hardest for AI-assisted writing to crack. (Most of them seem to be advocating AI as a rough draft tool, then a polish & Bob's your uncle.)
I'd love to know your thoughts: which genres will be most AI-proof?
(sighs) OK, get comfortable.
First, I’m going to suggest that AI is, for a certain diminished quadrant of neurotic writers, going to become a fetich object—something onto which they can project their fears of failure and success. You know writers who can only write if it’s raining, or if they’re newly in love, or if they’ve had a few drinks? Same kinda thing.
1/x
It’s the plagiarist’s paradox, really. Plagiarizing WELL is just as much work as writing from scratch, but it’s a different kind of work so, if you steal instead of build, you can do an end run around blank-paper paralysis. Some jabroni is going to get an AI to write a novel, then rework it until it’s OK, and get it accepted, and TRULY BELIEVE the AI wrote it because he thinks he’s too stupid to do what he patently actually did.
2/x
It’s sad, but, no sadder than the alcoholic writers of the Dorothy Parker era. :/
As for what genre is most vulnerable to being REALLY replaced EASILY by LLM? None of them, for the same reason that mirrors aren’t going to replace faces. People can tell the difference between a THING and the empty shape of a thing.
Which is not to say there won’t be SO much wasted time and economic harm and stolen effort and scamming and misery before people realize this.
3/
What’s actually replaceable by AI is written porn, because nobody really wants surprises and vulnerable genuine connection from their porn. (People who want surprises and vulnerable genuine connection read erotica, which looks the same but isn’t, IMO.) AI is, I’m sure, able to churn out loads of perfectly acceptable wank-fodder.
Next most vulnerable is literary fiction!
4/
Literary fiction LOOKS like a soft target because you don’t need world-building like SF and fantasy, and you don’t need a clever twist like a mystery. But the special sauce of literary fiction is the beautiful LINE, and here’s where AI is going fail except when it creates some unique metaphor by accident and its human minder pulls that one OK phrase out of the hundreds of humdrum cliches. Now do that twice for each page and you might fool somebody.
5/
@ShaulaEvans But for even a 200 page novel, that’s 400 “good bits” and assuming a generous 1-in-20 random good bit generation rate, that’s some poor bastard having to read 8000 chunks of absolute dreck to mine out the quality. That’s AFTER setting up 400 separate and unique and fussy queries to get those 20 options. Sounds suspiciously like work…
6/
Romance? Similarly, it looks simple on the surface because you don’t have to invent elf culture or alien biology, plus the ENDING IS PREDETERMINED. It’s a HEO! The lovers unite! It’s practically an outline, right?
But again, the devil is in the details. Good romance is all about human interaction—painful, funny, flirty, coy. A romance novel isn’t actually “the lovers meet” it’s “these PARTICULAR, well described and unique people, meet.”
7/
The beauty of a love story is not that they fall in love, which is hardly surprising given the label on the genre. It’s seeing the particular way the pair is right for each other, even when THEY don’t see it. It’s the specifics.
AI is shit at specifics. It is built SOLELY for cliche and generalities.
8/
Finally, the most impervious novel genres are SF and fantasy, IMO, because they need unique and elaborate world building and—see the pattern?—AI can only do what is most typical and, therefore, what you expect.
But the LEAST likely is poetry. Because AI writing, as has been noted elsewhere, is the exact opposite of poetry.
9/9
(And apologies for going on and on. I didn’t have time to edit it down to “pithy.”)
I'm curious what you mean by "opposite" of poetry. (I can see how attempts to get a language model to create poetry are missing a large part of its concept, but I don't see the qualities that would make it opposite of it.)
@robryk @ShaulaEvans
Poetry is, at best, expressing a universal sensation or idea in a novel way.
LLM can only do predictive text in expected ways. The entire purpose is to NOT be novel.
And of course, it can't shed new light on sensations or ideas because it has neither. It's just a jumble of words and probability.