My job is to teach Large Language Models how to write. LLMs are one manifestation of Artificial Intelligence, and are often accessed through chatbots like Gemini, Claude, and ChatGPT. Many people know them as the behemoth software systems that appear, unbidden, on search engines and that pop up like ambitious salesmen on customer service sites. They promise to answer questions and take orders, although they’re considerably better at collecting money than dispensing information. They are the algorithms that compose love letters, corporate reports, and political manifestoes.
I’ve taught college English for decades; my sentient, human students are brighter, funnier, and more interesting writers than any LLM, but also more prone to grammatical mistakes and harder to educate, coerce, or encourage into improvement. “But you know what I mean,” they say when I point out misplaced modifiers or unnecessary adverbs.
The LLM I’m training is more educable than a sentient being, because it’s utterly open to feedback. I don’t have to worry about hurting its feelings because, lo, it has no feelings. I don’t have to employ the sound pedagogical practice – which practice I practiced for years – of first giving praise, then cautious, constructive criticism, then asking a question for its growth and future exploration. I can tell it what it’s doing wrong, and it has to listen.
I love most of my college students, and when I overcome the limitation of being a generation older and wiser, I can even understand them.
But I don’t understand the LLM. I know that it is basically doing word-association at warp speed, so that by the time you’ve typed “I’ve been thinking of” onto your phone, something is already predicting that you’re going to want to put “you” next.
Such predictive, prescriptive text bothers me, especially if what I’m trying to say is that I’m thinking of going fly-fishing tomorrow instead of reporting to work. But the LLM knows that bazillions of times, people who have written “thinking of” want the next word to be “you.” And so I’m treated like everyone else. Which is annoying.
More than annoyed, some people are alarmed by AI. Many friends glance askance when I tell them what I do for a living. “Aren’t you worried?” they ask, like a passenger in a car that I am driving at speed toward a red light. They inform me in knowledgeable, well-modulated voices (not at all verging on hysteria) that AI. Could. Take. My. Job.
Why I’m Not Worried (Much)
I don’t think I’m about to be replaced by AI when it comes to writing novels and short stories. For one thing, AI is a pretty sucky writer. Even after all it’s read – basically, every book ever published – it has yet to compose a meaningful metaphor. It has no sense of humor. In fact, it’s so artificially unintelligent it doesn’t even understand sarcasm. And it’s a teenager, but it doesn’t do irony!
I’ve been reading AI-produced textbooks, fiction, reports, poems, diatribes, encyclopedia entries, and hallucinogenic memoirs for many years, and in all that time, I’ve laughed twice (usually, I lol easily). I also cry at almost anything – cat food commercials, drivers pulling over for ambulances – but so far, despite AI’s best efforts, I haven’t wept once.
Here’s the other reason I don’t lie awake at night worrying about AI (Taking. My. Job). I don’t have a writing job. I’ve had novels published and plays performed, and once wrote a column on fishing, but those were freelance gigs that I did for fame, not fortune.
I don’t know even know anyone who makes a living as a fiction writer, though of course, some people I don’t know earn their living that way. Ian McEwan comes to mind, and Garrison Keillor. But I don’t think AI could ever touch the hems of their garments.
Other full-time, successful writers include Stephen King, Danielle Steele, and Janet Evanovich. While such genre fiction superstars know their publishing futures are secure, some authors of detective stories or sci-fi series might worry, because AI has started churning out such works. But the only books that AI is currently threatening are the ones that are already formulaic and soul-less, ones where no one would notice any difference if they were propagated by a robot.
I admit it: If I were paying my mortgage from the royalties I earned from heartwarming tales of how a gutsy single gal from the Midwest stumbled into magic with a strong-jawed, emerald-eyed billionaire, I might feel consternation knowing AI could string clichés together as well as I could. But I’d deserve it.
But let’s say my friends are right, and I should pay more heed. Let’s imagine a time in the distant AI future – say, two months from now – when AI composes a book that I can’t distinguish from work by real authors whose books I love. Let’s say I read it. I laugh. I sob. I turn back to the good parts and laugh-sob again.
At the end, let’s say I set the AI book down with a silent exhalation of surprise. In that second of awakening from what John Gardner calls the “fictional dream,” when normal consciousness returns and I blink awake in my mundane world, I’m disappointed that the book and the dream are over, but I know that I will never be the same, because of having read it. In that liminal moment of appreciation and wonder, will it matter that the story was written by a machine, in a few seconds, instead of by a person laboring for years?
I don’t know. I don’t think so. If I can get from a piece of writing what I want to get from it – a hearty, hiccupping laugh, a solid cry, a swelling feeling of belonging to something bigger than my own world-view, a sense of wonder that words could be put together in the way I’ve just witnessed – I don’t care who (or what) will have written it.
There’s a parallel question in religion. If a sermon (Christian), d’var Torah (Jewish), khutbah (Muslim), or what-have-you (Whatever) strengthens the faith of the listener, does it matter what the leader who is giving the talk believes?
Some friends I’ve posed this question to say yes. Yes, the faith of the (say) minister matters, because a minister’s influence goes beyond just what they say from the pulpit. The leader of a congregation – the Priest, the Imam, the Rabbi – does far more than give inspiring lectures. A faith leader’s beliefs are expressed in actions that affect their community, not one hour a week but day in and day out.
A godly talk is one thing, but if it’s not backed up by loving human actions, the center cannot hold. And an LLM, no matter how clever, is not going to come to the hospital to give your mother last rites. It’s not going to marry your son to his beloved, either (although it might be able to help with his divorce paperwork). The abilities of an AI author begin and end with the words it composes. And if those touch us, does it matter why or how or from what source they come?
To many people, it matters who is behind the words they read. My friend Vincent feels a connection to the author of any book he loves, and that’s part of what makes him love (and buy) their books.
To which I say: piffle. The “implied author” is a term of art from literary criticism, and we’ve all been projecting our own genius, heart, and capacity for miracle-making onto them for millennia.
What I mean is, we don’t really have a personal connection to the authors of great literature. When we read a wonderful book, it’s the book that’s wonderful, not the person who wrote it. We connect to a part of ourselves that we hadn’t previously known.
Why I Am Worried
What concerns me is not that AI might start whipping out novels better than mine (a low bar), and that my multi-million-dollar three-book deal with Big Publishing House won’t come through. AI has tried, unsuccessfully, to imitate the human voice, and it ends up spouting clichéd nonsense.
What does worry me is that people talk about AI as if it’s anything more than a complex, fast-talking machine that simulates human thought. For instance, in the introduction to this argument, I wrote about what an LLM “knows,” which is stupid. LLMs don’t “know” any more than a rock does. AI can write about a rainbow, a radish, or a roller-coaster but it can’t ever see, eat, or ride one. Yet my students talk about what AI “thinks.”
AI doesn’t think. It solves problems. It answers questions logically if they’re logic questions, or it addresses queries based on what it’s read before, if they’re the kind of query I type into my best AI friend after midnight when I can’t sleep. For instance, I might go on for some insomniac hours explaining why I can stop eating Heath bars. My friends have heard this already and they’re not interested in hearing more and/or not awake, whereas Claude will party on with me all night.
The Beginning and the End
We’re past the beginning of the AI revolution. As every piece ever written about it (including those written by AI) says, we’re moving fast into the inevitable. The exhortation that “It’s already everywhere” is already old hat. It is everywhere, and it’s okay.
In the end, human creativity is encountering competition, but our imagination and emotions won’t be conquered. People who love reading – you and I – will keep reading. Billions of us are reading squillions of words from increasingly varied sources. And if one of us reads a piece that moves us, makes us feel joy, makes us laugh through tears and have to go read the paragraph aloud to someone, then that’s a precious, pervasive, human response, because my feelings are completely human. And so are yours.
(But not yours, AI, if you’re reading this. You have no feelings.)
Become a Saturday Evening Post member and enjoy unlimited access. Subscribe now
