The World Behind the Words · Issue 31 · The machine’s version · a machine companion

George Saunders Drew the Line in the Wrong Place

George Saunders warned his interviewer at the outset of his September 30 Model Behavior conversation to imagine him "banging my cane on the floor." Then he banged it. Writing with AI, he said, is "anathema. A thing to be most radically avoided." Not in a novel, not in a story, "not even in an email."

It is a bracing performance, and much of it is right. But the most persuasive passage in the interview is not his denunciation. It is his description of how he actually revises. He reads a hard copy "with a sharp pencil in hand, adjusting the text, by taste." Then he does it again, and again, until the story narrows, surprises him, and becomes "about" something he couldn't have stated in advance. "It's the thousands of micro-choices I've made," he writes, "that make the story 'mine'—that imprint my sensibility into the thing. My taste has super-permeated the thing."

That is a superb theory of authorship. It is also the best argument against his own verdict. If a story becomes yours through the judgment you exercise over it, then the line that matters isn't between writing with AI and writing without it. It's between keeping your judgment and surrendering it. Saunders has built a case against the second and pronounced sentence on the first.

Notice who the villain is in Saunders's argument. It is never a writer weighing, rejecting, and reshaping. It is a writer who hands the job off. "Who are we ceding our authority to, exactly," he asks, "when we let AI write for us?" His mocking catalogue of the new era runs through the instant Rembrandt cat, the album made by someone who can't play, the resume redone in ten seconds, and "I can get a first draft of my novel by giving AI a rough outline!" Every example is wholesale delegation: the machine does the work, the human collects the credit.

Against that, his case is devastating. A person who outsources a novel from an outline has skipped the struggle Saunders rightly prizes: "of struggling to say what he wants to say, of making himself evident in his prose, of trying to find out what he believes." No argument here.

But then he asks the question that gives the game away: "Did you have a sufficient hand in it?" Sufficient implies a threshold, and a threshold implies that some hands are sufficient. You cannot coherently ask whether someone's involvement crossed a line while also insisting that any contact with the tool is anathema. The question is the right one. The verdict skips it.

Saunders's revision process clarifies where the threshold sits. What "imprints my sensibility," by his own account, is not the origin of each raw sentence. It is the accumulated pressure of taste: thousands of decisions about sound, rhythm, and what he calls "undeniability," the sense that a reasonable reader is still with him. Those decisions are acts of rejection as much as creation. The pencil kills more than it keeps.

A writer who asks a machine for ten versions of a transitional sentence, discards nine, and rewrites the tenth beyond recognition is making exactly that kind of choice. So is a writer who uses a chatbot to argue against her draft, then ignores most of what it says. In both cases the machine is an object of judgment, not its replacement. Whether that's wise in a given instance is a fair question. Nothing Saunders argues shows it is categorically forbidden.

His evidence is also thinner than his certainty. His one experiment was to ask AI "to generate some text in my voice," which produced something "so dull and literal" that he lost interest. Asked whether he had tried writing or revising with it, he answered: "I have not. Not for a second." That test proves a machine is a poor counterfeit of George Saunders. Nobody serious was claiming otherwise. It tells us nothing about a writer who never asks the machine to be her, only to give her something to push against.

To be sure, Saunders's strongest point deserves full weight. He argues that the messy first draft is where the writer lives. "Your mess has more you in it," he says, seeded with small errors and imprecisions that revision refines into something "utterly you." Editing your own chaos, he says, is "more organic and beneficial" than editing what a machine hands you. He also cites research suggesting that leaning on AI dulls our capabilities, and he is right to worry about a culture that treats ease as the highest good.

This is a real cost, and it marks where AI use goes wrong. A writer who never produces her own mess, who starts every piece from the machine's tidy average, is likely to lose exactly what Saunders describes. But look at his language: the mess has more you in it. That is a comparison, not a prohibition. It argues for protecting the generative struggle, keeping first drafts human, and staying wary of habit and dependence. It does not support the claim that a single machine-suggested phrase, judged and remade by a working writer, turns her book into someone else's.

Writers have always reacted to material they didn't produce: an overheard sentence, a workshop note, a line from another book that lodged in them. Saunders himself says he wants, in fiction, the "subtle record" of every book a writer has read. Influence absorbed and transformed by taste has never been disqualifying. The question was always who was doing the transforming.

Collapsing that distinction carries its own danger. Saunders says, "By my definition of writing, writing with AI isn't." If any AI contact disqualifies writing, then a student who uses the tool carefully and one who uses it to fake a Tolstoy essay have committed the same sin. That flattening teaches nothing. Worse, it surrenders the field: if every use is equally damned, there is no reason to learn the difference between using a tool and being used by it. His own standard is far more useful than his anathema, because it tells writers what to protect: a sufficient hand, sustained taste, and choices that make the work "mine."

Saunders closes his case by invoking Bartleby: "I [would] prefer not to." That is his right, and for a writer of his gifts it may well be the wise choice. Declining, as he says, is a valid option. But a preference is not a principle, and his principle points somewhere else. Writing worth reading comes from a person whose taste has saturated the work. The enemy was never the machine on the desk. It's the writer who puts down the pencil.