In Idée Fixe, Paul Valéry presents some of his ideas in the form of a discussion between him and his friend, a doctor:
—I’m a Crusoe. With my home-made bow and arrows, I shoot down what birds I can—when there are any.
—And there usually are, I suppose?
—The sky of the mind swarms with them, parrots in particular. They have to be killed off first of all . . . before catching and taming the others.
—So you’re a parrot hunter? . . . You must work long hours, in that case?
—Often enough I do, Doctor. And you’ve been raising flocks of them for me.
—I have? . . . In what sense, what do you mean by parrots?
—Oh Lord. . . . I mean three words out of every six, pretty well. All the words that fail to survive that central scrutiny . . . intact.
Petrick’s note is a good example of a parrot that needs an arrow through the neck. Everything about it, from the animating idea to the smug formulation overly pleased with its rhetorical tidiness, fails to survive scrutiny. It’s a pleasing little epigram, especially if you already agree with the sentiment. But what does it actually say?
“AI-generated writing” is being used as a catchall term for everything from lazy high-school students using minimal prompts to produce generic text to fulfill an assignment, to the experimentation that someone like Venkatesh Rao is doing, and all points in between. A writer might use AI to brainstorm, test arguments, conduct research, generate alternative formulations, or produce a complete first draft. He might accept its suggestions, modify them, reject them, or use them to discover something he hadn’t previously considered. In some cases, the final text will be almost entirely generated by a machine; in others, no machine-generated sentences will survive at all. Are these activities sufficiently similar to be treated as one thing?
Why do people resist it? For a variety of reasons. Some are ideologically committed to opposing AI in the name of some concept of organic humanity. Some are inherently distrustful or anything new or strange. Others have no real knowledge or convictions, and are just mimicking a trendy attitude among their peers. But as we’ve been noticing, even those who are most fervently opposed to AI writing in theory seem to prefer it in the literary equivalent of a blind taste-test. Even “AI resistance” turns out to be hard to quantify. It may be more of a posture or an attitude than anything substantial.
“Something that no one wanted to waste time writing.” Again, the assumption seems to be that all AI use resembles a bored schoolkid doing the bare minimum to satisfy an indifferent teacher. But a sentence that took an LLM seconds to generate might have been preceded by a lengthy conversation, an afternoon of research, or an extended process of argument and revision. You can’t tell, merely by looking at a sentence or paragraph, how long the author labored at it, or how much thought went into it. Essays don’t have that sort of price tag hanging off of them. And why do we insist on trying to apply the labor theory of value to writing anyway? A writer can labor for weeks over something utterly worthless. Another can produce a memorable observation in a moment of inspiration.
Someone who uses AI to explore an idea, challenge his assumptions, or help express a difficult thought may be interested in the subject while having little attachment to the particular business of composing every sentence himself. Whether this is an admirable or impoverished way to write is worth discussing. But we have to describe the activity accurately before we can judge it.
The familiar objection that writing is thinking contains an important truth, but not the whole truth. Writing can be a means of discovering what one thinks, and outsourcing the entire process can deprive someone of that discovery. It doesn’t follow that every mechanical act of composition is itself an intellectual accomplishment, or that useful thinking cannot occur through conversation, revision, selection, and criticism. A writer using AI can evade the labor of thought, certainly. He can also undertake more of it. The instrument doesn’t settle the question.
Petrick’s note ironically exemplifies the very intellectual laziness it purports to condemn. The language has the familiar rhythm of a sentiment designed for immediate recognition and approval. The terms are left conveniently undefined. The distinctions that might complicate the conclusion are omitted. It asks almost nothing of the people applauding it, except that they recognize themselves as members of the discerning “we.”
This is what Valéry’s parrots do. They reproduce expressions whose meanings have ceased to require examination. The words are familiar, their associations are reassuring, and their repetition performs a social function. The fact that someone has personally typed every word provides no assurance that any independent thinking has occurred.
Petrick might have spent five minutes composing his note or five hours. I have no way of knowing, and I don’t particularly care. What matters is whether the thought can withstand examination. If we really wish to avoid wasting our time on writing that required no serious thought, we have a much larger problem than generative AI.