We Know You’re Using AI

Let me save you the embarrassment.

Rembrandt's Belshazzar's Feast depicts the origin of the phrase "the writing on the wall." The words “Mene, mene, tekel, upharsin." are not in future tense. They are already doomed and all that’s left is there recognition.

To give you an idea of how embarrassing it is to try to pass off AI writing as your own: I won’t even tell you which family member clearly used GPT to write a congratulations for our wedding. I feel it would be disrespectful, even though you don’t know this person.

I would rather they texted me a picture of a dead dove captioned “thinking of you two” than the slop we actually did receive, “not X, but Y” maudlin sort of prose. It’s especially awkward because I can’t even tell the person that I know because it would embarrass them. I just have to grimace and hope for better in the future.

We know you’re using AI and nobody will tell you because it would embarrass you. We don’t necessarily want to do that. Except sometimes when someone is really asking for it.

Robert M. Hamburger (you can’t make that up) wrote an open letter to Substack leadership complaining about, I wanna say, promulgating short-form media as opposed to more thoughtful long-form essays. I don’t know, I didn’t read it carefully. Detection software tells me his letter is top-to-bottom AI-generated. How this might count as irony is apparently lost on Hamburger. I don’t even need the software to know. I can smell it. It’s filled with telltale AI-isms, sure. But I don’t need those either. Even if he had used it “just as an editor,” I can feel it like plaque on my teeth. When he was called out for the AI by Chris Best, he doubled down with what read like the prose of an AI-generated presidential concession speech. The humiliation was almost too intense to watch.

By the way, I am not at all moved by those quizzes that suggest people are no better than chance at guessing what is AI text and what is not. The same sort of people who think that GPT is conscious are the ones that set up these studies, apparently. First of all, I can still tell. Second, a piece of random text presented in a flashcard setting might as well be AI-generated, anyway. I don’t care if you show me a line of Tolstoy and an AI generated plausible faux version and one day I can’t tell the difference. The overlooked important bit is that Tolstoy actually existed and we all care about his existence enough to make the machine sort of convincingly imitate him, and that is only if it is taken out of any plausibly natural context and entirely in hindsight. What is unique about actual human writing is that it comes from within lots of context. We know what bits Tolstoy actually wrote because we have agreed-upon records, and that is what actually makes any of it interesting at all.

If, for example, you tend to inarticulately bloviate like a Cockney, and then suddenly start texting like a museum plaque right when things start to get intense, I’m going to know that what you’ve presented to me is, if not AI-generated, at the very least bizarrely inauthentic. In that sense, I can understand why Hamburger would be motivated to post his slop. He wants to make sure his ideas are palatable and legible. And it worked. It has almost 5,000 likes and 2,000 comments or something like that. Most people don’t notice that it’s AI-written.

The cynic in me wants to say that, therefore, people en masse are morons. I don’t see it that way. People en masse are busy and AI is particularly excellent at pressing mass-appeal buttons to spiritual passersby, extremely legibly, without upsetting or disrupting their common sensibilities. Somebody not reading thoroughly is much more likely to skim it, share it, like it. Thus, these things spread like a three-day cold. In the short term, it’s exponentially more popular than real writing.

People look at those numbers going up to the right and project that slop is therefore soon to take over everything. This is just the classic mistake of a too-short time horizon. As a category, it tends to bloat in the short term and then quickly rot from public memory. It has a negative capacity at getting the approval of the right people: the people who you love, who have things you need, who may employ you, etc.

You burn bridges, in fact, when you indulge in not thinking. If you commit to this strategy, you’ll end up frenetically chasing your own tail, trying to spike your slowly declining popularity with people who, by definition, go away as they pay more attention to you. You enter into a cycle of self-humiliation, finding more and more ways to justify that to yourself and others. People who get caught in this loop accidentally seal themselves into a self-contained yet declining ecosystem, which will be their creative end. It’s not unique to AI. You see it all the time. I knew viral listicle writers who were very popular on Medium. They were naturally taken with AI and produced a bunch of slop, and then humiliated out of existence. I don’t believe real writing is in any danger, the arc of justice is just longish.

I mostly worry about my friends and family suddenly having access to this thing. It’s so embarrassing and I just don’t know how to tell them. Here: to everyone using AI in your texts, your Facebook posts, and your job applications, stop it. We know. We don’t like it. The short-term thrill of sounding more crisply articulate than you are will hurt you, not even in the very long term. It makes it too easy to not do the vulnerable thing of revealing what sort of person you are through the prism of what you would plausibly know about off hand. People try to make the following argument: we’ve always had editors and thesaurus and so on. How is AI any different? They haven’t thought it through: that’s the classic joke of someone writing inauthentically with a bunch of pompous similes: “writing with the thesaurus.” Imagine that times however many datacenters.

My wife and I met the superintendent of the public school at karaoke the other night. She told us how easy it is for her to tell when parents email her using AI. They always start with, “I hope you are well.” Of course, the AI style is also obvious and becomes even more agonizing once they meet in real life and they’re nowhere near that erudite. Unlike what I predicted of AI early on, it has not bothered to adapt itself to fool the most scrupulous. It has adapted to the greater market pressure to be like a semi-useless degenerate Tolstoyan aristocrat. Everyone knows the kids are all cheating with AI. The only reason they get away with it is because the teachers get to use AI to grade the AI homework. Then, parents complain about how uneducated their kids are with AI. For now at least, we have allowed a total mutual waste of time at the expense of the taxpayer. It’s doomed because everybody knows it’s happening. The schools that ban the phones and go back to reading and writing, oral exams, handwritten essays, etc. are going to outperform, embarrassingly, the ones who don’t. Just give it time.

Despite what the most flagrant tech bros tend to think, you can’t become capable of thinking if a machine does your thinking. You may think you’re “high-agency” enough to skip Moby-Dick and cheat your way through school and drop out and do a startup. That might have happened a few times as a cash-out of generations of accrued societal and familial wisdom, but a society that makes a habit out of that behavior becomes a dead society. One that can’t ensoul children to be able to think anything at all. It doesn’t matter if you avoid the em dashes and the “I hope you are wells,” and the “not X, but Ys” and if you always keep up with the new ones. What is always happening behind the scenes is that you’re stealing wisdom that you haven’t earned. Again, I’m not moralizing at you. I’m just telling you that we can tell. You’re trying to get away with something. You’re not going to get away with it forever. So, don’t get too used to using it for your job applications, your sermons, your cover letters, your texts. Especially if it’s difficult, high stakes, or emotional. Don’t do it. You will eventually be embarrassed and no one will even be able to tell you. They will just stop reaching out.

I’m not above it, either. I’m tempted to run these drafts through AI to make sure they’re tight and that I haven’t forgotten anything. But it’s better to miss something. It’s better to show the pattern of your ignorance as it truly appears than to buffer what you already think through a sycophantic machine that’s going to find every piece of evidence to agree with you, and therefore never allow you to have a difficult experience of transformation.

AI is a tool, yes. “Morally neutral” as people like to say. Sure. But its truly “morally neutral” applications are extremely limited. It’s a good spell checker. It’s good for transcriptions. If you’re a lawyer, it’s good for summarizing documents, but you still have to check them. It’s good for cooking and translations and phone plans and things I’m forgetting and don’t know about, I’m sure. Somewhere just a hair beyond, it’s suddenly a complete deal with the devil. The moment the thing spits out the text, you have your reward.

The people who matter can tell and it’s embarrassing.