Stop Using AI to Write

How René Descartes and the man from Nantucket can defeat the evil robots

Welcome to the weekend edition of The Experiment, your official hopepunk newsletter. I took the last two weekends off because of travel schedules, and while I don’t think I have a free weekend this summer until August, I’ll get better about getting things done. The essays are piling up in my head and I need to get them out. If you’d like to support my work, become a paid subscriber or check out the options below. But even if you don’t, this bugga free. Thanks for reading!

---

All of a sudden, we’ve been overrun by ink-stained robots. Half of all new English-language books on Amazon last year were written by AI. Litigants representing themselves filed 6% more lawsuits in 2025, many of their filings rife with fake citations invented by robots, and the problem is even worse with submissions to scientific journals. Almost half of all new music uploaded to the streaming service Deezer over the same period were fully AI-generated, and as much as a third of all new online content every month is AI-generated.

The problem with being overwhelmed by a glut of writing is not new. Like it says in Ecclesiastes, “Of making many books there is no end,” but as Jason Koebler wrote more recently in 404, “To browse the internet today, to consume any sort of content at all, is to be bombarded with AI of all sorts.”

…increasingly I have the feeling that it’s everywhere, coming from all directions, completely unavoidable. It’s not exactly that I have a revulsion to AI-assisted content or don’t want to get fooled by it. It’s that something is happening where my brain has become the AI police because everything feels incredibly uncanny.

To read anything now is to get lost in the uncanny valley where people save time by asking robots to produce shiny nothingness at their behest. Koebler, who describes the effect of AI-generated text as “overly generic,” pointed to a discussion on Reddit in which both the poster and the respondent used AI to generate what they call thoughts. Their topic? Injuries on the Baltimore Orioles. (Though to be fair, if there’s any subject right now that I would happily trade my human heart for a robot’s lack of feelings, it’s the Baltimore Orioles.)

One thing that bugs me about writing is people who misunderstand what writing is. A decade ago, a colleague told me she was going to write something and then send it to me to “wordsmith,” she said, miming typing with her hands, as if writing were a parlor trick to zhuzh-up text like touching up one’s lipstick before heading out. To others writing is a task at the end of a chain after the scientific method: observe, ask a question, make a hypothesis, do an experiment, analyze data, and draw a conclusion. Oh yeah, and then publish. If I thought this way, I might ask AI to write for me as well.

The temptation to use AI even among professional writers is so great that the nut graph that contextualizes the problem in every story about journalists misusing AI keeps getting longer, so long in fact that it’s now two paragraphs:

The reminder [to New York Times freelancers not to use AI] comes as the paper of record continues to grapple with AI-generated content, including preventable AI-spun errors, making its way into its pages. Back in March, the NYT faced scrutiny after a contributor to its competitive “Modern Love” column was publicly accused of using AI to generate an emotional personal essay; that writer later told Futurism that she’d used chatbots to conceptualize and edit the piece. Then, in April, the paper cut ties with a freelancer who admitted to using AI to cook up a book review that was found to be riddled with plagiarism after its publication.

And while these controversies indeed stemmed from the work of freelancers, the institution found itself in hot water yet again last week, when a substantial correction revealed that an article bylined by the NYT’s Canada Bureau chief contained an AI-fabricated quote weeks after publication.

There are plenty of good uses of AI for writers, including suggesting copy edits, advising on headlines, and finding contact information for sources and experts, but not generating copy, which is to say writing, because asking a robot to write—even if you direct it to write it in your own voice based on everything you’ve ever written—betrays a fundamental misunderstanding of what writing is and, in doing so, betrays what it means to be human.

A couple years ago, I was sitting in a conference room with almost as many doctorates as chairs while experts talked about the Science of Reading, the latest pedagogical theory of how to teach kids to read. (These theories come and go, and the good news is that this one is working.) As they bandied grad-school terms about, I understood: Reading is decoding. The letters and words and punctuation have been arranged in such a way that if you know the code, then you understand the meaning.

And then I understood that writing is encoding, the arrangement of letters and words and punctuation in such a way as to pass a message to the reader. This code can be and in fact must for the most part be replicated according to the same rules so that any reader possessing the code can decode it. AI agents know this code; AI agents are perfectly capable of replicating this code; but can AI agent encode?

When I, Emily Dickenson, or the limericist from Nantucket encode, we don’t just follow the rules. We’re playing variations on them, stacking rhythms, embedding callbacks, and hiding jokes in the prose for you to trip over. It’s what’s called voice, the trace material of a particular thought process. Paul Burka, the late Texas Monthly editor, once told me, “I write to find out what I think.” Put another way, writing is a way of thinking, and voice is the fingerprint on the smoking gun left at the scene.

AI can’t think; AI can only pretend to think; and yet: So many people are using AI to think for them. What ground are we ceding to the robots when we ask it to think not with but for us in our stead? “For as he thinks in his heart, so is he,” says Proverbs 23:7. The baseline temperature of Western philosophy is René Descartes’ Cogito, ergo, sum: I think, therefore I am. The act of thinking proves you exist.

A recent piece in Scientific American argued that every innovation in media from Johannes Gutenberg’s moveable-type printing press to the Grub Street boomlet in 1700s London to nickelodeons to B-movies produced its fair share of crap while also giving us Shakespeare, Jonathan Swift, and Francis Ford Coppola. But at least that crap was human. Even the most staged, hackneyed “I don’t know how I can pay for this pizza” pornography is recognizable as part of the human experience, if not one of its more laudable ones.

Thinking proves we are humans, and if writing is thinking, then asking ChatGPT to write for you is to hand over the keys of humanity to a drunk driver who cannot imagine the pain of a broken bone or a call at 3am. Asking AI to write for you is asking to be replaced by robots.

The good news is that all that’s required to not be replaced by AI is to do the relatively minimal work of being a human: Think, and therefore exist.

Jason Stanford is a co-author of the NYT-best selling Forget the Alamo: The Rise and Fall of an American Myth. His bylines have appeared in the Washington Post, Time, and Texas Monthly, among others. Email him at jason31170@gmail.com.

---

Further Reading

---

Buy the book Texas Lt. Gov. Dan Patrick banned from the Bullock Texas History Museum: Forget the Alamo: The Rise and Fall of the American Myth, cowritten by yours truly.

We set up a merch table in the back where you can get T-shirts, coffee mugs, and even tote bags now. Show the world that you’re part of The Experiment.