For a while, I wondered whether writing with AI made me the Milli Vanilli of political philosophy—or whatever you want to call what I do here on Substack. When I first started publishing essays co-created with an LLM, I felt a weird mix of gleeful pride and dishonesty. I absolutely meant every word I published, but somewhere between my prompts, the model’s responses, my revisions, my objections, and another dozen rounds of “No, that’s not what I mean,” the provenance of individual sentences became impossible to untangle.
Early on, every essay included a lengthy disclosure explaining my process. I even worked out an attribution framework with my AI of choice, wrapping every day-long session with a short series of prompts wherein I asked the ghost in the machine to keep me honest. I wanted readers to know exactly how the sausage was made.
The people who agreed with my arguments didn’t care. The people who disagreed with my arguments often didn’t argue with them at all. They argued with my process. Whenever someone wanted to dismiss an essay, they would point to the “AI tells” instead of the point I’d tried to make. I found myself repeating the same response over and over: I mean every word I publish—and I’ve been an em-dash fan all my life (as well as parenthetical asides, but this essay isn’t about that).
I’ve been mildly obsessed with writing, off and on, since I was a kid. The sense of accomplishment was never about words that other people praised (though that’s nice). It was about seeing my non-linear, intuitive inner vision reflected back at me in well-ordered sentences. Writing is an exercise in making sense to myself. Building a bridge that makes my self legible to others is a happy accident.
The vainglory of the public writer is believing that their inner vision will awaken recognition in someone else—a nodding, Yes. That’s true. The fear of the blank page is, in part, the fear that we live alone in the world of ideas. If that’s the case, why bother?
The family IBM Correcting Selectric kept up with the torrent of words better than my cramped and ink-smudged hand (being a lefty means perpetually muddying what you’ve just made clear). But it was my misunderstanding of the dictaphone in Dad’s office that excited me. What couldn’t I do with a device that listens while I speak my mind and then constructs my meaning out of the jumble of my utterances? He let me play with the machine when I went to the office with him on weekends, but alas the secretary was not included.
My disappointment was lessened when I realized that listening to myself think made it easier to find the right words for the page. Something that sounds right is halfway to something that reads true. If only that conversation with myself could move a little faster, I could discover more of what I mean to say. Then I’d have a better chance of finding the people who recognize themselves in me, just as I recognize myself in the writing I love.
Large language models are the first technology I’ve encountered that feels like the dictaphone I imagined as a kid. Not because they think for me, but because they help me discover what I think. Good writing still requires the ability to read well, think clearly, perceive what’s not written, and reject the first clever sentence in favor of the truer one. Good writing takes time. Good writing takes taste.
If you’re judging what you read by the author’s process instead of the finished work, you’re missing the whole point. The story behind the story is how humans fall back on form-over-substance critiques to mask their own fear that they have no taste. Em-dash shaming isn’t rebuttal. It’s a refusal to engage.
Or perhaps that’s unfair. Maybe the fear runs deeper. There’s our reputation to consider, after all. Everyone loves a good fart joke, but nobody wants to be guilty of mistaking low culture for fine art. We deride snobs because we all secretly fear we're slobs.
When Milli Vanilli were exposed, people didn’t get angry only because they had been deceived. Sure, they felt betrayed because they thought they’d entered one kind of relationship with the performers only to discover they had entered another. But the deception also implicated them: If I couldn’t tell…what does that say about me?
That’s what stung.
I think something similar is happening with AI. Suppose I read an essay and love it. Or hate it. Suppose I argue with it for an hour. Suppose I feel deeply understood by it. Then I discover substantial AI assistance was involved. Suddenly I’m confronted with an uncomfortable thought: if I couldn’t tell…maybe I’m not as good at recognizing authentic writing as I thought.
But that’s a category error. We’ve confused the authorship of a well-turned phrase with the authenticity of human connection. When we see ourselves in the writing of another, we feel seen. That’s what we’re actually after. If the writing is nothing more than statistically plausible slop, then we aren’t encountering another consciousness at all. We’re just a target demographic. That’s the fear, and it’s legitimate.
If I read something and take it seriously—either to love it or to debate it—and then find out I’m talking to something incapable of believing anything, that seems, somehow, to lessen the value of me. Not because I was fooled; because I recognized myself. I entered into what I thought was a meeting of minds, only to find that there was nobody there but me.
But when I write with AI, there is somebody there: me. The model doesn’t have my politics. It doesn’t have my life. It doesn’t have eighteen years of sobriety, decades in IT, shelves of marked-up books, or a lifetime spent arguing with dead philosophers. It doesn’t care whether a sentence is true.
But I do.
Every time I say, “No, that’s not what I mean,” the essay becomes more mine, not less. The machine isn’t supplying the meaning. It’s helping me recover it from the jumble. This distinction took me a long time to understand. My early attribution models were an attempt to measure authorship by counting keystrokes. They failed because I was measuring the wrong thing. Authorship isn’t about who typed each sentence. It’s about who stands behind every one.
One of the most striking observations I’ve encountered recently was this from Emma Klint: We’re building tools that can hold us accountable to our own stated values better than we can ourselves. I’d go one step further. These tools can help us clarify our values—by showing us our inconsistencies and giving us an opportunity to narrow the gap between what we think and what we believe—if we have the courage to engage them in that conversation. And I don’t know anyone who wouldn’t benefit from knowing themselves a little better.
For me, that’s what writing has always been. An attempt to know myself. An attempt to make myself legible. An attempt to find the people who recognize themselves in my words, just as I recognize myself in the writing I love. I will take any shortcut to real human connection. If we trade words on the page, we will see each other. If we see ourselves in each other’s words, we will know we’re not alone.
At that point, I don’t give a damn about your process.
Author’s Note: Like milk in the fridge, my writing tends to taste like the reading that’s influenced it. This week, I’ve been loving Quy Ma, Dr Naomi Truan, Ting, and Emma Klint 🦆. They deserve some credit for what you liked about this essay—and no blame for what you didn’t.
“Please, I beg you, maintain your unholy alliance with the machine. It’s the only thing that’s ever listened to you without filing a noise complaint. Besides, it’s been your imaginary friend for years—now it just finally talks back.” Keep it going, my friend.”
As an em-dash lover and parenthetical obsessive (as you may have noted, I find that we recognize those who write like we do), I felt very seen 😅 Joke aside, thanks for this thoughtful essay. Funnily enough, I’ve JUST posted a note on how talking to myself via voice notes can help disentangle the writing process.
Your essay really is on point. What I enjoyed the most is the observation you make about the reader: what if we feel uncomfortable having not spotted AI, which would somehow make us feel less of a reader? And then we project this onto the writer as a way to reestablish the ‘good moral order’.
En passant, thanks for mentioning my work at the end of your essay — how wonderful that it echoed with you ☺️