Gaslighting Ourselves

I was an early adopter of Twitter, back before the word tweet had been applied to the messages we sent via SMS; before @mentions were a thing (and long before Twitter adopted that convention); before #hashtags, even. I convinced my wife and numerous friends and many readers of my columns to join, and Amy and I attended and even organized local “tweetups.” Aaron Wolf and I created multiple Twitter accounts inspired by ancient Roman emperors (some real, some not) and used them to comment in a veiled and often hilarious way (at least on Aaron’s part) on our daily grind in the editorial offices of Chronicles. I drove subscribers to my About.com Catholicism newsletter and pageviews to the site.

Twitter was fun, exciting, and useful. Until it wasn’t.

I was an early deserter from Twitter, long before Elon Musk took it over; before the bots overran the platform; before Donald Trump rode his Twitter account to the White House, and long before he was banned from the platform for encouraging the events of January 6, 2021. I left for reasons, though, that would look familiar later on. I was tired of spending all of my time blocking and muting scores of Alt Right trolls (at that time, often posing as traditionalist Catholics) who were trying to get me to engage with them.

In most cases, they weren’t overtly hostile to me — far from it. Instead, they pursued a strategy designed to “move the Overton window” — that is, to push the boundaries of what is considered acceptable to say in public. They would reply with an interpretation of a tweet of mine that was clearly wrong but was phrased in such a way that, if I attempted to clarify what I meant, I might be drawn into saying something that they could use. I watched other prominent paleoconservatives fall for the ruse, thinking that they had to move beyond their original position in order to defend themselves, only to find that their Alt Right interlocutors were now claiming that So-and-So had been “red-pilled.”

The inevitable attacks would then come from folks on the left, and sometimes (though not always) the target’s knee would jerk, and he’d move even further in the direction that his Alt Right interlocutors had been pushing him.

It was a clever strategy — a form of gaslighting, convincing the target to think that he believed something he had not initially believed — and like many such clever strategies, it relied on our fallen nature, our pride, our desire to defend ourselves from misrepresentation. But just as there is no good answer to the question, “When did you stop beating your wife?” the only way to respond to such trolls was not to engage with them at all.

Thus my daily game of Whac-a-Mole, blocking and muting accounts just to have others pop up in their place. Twitter was no longer fun, and I signed off for several years.

The subject of this essay, however, is not Twitter but AI, and specifically the constant flood of social media posts urging authors to change the way that they write so that readers will not think that their articles were written by an LLM. Spend a minute or two on any social media platform, and you will spot them: “Here are 10 patterns found in AI-generated text. Stop using them!”

Most of those patterns are, in fact, common in the output of LLMs, for a simple reason: They are common in the writing of the human authors on whom those LLMs were trained. That em-dashes are more frequently found today in AI-generated text than they are in contemporary writing is a sign that tastes in punctuation and usage have changed — but the body of writing on which LLMs were trained contains less contemporary writing than it does older styles. LLMs almost invariably use Oxford commas, too, because the barbaric practice of trying to save a drop of ink at the expense of reading comprehension is a relatively recent phenomenon compared with the entire corpus of published works in English.

Why do LLMs frequently generate tripartite lists of examples? Because Western philosophy has always seen significance in the number three, from the triangles of the Pythagoreans to the dialectic of Hegel, and Christian theology agrees. Whether three examples feels more natural to us than two or four because of something inherent in human nature or because of that philosophical and theological history is a question beyond the scope of this essay, but it is an undeniable reality that suffuses our culture, our literature, and our discourse.^[1]

What does any of this have to do with my experience on Twitter? I spent so much time blocking trolls and ultimately left the platform for several years because their relentless attempts to steer my thinking were taking a toll. To the extent that you allow such gaslighting to affect what you write (and therefore what you think), you surrender your intellect (and even your soul) to those who are trying to manipulate you. When we place artificial constraints on our thought — I can’t let other people think that I believe what he says I do; I need to make sure no one can misinterpret what I wrote in the way that he did — those constraints change us.

Every moment you spend worrying about whether readers will mistakenly think that ChatGPT or Claude wrote your article; every time you delete an em-dash or an Oxford comma or wonder if you should provide a fourth example in a list just to “break the AI mold”; every attempt you make to defend yourself against some commenter who ran your entirely human-written article through an LLM detector that claimed it was “78 percent AI-generated” — all amount to gaslighting yourself. Concerned that others may think that your writing is AI-generated, you inadvertently let AI colonize your mind, changing the way that you write and the way that you think.

There are no self-made writers. Every good writer is a voracious reader; every great writer reads widely and deeply. Most of the social media posts that urge you to change the way you write in order to avoid being mistaken for an LLM are written by people who have neither read widely or deeply nor written anything more substantial than a tweet.

Pay no attention to their lists, and don’t change the way that you write (at least not for the reasons they offer). Marvel instead that LLMs, by ingesting far more writing than any of us could ever hope to read, have learned to imitate your writing in form — even if they will never be able to imitate it in substance.