Never underestimate our ability to normalize the extraordinary.
Less than two years ago, I witnessed the rapid adoption of ChatGPT in trivially outputting prose: translating messages, prettifying emails, and, of course, generating high school and college essays of arbitrary length. At the same time, there were telltale signs of AI assistance in writing, but the tech was novel enough that most people reading AI-generated text weren’t sufficiently exposed to its style. Since then, LLMs have enjoyed a couple of generations of improvements, but users—on both sides—have gotten more sophisticated in using and detecting the use of ChatGPT and its ilk.
The natural result of this mutual escalation is that readers are now naturally suspicious of any text, particularly when it uses words that are known to be preferred by ChatGPT, such as the likes of “delve” and generous use of emdashes1. Furthermore, a fascinating second-order effect cited in that article suggests that the prevalence of AI-generated writing is, in turn, affecting human speech! It’s a true symbiosis between human and machine, with the language generated by each entity exerting some influence on the other.
There is a pattern that emerges from this kind of technological advance. First, the tech is rudimentary and toy-like, almost embarrassing in how badly it’s trying to emulate its source material—a bit like the Will Smith spaghetti test of yore. Over time, the tech improves, and occasionally the advancement goes beyond raw human capability, like making copious use of Photoshop to sculpt bodies to idealized proportions. Rendered and realized perfection is recognized as machine output, so the next iterations painstakingly reintroduce flaws. To better imitate humanity.
Underneath this evolution is the desire for authenticity, where it’s ultimately more valuable to know that another person is behind the work, warts and all. It’s the proverbial callouses that layer on some amount of sincerity, and in a roundabout way, differentiate from polished but bland conformity. For fun, I took the quick notes that I jotted down before I wrote out this post, and ran it with a few prompts through Claude, asking it to mimic my writing style to summarize the point I’m trying to make:
We’re witnessing a fascinating feedback loop: AI trained on human writing now shapes how we communicate, measurably altering word usage patterns beyond major global events. As we unconsciously adopt AI’s distinctive vocabulary and structures, something curious emerges—growing fatigue with formulaic AI text and renewed appreciation for authentic, imperfect expression. Democratized “correct” writing has paradoxically made genuine voice precious.
Claude did a pretty good job! Grammatically flawless, varied sentence structure with colorful vocabulary to invoke layers of analogies to spur additional thinking.
It just lacks a distinct, authentic voice.
I’m also a fan of the punctuation, but it’s been in my writing for years before AI co-opted it for its style.↩
