I’ve been writing about AI on a Substack, but like everything else digital I don’t trust that it won’t one day get slurped away in a corporate shuffle, so… reposting here. Check out my Substack called of poets & processors and subscribe there!
What is that mysterious double-dash and why isn’t it a red flag for inhuman writing?
Maybe you’ve heard this one before: as I write these words there’s a post actively circulating written by some guy who can tell you “one simple trick” for spotting generative AI content online.
“Look for the emdash!” he writes. “It’s a dead giveaway.”
Thus we come to the problem of the emdash.
Oh, you know what an emdash is, right? Oh, sure, you know—but that guy reading over your shoulder doesn’t so I’ll explain here so that he can keep following along.
Simply, an emdash is just punctuation.
We use all sorts of punctuation in English writing and the kinds of punctuation one uses is often a matter of the form and formality of said writing. There are punctuations that get used to mark the end of sentences, say, most commonly periods, exclamation marks, and question marks. There are “quotation marks” both double and ‘single’ that call out words or phrases as a kind of contextual clue that these are someone else’s thoughts, words, ideas, or have broader meaning beyond the text one is reading. And then there are all sorts of helper punctuations that get used to help simulate the cadence of speech patterns like pauses and passes towards new ideas. These include commas, parentheses, colons, semi-colons, ellipses, and—you guessed it—emdashes.
Emdashes are probably the least well known, and infrequently used of the bunch, and basically are just a double-dash. A single dash might tie a pair of words together, where an emdash might tie a pair of concepts or sentence-fragments together, and are often employed (at least as I have found) as a more informal version of the semi-colon, a hint to the reader that a conversational tone is implied as one reads and very much used like a pregnant pause or a “get ready for this” beat in the reading.
And I’ll tell you what else: speaking as a writer myself, they are fun to use stylistically once one starts to think in that vibe and to think of someone reading the words on the page better matching the cadence in the writer’s head. Also, with modern variable-width fonts pretty much standard now, they make even more sense than even a decade or two ago with their less-relevant strict type-setting rules. In other words, people are using them a lot more these days, particularly for casual writing.
One guess what is busy slurping up a lot of modern, casual writing these days and using it to emulate human conversational writing styles.
Yeah—AI.
So. Here we are at the emdash problem: when an increasing majority of content is rapid-generated by AI engines, and those engines are emulating the most modern of casual writing that they have pilfered and scraped from public websites, it was then almost inevitable that (a) those generated AI texts were going to use punctuation trends that are common in text that was written in the last decade and (b) any human reader who is used to more formal writing would immediately misidentify this less-common, human-mimicked punctuation as a red flag gotcha for generative text.
Yet, it will never be so easy. Don’t let down your guard.
I can tell you this with confidence because almost everything I’ve written (at least, written casually) in the last couple of years has made frequent use of the emdash as a stylistic choice. I like the emdash. I use the emdash. And your objection to my use of the emdash is no more valid than telling an artist they use too much blue paint or a musician that their choice of chord progression is wrong: these are stylistic choices and—fuck off, I’ll write how I want to write.
People—human people—use the emdash and it is not a dead giveaway for anything, not even AI.
Like everything else we see online, we need to be a little suspect and cautious: we now have the job to use our brains to unravel the source of authorship, and yeah—guess what!—there is no easy quick trick to deducing origin anymore. It’s a toss up if a human or an algorithm wrote it.
The author of this particular viral meme accusation against all emdash-containing text is not entirely wrong. I mean, kinda mostly wrong, but not completely wrong. There will almost certainly be a trend towards greater use (and misuse) of emdashes in generative text for a while and for the very reasons I wrote above… but emdashes are no smoking gun nor flawless clue. They are but a single part of a complex profile of the origins of the modern written word, a profile that will get more complex as each day passes and more algorithmically generated content floods our feeds. We need to use our very human brains to detect these things and always be skeptical of sources and authors, but this means doing research to understand those sources by seeking to find profiles and consistent histories of the real people writing things, testing ideas against multiple perspectives, and shining sunlight on simple and stupid solutions to the complex problems we will face in the challenging of our own humanity.
AIs didn’t invent the emdash, and insisting they did is an insult to the thousands of real humans who have adopted this as part of a stylistic toolkit and are trying to write interesting things in what is already an uphill battle against the processors.