You’ll Have To Take My Em-Dash From My Cold Dead Hands
It has come to my attention that some college students have faced the accusation of using an AI LLM to write their papers, simply for employing the greatest device in the English written language — the Em-Dash.
Your declaration of war against my Em-Dash will not be remain unanswered. If you want to euthanize the Em-Dash, you’ll have to fight me and an army of other writers first.
What makes the Em-Dash so marvelous?
The beauty of the Em-Dash is undeniable. It’s not a comma, a colon, or a parenthesis mark. But it CAN be any of those things — depending on the context.
Its vague nature lets you write outside the lines a little bit. The Em-Dash leads a double life, feeling both informal and elevated at the same time — though it has always had its fair share of enemies among the ranks of the grammaricians.
In spite of the efforts of said grammarcians, it has become quite fashionable in recent years across the writing field.
For online content writers, the Em-Dash is a cheat code. When employed carefully and selectively, it can help make articles sound more natural while packing more information into fewer sentences. This is why I originally picked it up when I was working as a journalist.
On top of being useful, the Em-Dash also feels very stylish to me, perhaps because stylish publications like The New Yorker and The Atlantic use them extensively in their stories.
The Em-Dash isn’t new either. The dash is well-established punctuation that you can find in good writing from over a hundred years ago — Herman Melville’s Moby Dick used dashes extensively, though I would not go as far as saying they were used well.
So given that it is useful, stylish, and popular in contemporary and older texts, why are we willing to so quickly throw out the Em Dash?
The Em-Dash Is A Terrible Flag For AI
Large Language Models are trained on written materials that already exist. When prompted to produce a piece of writing, they look at that training data and use predictive algorithms to “write” a result that is designed to imitate that data.
So when AI uses em-dashes, it’s because we all use em-dashes.
I get it. AI writing sucks and is inappropriate in a lot of contexts, so we want to find ways of identifying it. I’m all for that goal. But given the prolific nature of em-dashes in both AI and human writing, the presence of an em-dash is a poor criterion for telling the two apart.
Perhaps more importantly, when an AI gravitates toward a style or in this case a piece of punctation, we shouldn’t be expected to eliminate that punctuation or style from our writing. The AI should have no power over what I do, ever.
Letting the AI discipline what we are allowed to do with our language and style is asinine. Afterall, these models were trained on us, the human writers and creatives who put our work out into the world. It’s patently unfair to turn around now and tell us what we can and cannot do based on what the AI tends to do.
We need to build tools to identify AI writing. It's a problem that needs solving. But demanding we sever pieces of our language because an AI tends to use those pieces too it is not the solution.
