Them robots - they took mah em dash

I like to use em dashes in my writing where they work and it absolutely is the right device, for me, at least some of the times I’ve reached for it. What I’ve noticed lately is that I second-guess and wonder if people are going to think the thing I wrote with my own hands is just more AI slop. Having that in the back of my head both slows me down and makes what I write feel homogenous. Maybe I become the human version of a humanizer skill. Or worse, all the tells that say “this is what human writing looks like” converge both AI and human writing into a vapid new language — the new corporatese, passive voice, indistinguishable content bloat that grows far more quickly than knowledge accumulated over the millennia which gets all but completely drowned out.

I noticed this before the meat proxy thing (of which I’ve certainly been guilty, fwiw), but at least some people have been trying to avoid sounding like AI even when they’re clearly writing themselves. What they do, though, is not stick to their human writing habits but rather they suppress the obvious tells. We went from chuckling at “that’s the load-bearing finding that flips my earlier recommendation — it’s not a bug. it’s a feature” to replacing every em dash with a colon, avoiding rules of three, punchy sentences, and being concise. Not that those are signs of good writing, but strict pattern matching like that is probably making our own writing worse.

Beyond norms, I imagine students who write with their own hands, if they exist, also must pass their writing through AI detectors because their teachers surely will. The consequences of a false positive are so high, better to ask the AI to rewrite your text for you until it passes than submit what you actually wrote. I figure that’s a variation of the same homogeneity dynamic, to wit, everything that’s bland is beyond reproach.

What’s funny is that even the AIs know their tells. When I’ve run the humanizer skill on a big AI-generated analysis, it just says “yup, I took out all the em dashes, boss: there were 237. now they’re all something else”.

As I think about these dynamics, what worries me more is that our own writing is ruined in persistent ways. In Slack (which I think is the destroyer of companies, more on that someday) threads with colleagues, I’m finding that some sound like Claude-of-today. I would be stunned to find out that they are actually piping every chat message through AI, but they definitely sound like AI. I think we pick things up from our interactions and absorb style from what we read. I suspect that AI writing will get so capable and varied that it will be impossible to tell — it might not be universally good, but it will get undetectable, such that the only human tell will be mistakes. Meanwhile, some people have been tripping my brain’s AI detectors these days and while the AI writing badness is temporary, the human habits that form from trying to avoid it will be lasting. The tropes will be lasting. My em dash — gone forever, y’all.

I suspect there is no solution. My personal attempt at my personal solution is to keep reading books born before 2022. I’m mostly formed and haven’t had the majority of my lifetime shaped by AI writing. What happens to the rest of the world? Artisanal writings somehow flagged with their fair trade, organic, non-GMO stamp? Maybe we stop caring or reading altogether, for that matter. For now, at least, remember the tells are temporary and the current slop sense we have only hurts our own writing. Don’t just forward garbage, sure, but do take back your em dash if you want it and remember that “did what I just wrote sound like AI?” is futile noise in your wet brain.