We're All Scared, Son

---

I hit publish on the Hugging Face Breach post last week and a little panel popped up before it would let me send it. Pangram, the AI-detection outfit Substack partners with, had scanned the text. 55% AI. 10% AI-assisted. 35% human. "Partially AI-assisted text," it told me, "written with some AI assistance."

I wrote that post the way I write all of them--me at the keyboard, an LLM in the room, drafting and rewriting and arguing about word choice until it sounds like me. That's the whole stated premise of this publication. It is also, apparently, a post that a purpose-built detector looked at and called majority machine.

I don't hide my AI use and I don't mind Substack or Pangram doing all of this, but I do wonder at what people think they're achieving here. A regular part of my 'writing' method is to ask the AI to apply my own style guide. In 2022 I made fun of GPT for being a useless waste of time, and here we are a scant four years later and the AI I saw back then is so much different, so much more capable, present, thoughtful, etc, that I wonder how anyone can possibly think that an AI detector is useful over any reasonable timeframe. We can make AI sound like any given person given a few minutes of voice recording, what makes them think we can't do it for text?

The Discourse

Spend any time online lately and you'll run into some version of the same complaint. "AI slop." "AI is coming for people's jobs." "AI datacenters are going to pave over the world." The paperclip references regarding misalignment are never far behind. It has the shape of a moral panic and the energy of a religious argument, it's everywhere.

It reminds me of the US gun debate. "AI is coming for your job" is the same sentence structure as "guns kill people." The rejoinder is obvious once you say it out loud: "people, using AI, come for jobs," just like "people, with guns, kill people." The tool doesn't want anything.¹ It doesn't have a plan. Somebody decides to run the layoff, and AI is the excuse on the memo. I'm not trying to relitigate gun control here, but I'm pointing at the argument's skeleton, because it's the same skeleton.

Bans Have a Bad Track Record

The instinct under all of this is: stop it. Slow it down. Legislate it, shame it, wall it off.

It's an old instinct and it has an old track record, and the track record is not good. Medieval Christendom banned usury outright--interest on a loan was a sin, full stop. So Christian borrowers who needed capital went to Jewish lenders, who weren't bound by the Church's rule, and the actual economic activity the ban was supposed to prevent kept happening, just routed around the prohibition instead of through it. The ban didn't stop lending. It just decided who was allowed to do it, and church officials--bishops, monasteries, allegedly popes themselves--kept borrowing from the very lenders the rule existed to shut out.

There's a similar story, less rigorously documented but too fitting to leave out, about a pope banning meat on Fridays--allegedly, at least in part, to prop up a fishing industry he allegedly had a stake in. I'm not sure about the historical accuracy of that claim, but whether or not the fish story is true in its particulars, the pattern it illustrates is real: prohibitions dressed as principle have a way of also serving whoever's holding the other end of the rope.

I've made the fuller version of this argument twice already. Weights and Measures is about AI specifically: the capable open-weight models are already sitting on hard drives all over the world, already downloaded, already impossible to recall. You cannot ban a checkpoint that's already on ten thousand machines. The Technology That Will Make Us Forget Again is the longer historical version--printing, movies, television, Wikipedia, every panic that followed the same arc and every ban that failed the same way. I'm not going to rehash those arguments here--go read them if you want the full case. What I want to get at today is a different question: if banning it doesn't work, and everyone arguing for the ban half-knows that, why do they keep arguing for it anyway?

What's Actually Being Fought

Most of what gets called "AI slop" isn't really a complaint about output quality. It's two much older complaints wearing a new coat: I don't like watching people be lazy, and I'm afraid a machine is going to take my job. Neither one is new. The laziness complaint just has a new edge now because the laziness produces something that has the thin veneer of being a finished product. The job complaint has a name--Luddite--and it's usually told as the story of people who were simply wrong about machines, and left there. They weren't wrong--their jobs did, in fact, go away--and being afraid of that isn't stupid. It's just aimed, historically, at the machine instead of at the people who decided how the gains from the machine would be divided.

You can eat better, travel further, get better medical care, and live longer than kings and emperors did a hundred years ago, and still be furious--furious at companies that have grown more powerful than nation-states, run by men you'll never meet and can't vote out; furious at governments that feel too broken or too captured to do anything about either. The system that's failing you is also the system keeping you alive. Both things are true, and being angry about the first without being able to hold the second is exhausting, and AI is standing right there when the anger needs somewhere to land.

We're All Scared, Son

The very first thing I wrote here argued that AI was a mutualistic partner, not a tool at all. But I'd add: fire, iron-smelting, the physics that eventually gave us the atom bomb--these weren't just tools either, and every one of them enriched us and could kill us, sometimes in the same decade. Calling a thing a partner instead of a tool doesn't make that tension go away. It just changes who you're asking to hold it responsibly.

There's a scene in Ernest Goes to Camp--not exactly prestige cinema, but it's stuck with me--where the camp counselors are about to catapult a bunch of snapping turtles at a construction crew. One turtle turns to the other, right before launch, and says: "I'm scared, Sarge." And Sarge says back: "We're all scared, son."

That's the reality of "AI slop." Not a quality complaint. Not really even an argument. Just: I'm scared. Scared of losing work, scared of companies I can't hold accountable, scared of a government that can't or won't protect me, scared that the thing keeping the lights on doesn't actually have my back. Scared of a future that is opaque and uncertain.

The fear is not unreasonable and I'm not saying to anyone to "just get over it." I'm writing it because "AI slop" is a bad container for that fear--it aims a real feeling at the one part of the situation that was never going to yield to complaint, banning, or a browser extension. The usury ban didn't stop lending. It just picked who got to lend. Banning AI, or shaming everyone who uses it, won't stop the thing people are actually afraid of either. It'll just decide who gets to use it in the open and who does it quietly, Pangram score or not.

We're just scared. Say that instead. History is full of people being scared and figuring out a way through the fear and on into new adventures that novel technology opened up. Let's do that.

---

Related: Anthropotechnic Mutualism, The Technology That Will Make Us Forget Again, Weights and Measures.

---

James got told by a machine that his own writing was 55% machine. Sounds about right.