Inside our search for a way to tell human writing from AI slop

Two theories, one dumpster fire

I guess first we should answer the question of why we care if someone uses ai to write, right?

I think it’s a fair question. And the best comparison I can give is that we care for the same reason people care when athletes use steroids. Except, in this case the steroids are made out of the liquified remains of other athletes¹

Being the only tech company that picks up the phone in the literary space during this AI crisis feels like being the only firefighter at a party after everyone’s been set on fire. Here’s the thing though, we’re just the strippers someone ordered. We dunno what the fuck to do about this fire.

So we’re doing what we always do. We’re talking to some really smart people to help us figure it out. Specifically, two companies with two different approaches. Each represents a theory about how to stop ai in writing².

Theory 1: Use AI-detectors to vet writing

I am skeptical of this approach. If you asked me a few weeks ago, I would have said it was all bullshit. In fact, an earlier version of this piece read:

It’s official. AI detection software is dead. There is only one way to tell if someone used AI to write their work. We all bet on it on Kalshi and let the stupidity of the crowds decide. Fuck it, embrace the fact that we live in the worst possible timeline, bring on President Hegseth³.

But when I reached out to the folks at Team AI-Detection to talk about it, they insisted their software is different. Also, they beef it up by using a combination of manual review + ID verification + ai checkers + legal maneuvering (like making you sign a contract) to try and verify as much as anyone can that something is not ai.

Their pitch is to create a badge, tag, verification t-shirt?, or what have you, that indicates to people that something was created by humans. They are already working with some of the Big 5 publishers to verify books. They want to be the “Certified Organic” of the 21st century.⁴

And, well, have you ever seen one of those prison movies where some guy walking down the hallway gets cornered by a couple other dudes with sharpened toothbrushes and they go all stabby-stabby?

That’s a little how I feel when I talk to them about this process. Like, what about liars? What about pen names? What about people who use AI to write one thing but not another? What about AI that just edits? What about people whose identity doesn’t match their ID? What about poetry? Flash fiction? Children’s books!? What about when AI gets better and better? How do you sue someone in French? Can you scale the manual review element?

Stabby-stabby-stabby-stabby-stabby-stab-stab-stab.⁵

And I feel really bad about it because I feel like I’m shanking someone’s dreams. Also I like them and I believe in the spirit of what they’re trying to do. But like–I dunno. It seems workable when it comes to a book, or a well-known columnist, but I’m not convinced it can handle short fiction, nevermind poetry or flash.

Next steps: they will be giving us an API to test how good their detection software is.

Speaking of, the best AI detector we’ve found so far is Pangram. I also have a meeting set up with them for next week so, maybe news soon.

Theory 2: Track the progress of a writer from the time they start till they finish every draft

The company in question wants to create an algorithmically monitored word-processory-thing that watches you write and is trained to recognize human-like patterns of typing and learns as you go.

I’m cautiously optimistic about this approach with big ol’ grains of salt. Aside from the Big-Brother-Watching-You-Work of it all, everyone else who’s tried to pull something similar off has had a rough go of it.⁶

Take Draftback, for example. It’s a tool that plays back the work you did on a google doc. I thought this was pretty clever until somebody cleverer came up with…

• EXHIBIT A: Undetectable AI - Human Auto Typer

• EXHIBIT B: Duey.ai — Auto Typer & Typing Simulator for Docs, Slides & Word Online

• EXHIBIT C: HumanTyper: Realistic Auto Typing

And so on. 11 ft wall, meet 12 ft ladder. So, anyone who is clever-er-er-er enough to beat this, hit me up. Have I got a party for you.

Next steps: Not sure. Team Track Your Progress also embedded an optional AI writing assistant in their tool which threw me. I told them I didn’t want to work with them unless they found a way to square that giraffe for me. I haven’t heard back.

So those are the main two folks we’re talking to. And I think they are representative of most others we’ve spoken to focusing on this.

Whatever happens, I think…now, let me get a little idealistic and wishy-washy, but I think everyone should sit for a minute and make sure we are not trying to build a torch-and-pitchfork tool, but are instead trying to build a torch-and-pitchfork deterrent tool.

What I mean by this is that we should be building something for writers and publishers to prove they did not use ai before accusations start, in situations where it matters, rather than make a tool to help us go around playing games of “Hey, fuck that guy!”

For those of you who don’t know the rules of the game “Hey, Fuck that guy!” It’s pretty simple.

1. Person on the internet sees someone doing something they don’t like and writes a piece titled, “Hey, fuck that guy!”

2. The rest of the internet follows this up with several dozen insightful articles titled “Yeah! Fuck that guy!”

And I know some people get off on being that guy, and no kink shaming here. But a gentle mindset-shift would be nice. For example, I don’t think anyone cares if grandpa used ai to help him write a steam-punk romance western to read for his love interest (Betty) on open mic at the retirement home. Roid up, grandpa. But professionals looking to enter a contest or get published in a magazine should go in assuming there will be a checks and balances and consequences — and publishers have a right to expect that.

So why not a, “Hey, let’s build a system where those types of questions don’t need to be asked because we’ve normalized the idea that human creativity is preciously human and go into situations with a little grace and faith in one another knowing that we are all in this together but keep your ai-slop out of our house.”...game.

Or some shit.

Anyway, that’s what we’re doing about it. We’re taking it slow. We’re talking to people with whole companies focused on the problem. And we’re dancing for strangers while it all burns.

1. …and make muscles so big they go out and do the sports while athletes sit home doing something they’ve always wanted to do with their lives. Write poetry, for example.

2. after torture and stalking

3. I love that joke about President Hegseth. Every time I make it, my little brother yells at me.

4. If we partner would that make us Whole Foods?

5. In retrospect, I probably could have used a cuter metaphor like of a 3 year old asking about where babies come from or something but where’s the fun in that?

6. maybe some company could hire Gen Z to go stand behind people while they work after ai has taken all of the entry level jobs.