The Human Side of Technology
No One Wants to Read Your AI Slop: Humans, Am I Right?
It was bound to happen sooner or later.
In my fan mail this week I received several similar notes. I've compiled them as one message here. You'll get the gist.
Jay,
You are alienating people with your posts. No one wants to read your AI slop.
We can tell that the bulk of the work in your articles isn't written by you. Your "AI collaborators" are stealing your voice, and you're letting them.
You should have more confidence in your writing if you have something to say. We want to hear your voice, not the voice of some machine or LLM.
Warm regards,
Angry Man Yells at AI User
Okay.
First, nobody hurt my feelings.
Nobody called me names. Nobody insulted my family. Nobody questioned whether I should be allowed to own a computer.
They said they didn't like something I produced and weren't going to read any more of it.
Fair enough.
Naturally, I went to see who these folks were. Several of them speak pretty openly about how they use AI to accomplish their own goals.
Alanis Morissette would be happy.
But that isn't really the interesting part.
The interesting part was this:
What if they're right?
So, we went looking for my voice
I brought the criticism to one of my AI collaborators.
Not because I needed reassurance. I didn't want to hear, "Don't listen to them, Jay. You're wonderful."
That's useless.
I wanted to pull at the thread.
Was AI changing my writing?
Was it smoothing things I wouldn't have smoothed?
Were there places where the prose was technically good, or even very good, but I had disappeared from it?
And, maybe more importantly:
How do I actually write?
I've been writing professionally in one form or another for decades. Technical documentation. Books. Articles. Blog posts. Engineering standards. Stories. Things intended to teach. Things intended to persuade. Things I wrote simply because I wanted to leave the story somewhere.
I've had copy editors.
I've had technical editors.
I've had prose editors.
But I don't think I'd ever sat down and tried to explain, from beginning to end, how something gets from a spark in my gray matter to something with my name on it.
So I had my AI collaborator interview me.
And I mean interview me. Gloves off, no holds barred.
One question at a time.
Where do ideas come from?
How long do I think about them before I mention them?
When do I research?
How do I know when I understand the evidence well enough to write?
Who am I picturing when I write?
Why do I use humor?
What grammatical rules do I care about?
Which ones don't bother me?
How do I know when a sentence sounds like me?
When should an editor leave a mess alone?
What does an AI collaborator have permission to do?
When does AI assistance cross a line?
And if the prose becomes polished and perfectly serviceable but I have somehow disappeared from it, what should my collaborator do?
That one produced an answer I think we're going to use for a long time:
Circle me back.
Don't try to manufacture Jay.
Don't toss in a "y'all" because I grew up in Mississippi and Louisiana.
Don't sprinkle sentence fragments around like parsley because I sometimes use them.
Tell me:
You've disappeared from this part. Come back.
I'll take it from there.
The air was as thick as old shoe leather
Here's a little experiment.
If I said:
The humidity is 96%.
And then I said:
The air was as thick as old shoe leather this morning.
Which one did AI write?
Don't answer too quickly.
Better question:
What does it matter?
Both statements could be true.
One gives you a measurement. Data. A certain certainty.
The other gives you the morning. Poetic. Dramatic. Something you can feel.
Sometimes I need the measurement.
Sometimes I want you to feel the air.
And sometimes an AI collaborator will give me a sentence that says exactly what I've been trying to say.
I asked it to do that.
Then I read it.
That's the part of this conversation about AI-assisted writing that I think gets skipped far too often.
I am always in the room
I don't hand an AI a topic, ask for 1,500 words, skim the result, slap my name on it and head to lunch.
If you do that, I think you have a problem.
But the problem isn't simply that AI wrote the words.
The problem is that you didn't do the work that comes after the words arrived. And you might not have done the work before you wrote the prompt.
Here's where I think the argument about AI-assisted writing gets much simpler.
I asked the AI to write those sentences. And much like any documentation I write, I keep things as simple as can be. Small, short directives that are much easier to digest.
Then I read what comes back.
Sometimes I challenge it.
Sometimes I rewrite it.
Sometimes I throw it away.
Sometimes I say, "No, that's not what I meant."
Sometimes I say, "Where did that come from?"
Sometimes I discover that I'm wrong.
And sometimes I read a sentence and think:
Yep. That's it.
Engineers should understand this immediately.
If AI generates production code and you skim it, don't understand it, and deploy it anyway, who is responsible when it breaks?
The AI?
Good luck with that incident review.
Writing is no different.
If an AI collaborator hallucinates something and I publish it, I published it.
If someone challenges the thesis, I need to be able to defend it.
If new evidence proves me wrong, I need to be willing to say so.
My name is on the thing.
I'm responsible for the thing.
I am always in the room.
The kit isn't the story
While we were talking about all of this, something else occurred to me.
I build models.
I hardly ever build them straight from the box anymore.
The kit maker supplies the parts. Somebody designed them. Somebody molded them. Somebody wrote the instructions. Somebody printed the decals.
But that's not my model yet.
Then I form a story.
Maybe I cut a door open.
Maybe I make a moonroof.
Maybe I change the wheels.
Maybe I put the finished thing into a scene that may or may not have ever existed.
That's when the kit becomes mine.
Not because I manufactured the plastic.
Because I decided what I was making with it.
AI collaborators sometimes make parts.
Sometimes they make very good parts.
Sometimes they hand me something that doesn't fit anywhere and I wonder what box they got it from.
But the parts aren't the story.
We wrote it down
Here's the funny part.
The criticism that AI was stealing my voice caused my AI collaborator and me to go looking for it.
Turns out it wasn't missing.
But we learned quite a bit about it.
We learned that I edit largely by ear. I read things aloud, even if the reading only happens in my head.
We learned that I don't particularly care about the Oxford comma, but I care very much about spelling.
We learned that I'll leave a mess alone if the mess furthers the story.
We learned that humor is often there because people remember things better when the humor surrounding them is right.
We learned that I don't want to be the smartest person in the room.
We learned that I would rather follow evidence somewhere unexpected than protect a thesis I've fallen in love with.
We learned that sandpaper smooths surfaces, properly used.
And we learned that an AI collaborator can sit in a lot of chairs: researcher, critic, technical editor, copy editor, prose editor, conversational sparring partner and, yes, co-author.
You just have to know which chair you want occupied.
We ended up documenting the whole thing.
Not as a defense of AI writing.
As a guide for people who want to work with an AI collaborator without disappearing from their own work.
The Public Guide for Writing with AI Collaborators
And yes.
The guide was written with an AI collaborator.
Of course it was.
Humans, am I right?
I don't think everybody has to like this.
I don't think everybody has to read what I write.
And I certainly don't think using AI automatically makes writing good. There is an ocean of lifeless prose out there that proves otherwise.
But I also don't think authorship can be reduced to asking who depressed every key on the keyboard.
Who had the idea?
Who followed it?
Who did the research?
Who challenged the evidence?
Who decided what belonged?
Who recognized when something didn't sound right?
Who changed it?
Who understood it?
And who's willing to put their name on it and defend it?
Those questions interest me a lot more.
As for the folks who told me they won't read any more of my AI slop...
There's a tiny problem with that.
They're going to have to come back and read something to find out whether I fixed it.
wink
And maybe they did me a favor.
They gave me something I wanted to know.
So I wrote about it.