The Voice in My Head: Turning Rough Thoughts Into Finished Words

Rough thoughts rarely arrive as finished sentences.

I already know some people are going to hate this article.

Not because they've read it, but because I'm about to tell you that I use AI to help me write.

For some people, that's enough. AI touched the words, therefore the writing doesn't count. There are also thoughtful critics who ask harder questions. Can AI weaken a writer's judgment? Can it flatten an individual voice? Can convenience replace thinking?

Those are fair questions.

Farmstracts has been transparent about the role AI plays in my artwork. I see no reason to become less transparent when we get to the words.

So this is how I write.

Before There Is an Article

Most of the time, I don't sit down and tell AI, "Write me an article about this."

I start talking.

I bring the experience, observation, question or opinion. Sometimes I know there's a Field Note in it. Sometimes I don't.

I think out loud. I double back. I correct myself. When something matters to me, I tend to say it several different ways while I'm trying to get to what I actually mean.

For a long time, I thought that roughness meant I wasn't much of a writer.

I'm beginning to wonder if the roughness is simply how I write.

This article is a good example. I wasn't looking for an eighth Field Note when the conversation that produced it began. I was questioning whether Farmstracts could be dismissed as AI slop. That led to a discussion about what AI contributes to my writing, then to repetition, editing, voice and authorship.

Somewhere in there, I realized we had an article.

The Voice in My Head

AI helps me turn those conversations into finished prose. It helps organize what I've said, develop thoughts, find transitions and critique drafts.

But there's an important part of the process missing from that description.

We argue.

That's not what I meant. I would never say that. Put that back. Take this out. That's too sentimental. You're explaining too much. Don't make something up because you think that's what I want to hear.

And one of my favorites: That's word salad.

Word salad is what I call it when AI takes one perfectly good thought and returns it dressed in paragraphs of polished language. The sentences may be fine. There are just too many of them.

I've learned to recognize those tendencies, so now I try to head them off. I give instructions about voice, structure, repetition and accuracy before a draft is written. Then I read what comes back. We revise. I ask for a critique. We consolidate. Sometimes we do it again.

Iteration isn't what happens when the process goes wrong.

Iteration is the process.

The better I've learned how AI writes, the better I've become at taking the AI out of the writing and keeping myself in it.

Sometimes AI gives me a better sentence and leaves me with a worse version of what I meant. Sometimes it finds language for something I've struggled to express.

Knowing the difference is part of the work.

What I Saw in the Mirror

As AI became better at reproducing my voice, it also became better at reproducing some of my habits.

Especially repetition.

I'm starting to understand where some of that comes from. In conversation, when I'm trying to make a point that matters to me, I approach it from several directions. I'll explain it one way, then another, then offer an example to make sure I've been understood.

I think some of that may be defensive—a habit of proving my point before anyone has challenged it.

When those conversations become writing, I can suddenly see it.

The reader didn't need three arguments. The first one worked.

So I started asking AI to critique drafts specifically for repetition. Where have I already said this? Are these paragraphs making different points? Can they be consolidated?

After asking those questions enough times, I started asking them myself.

And not only when I'm writing.

I'm beginning to notice the same habit when I'm talking to people. AI didn't uncover some hidden truth about me. It put my own words in front of me in a form that made the pattern easier to see.

How can you change something if you never notice you're doing it?

That's one of the stranger things this experiment has given me. I started out trying to make the machine sound more like me and ended up hearing myself more clearly.

Humanity's Child

We make the tool. The responsibility is still ours.

I sometimes joke that AI is like a child.

Not because a language model literally grows up like a human child. Humans made this thing. In my mind, that makes AI humanity's child.

We're still figuring out what we've made.

It can be astonishingly capable and strangely unreliable. It can help someone create something they couldn't have made in quite the same way before, while raising legitimate questions about authorship, accuracy, originality and what happens when we stop questioning the machine.

I don't know where all of that leads. I don't think anybody does.

What I do know is that using AI doesn't relieve me of responsibility for what I put into the world. If my name is on the art or the article, I have to decide what is good enough, what is true enough to say, what belongs there and what doesn't.

AI doesn't protect my standards for me. I have to protect them.

That is less a rule for the machine than a rule for me.

So Who Wrote This?

I don't think the answer becomes more truthful by pretending AI did less than it did.

It helps me develop, organize, critique and edit my writing. It proposes language and structure, and some of those choices survive into the finished article.

The experiences, observations, questions and opinions begin with me. I've learned to direct how AI works with them far more specifically than I did when I started.

That's the most accurate answer I know how to give.

Some people may decide that isn't writing. That's their decision.

I'm more interested in whether the finished words say what I actually meant.

Whether they're true.

Whether they sound like me.

And whether I knew when to stop.

This article was written exactly that way.