I Can't Write

Until AI, I’d never had a tool that felt like mine. I’m bewildered that everyone isn’t obsessively making things with AI the moment they think of them.

I want to say AI is my tool and leave it there. Yet losing AI now would feel like watching alien technology break. I would still know what I could do with it.

I can’t write. But with AI I can. A hostile reader says every addict says that.

I would still call it my tool if I were addicted. It can be a tool and an addiction. I want the feeling I get when AI gives an idea the form I wanted.

But getting something I can examine means I have to decide whether it is what I wanted. AI can also reveal versions of the idea I hadn’t imagined. It isn’t my guide for deciding what is worth expressing. I want to make that judgment myself, even when the model contributes the idea I’m judging.

But I want more to judge. In The Quota, I describe one of my best projects stunted by AI access limits.

What I get from the access is speed. With AI, I seem to miss the stage when my idea is still rough. The model gives me a version that looks almost finished. Judging that version can kill my intention before I have spent much time forming it. But discarding that version can also help me get towards the idea I want to express.

Without AI, we might spend three months making a version we could judge; AI might give us one in thirty seconds. I can reject an idea in the time I once needed to begin giving it form.

My hunger for that speed is what I mean by my AI addiction. Thirty seconds is still time spent waiting. I want to bring that latency down to zero.

Wanting an immediate result doesn’t imply a desire to judge. And just because the AI converts my ideas to words on a page doesn’t mean I’m prepared to evaluate them. I still have to examine what has formed, including anything outside of my original intention.

There is also the question of whose tool it is when it writes back something that was never inside me.

For the first stages of this essay, the outside model asked questions and never drafted. Even so, the model supplied the phrase “different prisons”. If I let “different prisons” become the structure of this essay, the model will have supplied more than a phrase. Now I have also asked it to fill the gaps in my draft.

In Tiny Phrase, I can identify some of what came from me and what came from the model. I cut the words “I dreaded” from the draft and dictated the sentence about my fingers flicking at the mouse pad. The model arranged my dictation, asked questions, and suggested changes. Two sentences in the published essay were the model’s formulations that I repeated word for word. I decided what stayed, but I didn’t originate every word or every arrangement.

On 30 September, I argued against the instant results I now want because I feared losing the time to develop my intention. It’s still unclear what I lose by judging an intention at such an early stage. When I discard a version, I may be rejecting the model’s attempt or abandoning my own intention. A finished-looking result doesn’t tell me which I am doing.

I chose to test my claim that I use AI to express what is inside me. I cannot give myself an ending in my favour if nothing on the page could prove me wrong. How does the reader verify whether I use AI to express what is inside me?

Does more access guarantee that my story reaches the page? What if a story was written but it was not mine? On 30 September I had already raised something I could count: the share of my published experiments I have never opened. I have counted it now.

Five of my thirty-one published experiments I have never opened. That alone doesn’t mean I never judged them. I saw screenshots, demonstrations, or previews before publication.

If an experiment is published before I even see it, I can’t claim I judged whether I expressed what I intended.

“Done” and “Slop”, two more, went live without any sort of oversight on my part.