Written with AI" tells you nothing. The disclosure that matters is how.
retained text
Most of the argument about AI and authorship is stuck on whether a model was involved. I use these tools daily and this isn't an argument against them — I think "whether" is close to the least informative question available, and the fights about credit and royalties are all downstream of it. Two different things keep getting folded into one word. Contribution is what any participant adds. Plenty of things contribute to a piece of writing: editors, a decade of reading, arguments you half remember, and now a model. Contribution is cheap, and it always was. Authorship is narrower. It's choosing which question is worth a year of your life. It's cutting the claims that don't survive contact with your own experience, including the fluent and flattering ones. It's standing behind the result when the result turns out to be wrong. A model does none of that — not for want of capability, but because it has nothing at stake. You can hand a tool the work. You cannot hand it the consequences. It will accept them without resistance, and absolutely nothing will have happened. Which is why "AI was used" tells me nothing. These are three different things: \- Drafted by a model, lightly edited. \- Used to find structure in material already written. \- Used to argue against the writer until the weak claims fell over. Three different books. Right now they all carry the same silence, or the same vague line in an acknowledgements page. I'd rather we argued about a norm for saying which one it was than about whether the model gets a byline. Is a workable disclosure standard even possible here, or does it collapse the moment it's voluntary?
document fetched 2026-10-20 00:20 · canonical https://reddit.com/r/artificial/comments/1x9oxkd/written_with_ai_tells_you_nothing_the_disclosure