I spent about four months writing a book recently.
AI was involved.
That's a much less satisfying sentence than either side of the argument usually wants.
I could say I wrote the book myself, which is true in one sense. The ideas accumulated over decades. They connect to software I've built, a role-playing game I designed years ago, experiments that went nowhere, relationships I've had, things I've lost and observations I've carried around long enough for them to change shape.
I could also say AI helped me write it, which is true.
So what happens when someone reads something I've written and responds:
Thanks, ChatGPT.
My first reaction isn't especially philosophical.
It's annoying.
There is something strange about spending years developing an idea, months turning it into a particular form, and then watching the whole process compressed into the presumed location of a button.
Once the irritation passes, though, there's a real problem underneath it.
The reader can't see the process.
They only have the artifact.
That used to tell us more than it does now.
If someone produced a reasonably polished 240-page book twenty years ago, you could make certain assumptions about what probably happened before the book reached you. Those assumptions weren't always correct, of course. Editors, ghostwriters and collaborators have been around for a long time.
AI changes the uncertainty by degree.
A person can now produce competent prose about something they barely understand. Another person can spend thirty years thinking about something and use the same technology to make the final explanation simpler.
From the outside, those two artifacts may look surprisingly similar.
That's new.
I've noticed the problem partly because AI is useful to me for exactly the reason that makes authorship harder to see.
My natural tendency is to overbuild explanations. I can happily disappear into abstractions, relationships among abstractions and several layers of qualification while some poor reader is still waiting for me to finish the original sentence.
AI can help me notice that.
Sometimes the useful contribution isn't adding anything. It's helping me remove the scaffolding I needed while thinking so the reader doesn't have to climb through it afterward.
That creates a funny result.
More work can produce an artifact that looks like less work.
Software has taught me the same lesson.
One of the systems I've spent years building is the digital infrastructure behind a small regional newspaper. A subscriber in another state can log in, download the newspaper, read local stories or listen to a radio stream.
Ideally, nothing about that experience communicates how difficult the underlying system was to build.
That's the point.
Nobody turns on a faucet hoping to admire the plumbing.
Writing has plumbing too.
The reader shouldn't have to encounter every failed idea, abandoned metaphor, argument with myself, technical experiment or earlier version of the thought merely so I can prove that work occurred.
If those things did their job, much of them should disappear.
AI makes disappearance cheaper.
That's useful, and it has a cost.
We increasingly can't determine how much understanding sits behind an artifact merely by inspecting its polish.
Maybe that means we'll need to become less interested in polish.
Does the idea survive questions?
Does it connect to anything outside the piece?
Has the person actually done something with it?
Does the work change over time when reality pushes back?
Can the author explain why something is there, revise it without losing the underlying thought, or admit when part of it doesn't work?
Those questions don't establish some pure boundary between human and machine authorship either. I'm not sure such a boundary would be particularly useful.
They tell us something about the relationship between the artifact and the person presenting it.
That relationship matters to me because the book didn't appear by