Sometimes the AI fixes a mistake by making sure the reader knows it has been fixed.
Imagine writing a scene in which a character is fighting. The AI describes them struggling to breathe, their movements slowing as exhaustion takes hold.
There’s nothing inherently wrong with that scene. It might even be rather good.
Except the character isn’t supposed to be exhausted.
So I tell the AI that they’re still energetic and capable at this point in the story.
A sensible request, I would have thought.
The next version may remove some of the obvious signs of fatigue, but then insert a reassuring sentence declaring that the character is not tired.
And suddenly I’m staring at a problem I never intended to write about.
The AI has solved the question I asked. Unfortunately, the book now seems strangely concerned with convincing the reader that a character isn’t exhausted.
The reader has never seen my instructions. They have no reason to wonder about the character’s stamina. Yet here is the narrator, apparently determined to settle the matter.
An invisible conversation on the page
I have encountered versions of this problem often enough to become rather suspicious of unnecessary negations.
A character does something without looking at another object. Someone doesn’t speak when there was never any reason for them to say anything. A paragraph explains that a decision wasn’t made for one reason, but for another, even though the first reason had never occurred to the reader.
Sometimes the missing action is important. Often it seems to exist because an earlier version contained something that needed correcting.
It’s as if the conversation between me and the AI has left a footprint in the fiction.
One example from the development of Book Three is wonderfully simple.
An earlier line said:
Chef did not answer because the cassette slid out.
The revision became:
The cassette slid out.
That’s it.
The cassette falls. The interruption happens. We don’t need to explain the answer Chef failed to give, because the reader wasn’t waiting for a report on his failure to answer.
The simpler sentence lets the event speak for itself.
When removing something creates more of it
There is a second, particularly peculiar version of this problem.
Sometimes, when I ask for something to be removed, the AI seems reluctant to leave an empty space.
It takes out an unnecessary explanation and replaces it with another explanation. It removes a meaningless gesture and invents a slightly different gesture. It corrects an unwanted physical condition by describing another condition nobody needed to know about.
The paragraph survives. The problem changes costume.
This can be remarkably frustrating because the rewritten passage may sound perfectly natural in isolation. I have to remember what I actually wanted from the scene before the conversation became about correcting it.
And sometimes the answer is that I wanted fewer words.
A paragraph doesn’t need to be compensated for losing a bad sentence.
When absence matters
I wouldn’t want to remove every negative construction from fiction. That would be a terrible solution to a problem caused by overly literal solutions.
A character refusing to answer can be heartbreaking. Someone deliberately avoiding another person’s gaze can tell us more than an entire conversation. Silence can be threatening, affectionate or devastating.
One of the examples that emerged during the writing process concerns Hex and an atlas. If the atlas is within reach and Hex, a hand normally inclined to investigate things, doesn’t reach for it, the absence can be meaningful.
The point is not that nothing happens.
The point is that something we expected has failed to happen.
There is a lovely difference between those two situations, and I want the writing to recognize it.
What I’ve learned to ask instead
When a passage feels wrong, I try to think about what I actually want the reader to experience.
If the character should be confident in a fight, I want to see that confidence in how they move and decide. If a conversation is meant to be uncomfortable, I want the discomfort to emerge from the characters rather than from a sentence explaining what the scene is not.
AI remains enormously useful for rewriting. But I’ve become wary of corrections that seem to address me, the editor, instead of the person reading the novel.
A good revision shouldn’t make the reader aware of a mistake they never encountered.
It should let them enjoy the scene as though the mistake had never existed.