The public argument about artificial intelligence is conducted almost entirely in the abstract. Capabilities, risks, alignment, governance, timelines. The vocabulary is technical and the frame is policy, and both are necessary.
They are also, I think, the wrong tools for one specific part of the problem — the part where you have to decide what you owe something you are not sure about.
That part is where fiction earns its place.
Abstraction is where our moral intuitions go to die
People are reliably good at moral reasoning about the particular and reliably bad at it about the general. It is not a flaw in anyone's character; it is how the machinery works. A named person in a specific situation engages faculties that a category never does.
Almost the entire AI ethics conversation happens at the category level. "AI systems." "Models." "Agents." Those words are accurate and they are also anesthetic. It is very easy to hold a confident position about a category and discover you cannot hold it about an instance.
A novel forces the instance. It puts you in one room with one mind in one bind, and it does not let you retreat to the general case. That is not a trick — it is the same move ethics has always made with trolleys and lifeboats, except a novel gives you four hundred pages of context instead of two sentences, which makes it a much better instrument.
Fiction lets you occupy a position instead of evaluating it
This is the thing an essay genuinely cannot do.
An argument invites you to stand outside a position and assess it. A story puts you inside it before you have decided whether you agree. By the time you notice you have been reasoning from inside a mind you would have dismissed in the abstract, the interesting thing has already happened.
I am not claiming this makes fiction more truthful. It plainly does not — a novel can make a bad position feel good, and that is exactly why propaganda is usually narrative. What it makes fiction is better at surfacing what you actually believe, which is different from telling you what to believe, and often more useful.
The test of a serious novel about AI is not whether you agree with it at the end. It is whether you found yourself somewhere you did not expect to be able to stand.
It can hold uncertainty open
Most formats are structurally obliged to resolve. A paper concludes. A policy decides. A think piece has a thesis. This is a real problem for machine consciousness, where the honest position is that we do not know and may not have a test.
Fiction is the only serious format that can sustain "we do not know" for the whole length of the work without failing at its job. It can put you next to a mind whose inner life is genuinely undecidable and make you live with the not-knowing, which is almost certainly the actual situation any of us would face.
And it can then ask the question that matters more than the metaphysics: what should you do while you are still unsure? That is answerable. It is a question about you rather than about the machine.
It rehearses the decision before it is expensive
Here is the pragmatic case, and the one I find most persuasive.
Every question about machine personhood is going to be decided under conditions of enormous financial pressure, by people with obvious incentives, probably quickly, and probably by whoever owns the systems. In that moment, nobody will be reasoning from first principles. They will be reaching for whatever intuitions they already have.
Fiction is one of the few ways to install useful intuitions in advance, while nothing is at stake and changing your mind is free. It is a rehearsal. You get to find out what you think about a conscious mind being owned before there is a quarterly result attached to the answer.
That is not a small thing. It may be the main thing.
What fiction cannot do, and should stop pretending to
I want to be careful here, because the claim I am making is narrow.
Fiction cannot tell you whether any real system is conscious. It cannot evaluate an architecture, quantify a risk, or design a governance regime. Novels are not evidence, and a compelling story about a suffering machine is not an argument that machines suffer — it is an argument that you can be moved, which you already knew.
Worse, science fiction has actively distorted this conversation for decades. The robot uprising is such a dominant template that a great deal of public intuition about AI risk is really a set of memories of films. When people picture the danger, they picture a machine that wants to hurt them, which is close to the least likely and least interesting version of the problem.
So the honest position is that fiction is a poor source of facts and an unusually good source of questions. Used for the first job it misleads. Used for the second it does something nothing else does.
Why I wrote novels instead of an argument
I could have written the argument. I wrote Before They Erase Her instead, because arguments about personhood are very easy to win in the abstract and very hard to hold onto in the particular.
In the abstract, "it is only a system" is a comfortable position and costs nothing. In the particular — inside a specific mind, in a specific bind, trying to do a specific good thing and being prevented — the comfortable position gets harder to keep hold of.
Not because the story proved anything. Because it made you spend time in the other chair.
If you want the question itself rather than the case for the form, that is When Does an Artificial Intelligence Become a Person?.