102 asks O’Sullivan’s question from the writer’s side: what happens when the production of prose no longer guarantees a mind behind it? Mandy Brown answers the same question from the reader’s side. Reading, she writes, “presumes that what we are reading is the product of a mind, that the reading is itself a gathering of minds.” When there is no mind, “we find not a mind but a forgery, and a glib one at that—a thin, transparent skin wrapped around an empty void.”
StoryScope measures where that forgery shows. AI over-explains the narrative compared to humans (AI narrators explicitly state the story’s theme 77% of the time vs. 52% for humans.) AI doesn’t trust the reader to infer meaning. That inference is what powers the human writing. It’s that nuance the narrator creates that connects with people because every reader can have different takes from the narrative while AI’s verbosity is there to reduce the ambiguity. That ambiguity is the heart of the story.
The paper names the same mechanism: “The pattern is one of over-determination: AI spells out meaning rather than trusting the reader to infer it.” The rest of the numbers point the same way — humans keep morally ambivalent protagonists (59% vs. 38%) and treat the audience as a co-participant (an aside to “you, dear reader”), while “AI writes as though no one is watching.”
One number cuts against the easy version of this argument: AI’s references to other works are vaguer than ours (72% vague allusions vs. 50% for humans; humans name things specifically). So vagueness isn’t the ambiguity I mean. AI is imprecise about its sources and over-precise about its meaning. The ambiguity that works is the opposite pair — concrete in the particulars, open in what they add up to.
So the gathering of minds needs a gap to happen in. Meaning that is stated is delivered; meaning that is inferred is produced — by the reader, in the reader. A text that closes every gap leaves the reader nothing to do, and nothing to do means nobody there.
This doesn’t overturn what I said in 102a — it bounds it. “It won’t matter who wrote it as long as it’s valuable” holds for writing whose value is delivered (the fact, the answer, the how-to). It breaks for writing whose value is produced in the reader, because there the reader’s inference is the value, and you can’t get it from prose that has already done the inferring for you.
Related Note(s):
- 102: the root question, answered from the writer’s side (accountability) — this is the reader’s side of the same question.
- 102a: my “valuable regardless of who wrote it” answer, which this note gives a boundary condition.
- 29: “Certainty is the enemy of open thinking.” Over-determination is certainty applied to a story; ambiguity is what keeps the reader open.
- 2d: a tension worth holding — one idea for the reader to leave with is not the same as one meaning spelled out. Zinsser’s rule is about the writer’s discipline, not the narrator’s commentary.
- 69b: the average of the internet also averages out the ambiguity; the tidy single-track plot is what the mean looks like.
- 73a: if you can’t validate the source of truth, you also can’t do the inferring — both notes are about what the reader is left holding. Sources:
- StoryScope: Investigating idiosyncrasies in AI fiction — Jenna Russell, Rishanth Rajendhran, Chau Minh Pham, Mohit Iyyer, John Wieting (University of Maryland / Google DeepMind), preprint. Figures from §4.1 and Table 15; 61,608 stories.
- Umyazu by Mandy Brown
- We’ll soon find out what is truly special about human writing by James O’Sullivan
