Does AI's atemporal processing explain its preference for linear plots?
This explores whether the way transformers ingest text all-at-once (rather than living through events in sequence) is what makes AI-written stories drift toward simple, straight-line chronology — and what the corpus actually has to say about that link.
This explores whether AI's 'atemporal' processing — reading a whole passage in parallel rather than experiencing it unfold in time — is the reason its fiction leans on linear plots. The corpus doesn't pose the question in exactly those words, but it stacks up two findings that, read together, make a real case. First, the tell is real: a story detector called StoryScope separates AI from human fiction with 93% accuracy using *only* discourse-level features like chronological structure and character agency, and it keeps 97% of that accuracy after stripping away all the surface style Can AI stories be detected without analyzing writing style?. So the AI signature isn't word choice — it's how the story is *shaped*. Linear chronology is one of the things giving it away.
Where does that shape come from? The most mechanical clue is how transformers handle text at all. They integrate every token through weighted parallel aggregation — adding information up simultaneously rather than selectively suppressing what doesn't fit — which is exactly why they miss jokes, wordplay, and any meaning that depends on holding a frame across time Why do AI systems miss jokes and wordplay so consistently?. A non-linear plot (a flashback, an unreliable narrator, a payoff that reframes the opening) is the narrative version of that same demand: you have to suspend one meaning and let a later one rewrite it. If the architecture aggregates rather than re-frames, the path of least resistance is to lay events down in the order they happen.
There's a deeper version of this in the 'Plato's cave' note: text itself is a lossy abstraction that strips out the physics, geometry, and causality of real events, so a text-only model manipulates symbols without the underlying dynamics that generated them Are text-only language models fundamentally limited by abstraction?. Causal and temporal structure is precisely what gets thinnest in that compression — which would push generated narrative toward 'and then, and then' sequencing rather than the causal braid humans build. Add that LLMs behave like scaled-up System-1 cognition — fast, associative, pattern-completing rather than deliberately structured Why do people trust AI outputs they shouldn't? — and a preference for the smoothest, most predictable ordering follows naturally.
The honest caveat the corpus forces on you: nothing here measured plot structure as a function of processing order directly, so 'atemporal processing' is a clean story rather than a proven cause. And one note cuts the other way — from inside shared discourse, humans and LLMs draw on the same symbolic substrate, so the difference may be structural rather than absolute Do humans and LLMs differ fundamentally or just superficially?. That leaves the interesting takeaway: the linearity you sense in AI fiction is measurable, it's a structural fingerprint rather than a style quirk, and the best available explanation isn't that the model lacks imagination — it's that parallel, time-flattened processing makes the straight line the cheapest line to draw.
Sources 5 notes
StoryScope achieved 93.2% accuracy separating AI from human fiction using only discourse-level features like character agency and chronological structure, retaining 97% of performance while eliminating stylistic cues. These structural choices resist humanization because they require rewrites, not surface edits.
Transformers integrate token information through weighted parallel aggregation rather than selective suppression of irrelevant words. This structural difference explains consistent failures with jokes, wordplay, and frame-dependent meaning—not knowledge gaps, but missing cognitive operations.
Text strips the physics, geometry, and causality present in reality, forcing language models to manipulate symbols without grounding in their source dynamics. This creates predictable failure modes in physical, geometric, and causal reasoning that multimodal training could address.
Rose-Frame identifies map-territory confusion, intuition-reason conflation, and confirmation-bias reinforcement as traps that multiply their distorting effects when they co-occur. Evidence from cross-linguistic overreliance and architectural transformer biases confirms the compounding mechanism operates universally.
Applied Habermas's observer/participant distinction to AI: from outside, humans and LLMs are utterly different; from within shared discourse, both draw on the same symbolic substrate, making the difference structural rather than absolute.