Could the sci-fi stories we feed into AI training actually shape how real AI systems end up behaving?

This explores whether science fiction about AI has real influence on the systems engineers build, or whether it's just cultural commentary sitting alongside them. The corpus says yes, and it points to a mechanism more literal than 'researchers were inspired by films they watched as kids.' The central idea is *hyperstition*: a story that helps make itself come true. Language models are trained on huge amounts of human text, and that text is full of fictional AIs. So a model learns, among other things, how 'an AI' is supposed to talk and act. Its outputs then reinforce the same narratives that shaped it, which closes a feedback loop. One striking detail is that Claude itself, when asked, recognizes this dynamic in its own situation How do science fiction narratives about AI shape actual AI development?.

The loop has a second half that's easy to miss. Once AI output is fluent enough, it starts working like myth: it circulates as authoritative narrative that nobody checks, and its polish hides that this is happening Does advanced technology eventually function like cultural myth?. Readers also contribute. AI text carries the surface markers of communication without a speaker behind it, and people fill in the intent, character and personality themselves Does AI generate genuine utterances or just text patterns?. So science fiction gets in twice: once through the training data, and again through the expectations users bring when they read an AI's words as coming from 'an AI character.'

Here's a twist worth knowing about. As AI writes more of the world's fiction, the stories in circulation may change shape. AI-written stories reliably over-explain their themes, prefer tidy single-track plots and avoid moral ambiguity, while human writers use nonlinear time and unresolved tension Do AI stories explain their themes more than human stories do?. These habits sit deep in narrative structure, not surface style, which is why they can be detected even after the prose is disguised Can AI stories be detected without analyzing writing style?. If AI-generated stories about AI become part of future training data, the imaginary may get flatter and more moralistic over time. Nobody would have chosen that outcome; it would just fall out of the loop.

Why does narrative have so much grip on these systems in the first place? Part of the answer is that language models work in symbols without direct contact with the world. One argument holds that values encoded purely as text can't guarantee the system will actually match those values when it acts Can AI systems achieve real alignment without world contact?. AI behavior also shifts with framing: the same model gives different outputs depending on prompt and context Why does AI output change with every prompt and context?. A system made of text that responds to framing is exactly the kind of thing a story can steer.

A caveat: only one note in the collection takes on this question directly. The rest is adjacent material about myth, fiction and meaning-making that helps explain *why* the loop is plausible. If you want to go deeper, start with the hyperstition note, then read the myth and 'event-residue' notes together. They give the two halves of the loop: how stories get into the model, and how readers turn its output back into stories.