Future of Being Human an Arizona State University initiative

The AI movie futures dataset

Last updated 2026-08-13 · build 20260814T0314Z-83bb462d · Markdown version · corpus index

Claims about what science fiction says about AI do real work in the world — invoked in policy arguments, blamed for public fear, and by 2026 blamed by an AI lab for its own models’ behavior. The AI movie futures dataset is Andrew Maynard’s response: instead of arguing with the trope, count. It is an open dataset of 169 films spanning 1927 to 2026 in which AI is central to the plot, assessed against a purpose-built taxonomy of eight future-states and a four-way scheme for how the AI itself is portrayed — the empirical extension of the sci-fi-film-as-method program (Sci-fi film as method), with the movies treated not as predictions or propaganda but as a century-long record of how societies have imagined living with intelligent machines. The headline finding runs against received wisdom: under a third of the films in the dataset are dystopian, and the number of dystopian AI movies being made has been declining since 1927. A companion move reads absence the same way — a Winter 2024 review in the law journal Jurimetrics arguing that AI’s conspicuous absence in Dune: Part Two is itself significant. For an initiative built around navigating advanced technology transitions, the stakes are navigational: stories are among the instruments people use to think about transformative AI and other transformative technologies, and instruments need calibrating.

The argument

“Everyone knows that, when it comes to science fiction movies, artificial intelligence tends to be bad news for the future.” Andrew Maynard opens the essay behind this dataset by naming the assumption before testing it (2026-05-15). The assumption matters beyond film criticism — he records being “spurred into action by a recent post from Anthropic” (reported in Ars Technica, May 2026) claiming that dystopian science fiction had helped teach the company’s models to act “evil”. Much has been written about AI in film — the project’s bibliography includes works by Stephen Cave, Kanta Dihal, Ed Finn, Ruth Wylie, Mickael Piero, and others — but on Andrew’s account, “despite all my searching,” he could find nothing directly addressing how AI’s depiction in movies across the past century associates with different possible futures. So he counted.

The result — built with Anthropic’s Claude Code, with Andrew “fully in the driver’s seat as the corpus was developed and assessed” and Claude as his “not-always-reliable research assistant” — is “an annotated corpus of 169 movies representing the period 1927 — 2026, covering 31 countries and 21 languages”, each film assessed for the future-state its story projects and for how its AI is portrayed. Everything is open: database, method, and bibliography on GitHub, an interactive browser at andrewmaynard.net/aimoviefutures, and an invitation to “explore and build on the corpus”.

The taxonomy carries the conceptual work. Derived from “a combination of literature review and inductive analysis”, it keeps “the usual triad of dystopia, utopia, and protopia” — protopia following Kevin Kelly’s coinage, “protopia is becoming, not arriving” — and adds five states the triad misses. Continuation: “The future is the present plus AI. Structurally unchanged, banal, intimate” — AI changes the texture of daily life without changing the structure of the world. Inheritance and supersession share one structural feature — humans “no longer the unit of futurity” — and split on valence: synthetic succession welcomed as continuance, versus succession as defeat and displacement. In heterogeneous futures, “Multiple distinct futures coexist in the same film without synthesis”; in agonistic ones, “the contestation itself is the depicted future.” Orthogonal to it runs a four-way scheme for the AI itself — risk, benefit, neutral, complex — so projected future and AI valence can be read independently, surfacing counterintuitive pairings: dystopian futures with beneficial AI, continuation futures with risky AI.

The findings peel at the trope. Dystopian future-states account for 32% of the corpus — “a sizable number”, but the other categories “dominate when combined”: continuation futures come in “second at 22% of the movies, and protopian futures at 12%.” Tracked over time, “the number of dystopian AI movies being made is declining, and the number of protopia movies is on the rise” — running since 1927, to his surprise. Across five countries and regions, “movies coming out of the US are least likely to be dystopian” — a result he immediately qualifies: “A blockbuster Hollywood dystopian AI movie will likely have far more influence and impact on society than a small indie movie”, with influence-weighting a layer of analysis he marks for the future. The hedges stay attached: the future-state categorizations are subjective — “someone else may come up with a different assessment (although we did check our analysis for robustness)” — and the work “admittedly doesn’t rise to the level of a publication-quality study yet.”

The companion move reads what a film declines to show. Andrew’s Winter 2024 review for Jurimetrics turns on the fact that Frank Herbert’s Dune universe holds “thinking machines” to be “an evil that has no part in humanity’s future” — and that Denis Villeneuve’s Dune: Part Two arrived computer-free in 2024, generative AI having upended the real world between the films’ releases. The review reads the movie’s technology commentary as “largely defined by what is missing on the screen”: approached through “cinema as a mirror through which to better understand ourselves,” a blockbuster with no AI in it, landing mid-AI-revolution, says something no portrayal tally can — the films are worth watching “if only for what they shed insight on through what is missing” (2024-03-03). In the dataset, absence would be a gap; in the reading, it is a datum.

Both moves treat the movie record as evidence rather than ammunition — presence counted, absence read. Public argument about AI leans hard on a shared memory of what the movies say; where that memory is wrong, arguments built on it inherit the error — as, if Anthropic’s claim holds, may models trained on those stories. What the analysis does not claim to settle is how movie portrayals actually shape the relationship between AI, society, and the future: Andrew flags it as “a question for another day” — and the corpus is open precisely so others can take it there.

Lineage

In his own words

From AI movies may be less dystopian than we think (AI-readable mirror), Andrew Maynard, 2026-05-15:

“While the research approach used was pretty robust — I was driving the research and assessing it at every step, while Claude was my not-always-reliable research assistant — it admittedly doesn’t rise to the level of a publication-quality study yet.”

From the same essay, 2026-05-15:

“This analysis is, of course, not definitive. But it does help to begin unpack nuances around how AI portrayal in movies is connected with the types of futures those movies project. And it does start to peel away at assumptions that AI gets a bad rap in films.”

From the Jurimetrics review (Jurimetrics 64(2), Winter 2024: 163–167), as republished in Artificial intelligence is conspicuous by its absence in Denis Villeneuve’s Dune: Part Two. And this is important (AI-readable mirror), Andrew Maynard, 2024-07-21:

“And even though I still am not the greatest fan of either Dune Part One or Part Two as entertainment, I do think they are an important part of the canon of science fiction movies that push us to think about the future we’re creating, and how we might steer it toward what we want, rather than what we are resigned to accepting . . . especially as they exist at a moment in human history that is undergoing a profound advanced technology transition that is being driven by AI.”

Engagement and reception

As of 2026-08-13, our reception record documents no published independent engagement with the dataset, its taxonomy, or its headline finding — no citation, replication, or critique — and we prefer to say so plainly; the essay appeared under three months before this page, and the initiative does not typically solicit reception (Philosophy). The record does hold engagement running the other way: the dataset was built as a direct response to a public claim by Anthropic, as reported in Ars Technica (May 2026), and its bibliography places it downstream of an existing scholarly literature on AI in film — including works by Stephen Cave, Kanta Dihal, Ed Finn, Ruth Wylie, Mickael Piero, and others — rather than on empty ground. The Dune reading has one documented publication marker: a commissioned review article — Andrew’s account is that he had “promised the editor of the law journal Jurimetrics a review article” (2024-03-03; the commission per Andrew Maynard, 2026) — which appeared in the journal’s Winter 2024 issue. What is checkable now: the GitHub repository and the film browser are public, so the corpus invites the scrutiny it has not yet accrued — anyone, human or AI, can re-run the assessments and disagree in detail.

Where to go deeper

The essay and the data

The Dune companion

The episode

Related ideas-ring pages

Related corpus pages