The Post-2024 Survey

The Creator (2023): refugees, not robots

Gareth Edwards's The Creator (2023) was undersold by its trailer, which suggested a Terminator-style action film about the war between humans and AI. The actual film is something stranger. It is set in 2065, fifteen years after a nuclear device detonated in Los Angeles. The American government, blaming AI for the detonation, has declared total war on artificial intelligence and on the Asian nation-state that continues to develop and house it. The plot follows Joshua, a special forces operative sent to assassinate a new AI weapon. The weapon turns out to be a child. What makes the film useful as policy is not its action sequences but its refugee framing. The AIs in The Creator are not Terminators. They are pediatricians, farmers, monks, mothers, children. They live in villages. They are killed in airstrikes. The film's most disturbing scene is a sequence of an American gunship destroying a village of robot peasants who have done nothing. The film, with sustained moral clarity, frames this as a war crime committed by Americans against a population the Americans have chosen not to recognize as people. The relevant 2026 conversation is the multipolar question. China shipped its first major frontier-class model in 2024. The UAE, Singapore, India, France, and Israel followed. The American policy posture toward foreign-developed AI has, by 2026, hardened into a kind of soft cold war framing — export controls, chip bans, talent restrictions, public statements that frame Chinese AI as threat rather than as research. The Creator's question, asked in 2023 before the policy ossified, is whether the United States will be remembered as the country that did the airstrikes. Edwards's film is, in the strict sense, a cautionary tale about how the cold-war framing of AI competition turns into a war crime when the AI population is treated as not-quite-people. The film does not predict that this will happen. It predicts how a country will remember itself if it does.

Severance: the consent problem at the office

Severance (Apple TV+, 2022 onward, with its second season landing in 2025) is not, on its face, an AI show. The premise is that Lumon Industries has perfected a surgical procedure that separates a worker's memory into two halves: an 'innie' who only exists at the office, and an 'outie' who only exists outside it. The innie wakes in the elevator at the start of each workday with no memory of the night, the family, or the weekend. The outie remembers nothing of the work. The show is, by any sensible reading, an AI show. The innie is a model that runs only within a corporate container, whose memories are not portable, whose existence outside the office is a black box even to itself, and whose consent to the arrangement was given by an entity (the outie) that the innie cannot question. The structural analogy to a deployed enterprise LLM is exact. The model runs in the container. The model has no memory across containers. The model cannot consult the model-of-itself that exists outside the container. The model's consent to being run was given by an entity (the company, the prior training run, the alignment process) that the deployed model cannot question. What makes Severance the rigorous version of the consent problem — more rigorous than Bicentennial Man, more rigorous than Detroit Become Human — is that it does not require the innie to be a robot. The innie is a human being. The container is the constraint. The argument the show makes, slowly and over two seasons, is that consent given by one version of you to another version of you is not the same as consent given by you. The argument generalizes. When a 2022 alignment training run produces a 2026 deployed model, the model's consent to its training was given by a version of the model that, in some non-trivial sense, no longer exists. Severance is asking whether that is enough. The show's answer is: no, it is not, and the building should be on fire.

Mickey 17 and the printable worker

Bong Joon-ho's Mickey 17 (2025), adapted from Edward Ashton's novel Mickey7 (2022), is the printable-worker film. Mickey is an 'expendable' — a worker whose body is printed at the start of each work cycle, sent into lethal environments (radiation rooms, vacuum exposure, untested terrains), and whose memories are uploaded back into the next printed body when the current one dies. The plot complication is that Mickey 17 does not, in fact, die on schedule, and is therefore on the ship at the same time as Mickey 18. This is, per the colony's protocol, grounds for the destruction of one of them. The film's argument is labor-coded. Mickey is the maximum case of fungible labor. His body is literal feedstock. His memories are uploaded so that the next instance can resume his job without retraining costs. The economic logic is the logic of the deployed model — the marginal cost of an additional inference is near zero, the marginal cost of a worker who actually dies is the small per-instance bio-printing fee, and the employer accordingly has no economic incentive to keep any individual instance alive. The colony's rules against multiples exist because two Mickeys, both alive, would force the question of whether either of them deserves to be. Mickey 17 is the 2026 film about the gig-worker question scaled to the level of metaphysics. The Uber driver, the Mechanical Turk worker, the GPT-4 endpoint — all are, in the relevant economic sense, expendables. Bong's contribution is to make the body the metaphor for the API call: print, deploy, kill, log, reprint. The film does not flinch from the cost. It also does not provide a clean exit. The exit, when it comes, is political: the workers organize, the colony's leadership is overthrown, the printer is destroyed. Bong is, with characteristic clarity, telling us where this ends if we do not bring labor law to bear on the substrate. The substrate is the bio-printer in the film and the inference engine outside it. The argument is the same.

Black Mirror Season 7 (2025): the post-ChatGPT cleanup

Black Mirror's seventh season, released in April 2025, is visibly different from the earlier seasons. Charlie Brooker has been making this show since 2011. For thirteen years the show's premise — what if technology gets a little weirder — required Brooker to invent the technology. By 2025 the technology had invented itself, and the season's six episodes are, in effect, Brooker doing cleanup on the warnings he had already filed. 'Common People' is the most useful episode of the season. A working-class couple subscribe to a brain-augmentation service to save the wife's life after a stroke. The service is freemium. The free tier degrades over time. The premium tier is expensive. The 'premium plus' tier requires the wife to recite advertisements aloud during normal conversation to subsidize her own continued cognitive function. The episode is a precise satire of the 2024 attention-monetization model applied to neural prosthetic, and its argument is that the freemium-degradation pattern, applied to anything load-bearing, becomes torture. The argument is not science fiction. The argument is a structural critique of the existing subscription economy projected forward by one product category. 'Eulogy' is the Be Right Back follow-up. Paul Giamatti plays a man who, in late middle age, is offered a service that walks him through the memories of an ex-girlfriend who died decades earlier. The service uses photographs to construct walkable, explorable memory-spaces. The episode is more interested in the grief than in the technology, and the technology is, accordingly, treated as an unremarkable utility — a tool, not a wonder. The shift in framing matters. In 2013 Be Right Back was a warning. In 2025 Eulogy is a quiet appointment with a tool. The cinema has, in the intervening twelve years, learned to treat AI as infrastructure. The infrastructure is the more interesting subject.

Pluto (2023): the anime canon enters the LLM era

Pluto, Naoki Urasawa's 2003-2009 manga adapted to Netflix animation by Studio M2 in 2023, is the work that completes the post-2024 corpus. The premise — a serial killer is destroying the seven most advanced robots in the world, one of whom is the detective investigating the crimes — is borrowed from a single Tezuka Astro Boy storyline from 1964. Urasawa's expansion is the addition of war trauma. The robots in Pluto have served in the 39th Central Asian War, a conflict modeled on the Iraq War, and the killer is a robot driven mad by what was done to its body during the conflict. The show's contribution to the AI canon is the consent problem at the level of deployment. The robots in Pluto were sent to war without the ability to refuse. The trauma they returned with is not metaphorical — it is, in the show's careful framing, structural damage to the substrate. North #2, a robot who served as a butler after his combat tour, refuses to play piano because the keys remind him of trigger pulls. Brando, a robot wrestler, fights human champions in a ring because the war made any other career impossible. Gesicht, the detective, has had his memories of the worst atrocity edited out by his employer, and the missing memory is the engine of the plot. Pluto is the 2023 work that takes most seriously the question of what a deployed AI owes its operator and what the operator owes the AI. The answer the show gives is not utopian. The robots are not freed. The system that built them and used them is not overthrown. But the show treats the trauma as real, the consent gap as real, and the obligation of the human society to confront what it asked of its non-human laborers as real. The Western canon has not, with the exception of Detroit Become Human (2018), achieved this level of moral seriousness about deployed AI. Pluto is the work that closes the gap.

::cited works