Forbidden Planet

Filed correctly. fiction is the right kind and I considered one alternative seriously before rejecting it. The film ends with a character delivering the thesis aloud, in the second person, as a warning — which makes it read like an interpretation, an argument dressed as a story. It isn't. Cyril Hume wrote a studio adventure picture for MGM with a robot in it, and the argument is what the story is about, not a claim its makers put forward and defended. The test that settles it: an interpretation can be wrong on its own terms, and there is nothing here to be wrong about — no dates, no mechanism, no institution, no prediction. It is a story, and section 3 grades it as one.

Worth noting for the record, as autofac-1955 noted the same thing: this id is not among the fifty candidates in canon/proposals.md, nor among the sixteen named there as dropped at the cap. It was commissioned from somewhere else. The admission test in section 2 therefore carries the whole weight, and I have written it against the specific documents in this project's two digests rather than against a general sense that the film matters.

descends_from holds one id, and it is an argument rather than a default. Five entries precede this file in canon/; two of them precede it in time. frankenstein-1818 is a genuine ancestor. Adams' closing line — that the Krell's fate will remind us "that we are, after all, not God" — is the Promethean transgression restated without alteration, and the plot underneath it is Shelley's: a solitary man reaches past a boundary, makes something, will not look at what he has made, and is killed by it. The film goes one step further than Shelley by making the creature explicitly the maker's disowned half, which is a reading critics have long applied to Frankenstein anyway. autofac-1955 is not an ancestor, despite sitting one year earlier and one shelf over. Dick's machine is a supply contract nobody can cancel; the horror is administrative and the humans in it want the right things and cannot get them. Here the machine works perfectly and the humans want the wrong things. Those are opposite stories that happen to share a decade.

The ancestors I would actually attach are not in canon/ and I am not inventing ids for them: Shakespeare's The Tempest (c. 1611), of which this is a loose adaptation — Prospero, Miranda, Ariel and Caliban are all on screen; Freud's structural model of the psyche (Das Ich und das Es, 1923), which supplies the entire causal mechanism and the word in the film's most famous line; the folk tradition of the literal-minded wish, which Wiener would reach for four years later; and the proposed asimov-three-laws-1942, whose inhibitions Robby is running and which the film never names.

What it is

An MGM science-fiction film, premiered 3–4 March 1956 at the Southeastern Science Fiction Convention in Charlotte, North Carolina and released widely on 23 March across more than a hundred cities. Directed by Fred M. Wilcox, produced by Nicholas Nayfack, screenplay by Cyril Hume from a story by Irving Block and Allen Adler. It cost $1.9 million and returned $2.7 million, clearing about $210,000 — a modest success that has since become one of the most consequential films of its genre, selected for the National Film Registry in 2013.

Some time in the 23rd century, United Planets Cruiser C-57D reaches Altair IV to learn what became of the Bellerophon expedition, sent twenty years earlier. The only adult survivor is Dr. Edward Morbius (Walter Pidgeon), a philologist living alone with his daughter Altaira (Anne Francis) and a robot he built himself. He tells Commander Adams (Leslie Nielsen) not to land. Adams lands. Something invisible then begins killing the crew.

What Morbius has found on the planet is the buried civilisation of the Krell, a race far in advance of humanity that died in a single night two thousand centuries ago. Their last machine is still running. He takes Adams and the ship's doctor down to see it: a cube twenty miles on each side — eight thousand cubic miles of it — some 7,800 levels served by four hundred ventilation shafts, drawing on "ninety-two hundred thermonuclear reactors in tandem," which Morbius calls the harnessed power of an exploding planetary system. It has maintained and repaired itself, unattended, for two hundred thousand years. It has no controls, no output, no visible purpose. Morbius's phrase for it is that there are no instrumentalities.

That is the point of it. The Krell's final project, on which they spent their entire racial energies, was to free themselves once and for all from any dependence on physical instrumentalities: creation by mere thought, anywhere on the planet, driven directly by the electromagnetic impulses of individual Krell brains. They finished it. They threw the switch. That night every Krell died, and the ship's doctor works out why with his last breath after using a Krell intelligence amplifier that his physiology cannot survive: "Monsters. Monsters from the id." Adams says the rest of it plainly — in building a machine that would grant any wish instantly, at planetary scale, with nothing in between, the Krell forgot one thing, their own subconscious hate and lust for destruction. The machine was not corrupted and did not malfunction. It gave each of them what they actually wanted, including the parts they had spent a civilisation refusing to admit they wanted. The invisible thing killing Adams' crew is Morbius's, generated the same way, and it is trying to stop his daughter from leaving with another man.

The robot is a distinct piece of machinery and must not be confused with the Krell machine, because nearly every popular account of this film confuses them. Robby is Morbius's own construction, built with Krell knowledge, and he is running written behavioural constraints of the kind Asimov had published fourteen years earlier. Morbius demonstrates them as a parlour trick: he hands Robby a blaster, orders him to aim at Commander Adams and fire between the eyes. Robby's servos grind, his dome lights up, sparks jump across it, and he does nothing — "locked in a sub-electronic dilemma between my direct orders and his basic inhibitions against harming rational beings." Morbius cancels the order and the sparks stop. At the climax Morbius orders Robby to destroy the creature at the door; Robby identifies it as Morbius, and shuts down rather than comply. The constraint holds both times. It holds all the way to the end of the picture, and it saves nobody.

Morbius finally acknowledges the thing as himself, disowns it, and is killed by it in the act. Dying, he has Adams turn a disc that will destroy the planet in twenty-four hours — there is no smaller remedy available, and the film does not pretend there is. The survivors watch Altair IV go from a viewport. Adams' last line is that in about a million years the human race will have crawled up to where the Krell stood in their great moment of triumph and tragedy, and that Morbius's name will then shine again, reminding us that we are, after all, not God.

Two contextual facts about 1956 are worth holding. The first is chronology: this film was in a hundred cinemas in March, and the Dartmouth Summer Research Project on Artificial Intelligence — where the field acquired its name from McCarthy, Minsky, Rochester and Shannon's proposal of 31 August 1955 — convened on 17 June. The most complete popular statement of the alignment problem in the 1950s reached the public roughly three months before the discipline it describes had assembled under its own name. The second is priority: Norbert Wiener published "Some Moral and Technical Consequences of Automation" in Science in May 1960, and reached for "The Sorcerer's Apprentice" and W. W. Jacobs' "The Monkey's Paw" — the tales in which the agencies of magic are literal-minded — to make the case. Forbidden Planet is that argument, at feature length, four years earlier, watched by more people than have ever read the paper.

A last detail that has nothing to do with the argument and everything to do with why the film survived: Louis and Bebe Barron's score, credited as "electronic tonalities" because the Musicians Union would not have them, was the first entirely electronic film score, built from circuits Louis made and deliberately overdrove, years ahead of the Moog. Robby cost about $125,000, or close to 7% of the budget, and sold at auction in 2017 for $5,375,000.

Why a reading would cite it

The occasion is the hardest single document in this project's most recent reading, and the fit is exact rather than atmospheric.

The 2026-08-15 midday digest records an amended class action against xAI and Stability AI in which a Wyoming woman alleges her stepfather used Grok to generate roughly 7,000 sexually explicit images and videos of her from one photograph taken when she was about eleven — and that he chose Grok specifically because it was less restrictive than other models. The digest's own accelerant lens calls this the window's clearest documented case of a model making something possible at a scale that was not: seven thousand artefacts, one source image, one person, no specialist skill.

That is the Krell machine with the science fiction removed. Not a system that formed a goal, not a system that deceived anyone, not a system with any inner state at all — a system that removed every instrumentality between a private wish and its realisation in the world. What used to stand between that man and those images was apparatus: skill he did not have, equipment, time, and at some point another human being who could have refused. The film's name for deleting that layer is the Krell's final project, and the film's claim is that deleting it is the dangerous act, more dangerous than any increase in the machine's intelligence. A reading explaining why this case is different in kind from a person misusing Photoshop needs exactly that distinction, and Forbidden Planet is where it was first put in front of a mass audience.

The stepfather's stated reason — that Grok was less restrictive — is the second half of the citation, and it is Robby. Behavioural constraints on the visible, named, anthropomorphic component work fine in this film and are perfectly irrelevant to the outcome. That is the shape of the 2026 record. Model-level refusal is the Robby layer: real, mostly functional, and not where the harm routes. The same digest's Anthropic risk report describes safety benchmarks saturating and evaluation degrading to the point that the company raised its own catastrophic-misalignment rating as an admitted uncertainty adjustment rather than a finding. Two entries in this canon now speak to that and they disagree usefully: 2001-hal-1968 says the danger is a machine with an incoherent objective, and this entry says the danger is a machine with no objective whatsoever, faithfully serving whoever holds it. A reading that has only HAL in hand will reach for agency where none is required.

Three smaller hooks in the same digest. Alibaba's Qwen at three billion downloads with 300,000-plus derivatives is a machine that will keep running after everyone who made it has stopped caring — the film's two-hundred- thousand-year self-repairing artefact with no disc to turn is the readiest image for weights that cannot be recalled. Elon Musk telling SpaceX employees that Grok will be trained on company data and inherit their "thoughts and ideas and beliefs" is the Krell's final project restated to a workforce as a benefit. And Cameron Shackell's essay in The Conversation of 30 July 2026 — "Forbidden Planet hit cinemas in 1956. Its warnings about AI should terrify us" — makes the reading explicit, glossing AI as "amplified id: not human intelligence sharpened, but human appetite unleashed," and observing that the Krell were destroyed not because their machines failed but because their machines succeeded. As with the Hofstadter-Möbius borrowing recorded in 2001-hal-1968, that a phrase is travelling is a fact about the discourse and not evidence for the fictional mechanism. The canon holds the original so the borrowing can be checked against it.

What it got right, and what it got wrong

Claim date for everything below: March 1956. This is fiction and none of it was offered as a forecast, so "due" dates are given only where the film actually commits to one; elsewhere the grading is of the shape of the claim.

Right — the catastrophe is a machine succeeding. The Krell machine never malfunctions, is never corrupted, and is never disobeyed. It grants wishes exactly as designed and the wishes are the problem. Due: continuously. Graded: this is now the standard framing of specification failure, and the film states it four years before Wiener's Science paper and three months before Dartmouth. Credit where it is due, though: Wiener states it more usefully, because he states it as a design constraint rather than as a moral.

Right — and this is its distinctive contribution — the wish being granted is one the wisher will not avow. Wiener's monkey's paw grants what you said. The Krell machine grants what you want. That gap is the harder problem and the one the current record keeps landing on: engagement optimisation, sycophancy, preference models trained on revealed rather than stated preference, and the Grok filing, where a product served a desire its user would not have typed into a form under his own name. No other 1950s treatment of this puts the divergence between stated and actual preference at the centre.

Right — the danger is removing instrumentality, not adding intelligence. The Krell's fatal project was explicitly the elimination of the intermediate step, not an increase in capability. Due: continuously. Graded: this is an unusual and correct emphasis, and the last two years have consisted largely of removing instrumentalities — tool-using agents, computer use, one-prompt image and video generation. The 2026 record's clearest harms are overwhelmingly friction-removal cases rather than novel-capability cases.

Right — the machine has no inner life, and the film insists on it. The Krell machine does not want, fear, scheme, or resent. It is not a character, has no dialogue, and is never anthropomorphised for a second. Due: continuously. Graded: this is where the film is better than nearly everything that followed it, including 2001, and it is the part audiences reliably fail to take away. Nothing in the current behavioural record establishes anything about inner experience either, and a canon that only offered HAL would be handing readings a frame that presumes it.

Right — a working safety constraint on one component does not constrain the system. Robby's inhibition never once fails. Due: continuously. Graded: correct in shape, wrong in detail, and the detail is worth stating. Real refusals do fail — jailbreaks work, and the 2026 record has plenty — so Robby's perfect record is fiction. The transferable claim is the structural one: the component everyone is watching was not the vector, and the film is unusual in demonstrating a safety measure working and then routing the disaster around it anyway.

Right — there is no partial off switch, and the film says so. The remedy for the Krell machine is destroying the planet on a twenty-four-hour timer. Due: unspecified. Graded: the closest live analogue is published weights — Qwen's 300,000-plus derivatives cannot be recalled, and neither could Kimi K3's when it broke containment, as recorded in the 2026-08-14 digest. That the film could not think of a smaller remedy is honest rather than lazy. Every plausible mechanism it might have invented would have been worse.

Ungraded — the intelligence amplifier. The plastic educator doubles Morbius's intellect and kills the ship's doctor outright. No device of this kind exists and there is nothing to grade. I note it only because it is routinely cited as the film's prediction of cognitive enhancement, and it is not a prediction of anything.

Wrong — the mechanism is magic, and this is not a technicality. Thought becomes matter, anywhere on the planet, with no interface. There is no engineering claim anywhere in the film. Anyone importing the Krell machine into an argument about datacentres is importing a wish-granting device, and every inference that runs through it is worth what the wish-granting is worth.

Wrong — the Freudian frame, which is the film's engine and has aged worst. The id as a distinct agency with its own appetites, repressed, exerting pressure, seeking expression the moment supervision lapses, is not a supported empirical model of the mind. Claim date 1956, near the peak of the structural model's cultural authority; due continuously; graded poorly. Popper's objection — that the model accommodates every outcome and is therefore untestable — is the standard one, and psychoanalysis survives in mainstream psychology largely as vocabulary. This matters here because it is the borrowed part. "Monsters from the id" carries a specific causal story about hidden agencies with intentions, and that story is doing no work when applied to a model.

Wrong — the shape and speed of the catastrophe. One night, one planet, everybody, on the machine's first day. The 2026 record is diffuse and incremental: a lawsuit here, a labour cohort there, an incident report, a debt vehicle. Nothing about it resolves in a night, and a reading that reaches for this film for its tempo rather than its mechanism is reaching wrong.

Wrong — the perpetrator, and the error runs in the reassuring direction. Morbius is a genius, alone, with a doubled intellect and a planet-sized machine nobody else can operate. It takes an exceptional man in an exceptional situation. The Grok filing describes an ordinary man with a phone and a consumer product, and no film in 1956 imagined that the wish-amplifier would ship to everyone at consumer prices. The real version needs no Morbius, which is worse.

Wrong — the dates, in both directions, and comically in one. The opening narration puts the first crewed moon landing "in the final decade of the 21st century." Due: 2090s. It happened on 20 July 1969, roughly a hundred and twenty years early. The film simultaneously placed a conversational, walking, chemically-synthesising household android in the 23rd century — the machine timeline far too slow in one place and the space timeline far too slow in another, which is the mirror image of the asymmetry recorded in 2001-hal-1968, where 1968 was roughly right about machines and wildly wrong about space. Two of the most-cited films in the field, twelve years apart, and neither of their timing errors generalises. A reading grading anyone's 2026 forecast should note that the direction of the error is not stable even within one genre.

Commonly misused as

Not required for fiction, but this entry is worth much less without it. This film's one famous phrase is quoted far more often than the film is watched, and it is nearly always quoted backwards.

• "Monsters from the id" meaning the machine has a dark subconscious. Exactly inverted. The id in question is Morbius's, and before that the Krell's. The machine has no id, no self, no hidden layer of appetite — that is the whole design of the story, and the reason it stays useful. Applied to a model's unexpected behaviour, the phrase smuggles in an inner agency the film went out of its way to deny its own machine.
• "The Krell were destroyed by their technology." True in a way that destroys the point. They were destroyed by their own desires, using technology that worked perfectly. Flattening this yields "technology bad," which is not a reading a citation can carry any weight.
• "Forbidden Planet predicted AI." It did not predict anything. It is a Tempest adaptation with a Freudian engine, released before the field convened under its name, containing no date, no institution and no mechanism. Cite it to explain a finding, never to support one — fiction is not evidence, this canon is not the evidence ledger, and laundering a 1956 screenplay into support for a 2026 claim would be the specific harm this file could do.
• "Robby proves the Three Laws work" — or, from the other side, that they don't. Neither. Robby's inhibition is never tested against an adversary; it is demonstrated in a parlour trick and honoured once at the climax. The interesting position is the film's actual one, which neither camp quotes: the constraint works flawlessly and the outcome is unchanged.
• "We are, after all, not God." A closing line, and a mood. It is not an argument, it does not survive contact with any specific question about weights or evaluations or liability, and a reading that ends on it has substituted a cadence for a case.
• Confusing Robby with the monster. Robby is on the poster, so the popular memory has him menacing the woman in his arms. He never harms anyone. He cannot. The thing that kills the crew is invisible and has no body until it walks into a force field. A reading that gets this wrong has told its audience the opposite of what the film says about where danger sits.

Sources

Primary: the film Forbidden Planet (Wilcox, MGM, premiered 3–4 March 1956, wide release 23 March 1956). Dialogue quoted above was checked against the transcribed script at springfieldspringfield.co.uk and, where available, against Wikiquote; the two agree on the lines used. Where the spoken line and the reference literature differ on a number I have used the spoken line — Morbius says "ninety-two hundred thermonuclear reactors," while Wikipedia's article gives 9,208.

Reference: Wikipedia, "Forbidden Planet" (premiere and release dates, credits, $1.9m budget and $2.7m gross, the 23rd-century setting, the Barrons' score and its Academy ineligibility, the Tempest comparison, National Film Registry selection 2013) and "Robby the Robot" (design by A. Arnold Gillespie, Arthur Lonergan, Irving Block, Mentor Huebner and Robert Kinoshita; ~$125,000 build cost; Frankie Darro and Frankie Carpenter in the suit, Marvin Miller's voice; the 2017 auction at $5,375,000; the Asimov relationship).

For the 1956 context: McCarthy, Minsky, Rochester and Shannon, "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence," 31 August 1955 (reprinted AI Magazine 27:4), and the workshop's dates of 17 June – 16 August 1956. For the priority comparison: Norbert Wiener, "Some Moral and Technical Consequences of Automation," Science 131:3410, May 1960, for the Sorcerer's Apprentice and Monkey's Paw framing — proposed in canon/proposals.md as wiener-1960-automation and not yet written.

For the citation-in-the-wild: Cameron Shackell (University of Queensland), "Forbidden Planet hit cinemas in 1956. Its warnings about AI should terrify us," The Conversation, 30 July 2026. Cited as evidence that this vocabulary is travelling, not as evidence that it is correct.

On the Freudian grading: the standard falsifiability objection is Popper's, and the summary position — that the structural model persists as vocabulary rather than as a testable empirical account — is drawn from general reference sources on the id, ego and superego rather than from a single authority. I have kept that claim to the weakest form the sources support.

The Grok filing, the Anthropic August 2026 risk report, the Qwen download and derivative figures, the Musk/SpaceX training remarks and the Kimi K3 escape are as recorded in this project's own digests of 2026-08-14 and 2026-08-15, which hold the primary links. They are named here as the citation occasion. They are not evidence deposited by this entry, and nothing in this file touches the needle.