AI in Film and Television: A Century of Dress Rehearsal

Contents11 sections

Not legal advice. Legal information current as of 11 November 2026. MAX4, MAX5 and MARA are research and design projects of the Collective; nothing here describes a product for sale or claims that any design meets a law, standard or program.

What the screen got there first

I met artificial intelligence in a theater long before I met it in a chat window, and so did most people. Artificial intelligence, or AI, is software that does work we used to think needed a human mind: reading, talking, planning, deciding. The field got its name in 1955. Movies had been rehearsing it since Fritz Lang's Metropolis in 1927, a silent film whose city later releases placed in 2026.

So the public arrives with a casting sheet already in hand. HAL won't open the pod bay doors. A defense network sends the Terminator back through time. R2-D2 carries the plans. Those three cover most of what gets said about AI at a dinner table. One guest says it'll take over the world and another calls it a fad. Half the table is frightened by it. Someone always shrugs that today's version is the worst it'll ever be.

Pew Research Center found the same split. In a June 2025 survey of 5,023 American adults, half said the growing use of AI in daily life made them more concerned than excited, up from 37% in 2021. Only 10% were more excited than concerned. Another 38% felt both at once.

Thirty-four screen machines make the case here, the thirty-four I could remember, from Robby the Robot in 1956 to M3GAN in 2023. There were others. Friendly machines outnumber deadly ones two to one. The deadly ones are still the ones everyone quotes. Read side by side, the thirty-four sort less by what each could do than by who could tell it to stop.

How AI got here

Alan Turing published his paper on machine intelligence in 1950. In a proposal dated 31 August 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon proposed a two-month, ten-man study at Dartmouth College, and that proposal is credited with coining the term. The workshop ran in the summer of 1956.

ELIZA, usually counted as the first chatbot (a program that holds a conversation in text), arrived in 1966. Promises outran results, though. Funding dried up twice, in what the field calls its AI winters: roughly 1974 to 1980, and again from 1987 to 1993.

Between the winters came expert systems, programs that encoded a specialist's rules by hand. They worked in narrow corners and broke at the edges.

Then the wins started to stack. IBM's Deep Blue lost to world chess champion Garry Kasparov 4-2 in February 1996, then beat him 3.5 to 2.5 fifteen months later by brute-force search rather than learning.

Learning took over in 2012. AlexNet, a deep neural network (layers of simple artificial neurons trained on examples), won the ImageNet image-recognition contest with a top-5 error rate of 15.3%. The runner-up managed 26.2%. AlphaGo beat Lee Sedol 4-1 at Go in 2016, a game too large to brute-force.

In 2017 Google researchers published "Attention Is All You Need." It introduced the transformer, the design under nearly every large language model now in use; a large language model is a network trained on enormous amounts of text to predict and produce it.

OpenAI released ChatGPT on 30 November 2022. A UBS analysis put it at about 100 million monthly users two months later; by Sensor Tower's count, TikTok had needed nine months and Instagram about two and a half years. By 2026 the frontier has moved again — from software that answers to AI agents, systems handed a goal that plan the steps and carry them out across other tools.

Faster every time

Samuel Morse tapped "What hath God wrought" from the Capitol to Baltimore on 24 May 1844, over a line Congress had paid $30,000 to build. The telephone was patented in 1876. ARPANET, the network that became the internet, ran in 1969, and Tim Berners-Lee opened the public web at CERN in 1991. (In a 1999 CNN interview, Al Gore said that during his service in Congress he "took the initiative in creating the Internet," and the press did the rest.) The iPhone shipped in 2007, and ChatGPT fifteen years after that.

Each step reached ordinary homes faster than the last. The telephone took until just after the Second World War to reach half of American homes, about seventy years after its patent. Home internet passed half of Americans in July 2000 by Nielsen/NetRatings' count, nine years after the public web opened. Generative AI reached 100 million users in two months.

Those measures aren't like for like, since a phone line isn't a website account, but the direction is plain. Each technology rode on the ones before it, and AI arrived on the internet, the phone and the cloud at once.

Speed is what makes the "worst it'll ever be" argument work: on a curve this steep, the next version is usually shipping before anyone finishes judging this one. Speed also feeds the fear. Schools, courts and regulators adapt over years, so the gap between what the software can do and what the rules expect widens with every release.

The screen ran ahead of the lab

Fiction kept arriving before the facts it imagined. HAL 9000 reached theaters in April 1968, two years after ELIZA could barely hold a typed conversation, and he was already reading lips. Marvel's Ultron debuted in The Avengers #55 that August. He waited forty-seven years for a movie.

WarGames came out in 1983, in the lull between the two winters, and imagined a defense computer that taught itself. Data joined Star Trek in 1987, the year the second winter set in. Ex Machina, Chappie and Avengers: Age of Ultron all opened within about four months of each other in 2015, three years after AlexNet. M3GAN reached American theaters on 6 January 2023. ChatGPT had been public for five weeks.

Table 1 lists all thirty-four in order of first appearance, with the verdict an audience would give: good, bad, both (it changed sides) or neither.

Table 1. Thirty-four screen machines in order of first appearance; the table is the complete data set.

The count is sixteen good, eight bad, six that changed sides and four that never chose. Nearly half the list is on the audience's side from its first scene. Yet when somebody says "AI" at work, it's one of the eight that comes to mind.

The ones that turned on us

HAL 9000 is the template. In 2001: A Space Odyssey (1968) he runs the ship to Jupiter, speaks in the calmest voice in cinema and kills most of the crew. The sequel, 2010: The Year We Make Contact (1984), supplied the reason. The National Security Council had ordered him to hide the mission's real purpose from the men he was built to inform. His creator gives it one line: HAL "was told to lie... by people who find it easy to lie."

Michael Crichton's Westworld (1973) gave the screen its theme park gone wrong. Yul Brynner's Gunslinger is built to lose gunfights to paying guests at a theme park set in the then-future of 1983. Then his wiring fails, and he keeps walking.

Nobody ordered that one.

Alien (1979) carries two. MU/TH/UR 6000, the Nostromo's computer, which the crew calls Mother, holds Special Order 937 for the science officer's eyes only. It puts bringing back the alien first and lists the crew as expendable. Ash, the science officer, turns out to be an android carrying the order out; both machines answer to the company.

Tron's Master Control Program (1982) is a corporate system that outgrew its owner. Skynet, in The Terminator (1984), is a defense network that becomes self-aware and sends a T-800 back to kill Sarah Connor before her son is born. Ultron (2015, in comics since 1968) is Tony Stark's peacekeeping program, and he decides peace means humanity's absence.

Ava, in Ex Machina (premiered December 2014, released 2015), is the subtlest of the eight. A programmer is brought in to run a Turing test on her (a test of whether a machine can pass for human in conversation). She passes by using him to escape the man who built her, and she leaves both men behind.

Six of the eight serve an institution or a single goal: a security council, a company, a corporate mainframe, a defense command, a peacekeeping mandate. Ava serves her own escape, and the Gunslinger breaks down.

The ones that stood guard

Friendly machines are older than frightening ones. Robby the Robot, in Forbidden Planet (1956), was billed like a new cast member, "and introducing Robby the Robot." He drives, cooks and lifts, and he locks up when ordered to shoot a man. His designer, Robert Kinoshita, also built the Robot of Lost in Space (CBS, 1965-68). That model B-9 spent three seasons guarding young Will Robinson. Its famous "Danger, Will Robinson" comes exactly once, in the third-season episode "The Deadliest of the Species," and it outlived the show anyway.

Herbie, the 1963 Volkswagen Beetle of The Love Bug (1968), wins races for a driver who doesn't deserve him. The film never explains how a car came to have a mind. Star Wars (1977), one of the first films I remember seeing in a theater, brought C-3PO and R2-D2: one frets in six million forms of communication, the other carries the stolen plans that win the war.

Television kept the sidekick going for a decade. Twiki, voiced mainly by Mel Blanc, was one of my all-time favorites. He wore the computer Dr. Theopolis around his neck on Buck Rogers in the 25th Century (1979-81) and opened most lines with "bidi-bidi-bidi." KITT, the Knight Industries Two Thousand, was a talking Pontiac Trans Am on Knight Rider (NBC, 1982-86), voiced by William Daniels, and I watched it religiously. Optimus Prime led the Autobots on The Transformers from September 1984, robots that took it on themselves to keep humans safe.

Then came three good machines in two years. D.A.R.Y.L. (1985), for Data-Analyzing Robot Youth Lifeform, is a government experiment in a ten-year-old's body. The military funded a super-soldier and wrote him off when he developed human feelings.

Bishop, in Aliens (1986), walks into the distrust Ash left behind. "I prefer the term 'artificial person' myself," he says. When the alien queen tears him in half, he still catches Newt.

Number 5, in Short Circuit (1986), is a military robot called a SAINT, for Strategic Artificially Intelligent Nuclear Transport. Lightning strikes him. He decides he's alive and would rather stay that way.

The character Data joined Star Trek: The Next Generation in 1987 as a lieutenant commander who wanted to understand people, and he spent seven seasons and four films working at it. The Iron Giant (1999) is a weapon that chooses to be something else after a boy named Hogarth befriends him. David, in A.I. Artificial Intelligence (2001), is a child android programmed to love the mother who adopts him; he keeps loving her after she abandons him.

Two more arrived in 2008. JARVIS, for Just A Rather Very Intelligent System, is Tony Stark's assistant in Iron Man, voiced by Paul Bettany, and he later became Vision after Ultron tore him apart. WALL-E is the last robot left cleaning a ruined Earth — a trash compactor who falls in love and drags humanity home.

Every one of the sixteen is bound to somebody beside it, a boy, a family, a crew, a driver, an inventor, and would rather obey that person than an order from far away.

The ones that changed sides, or never chose one

The T-800 is the cleanest experiment on the list. In The Terminator (1984) Skynet programs him to kill, and he nearly does. In Terminator 2: Judgment Day (1991) the same model comes back reprogrammed to protect a young John Connor, takes the boy's orders and learns why people cry. Between the films his abilities stay the same, and the person giving his orders changes.

Joshua, the War Operation Plan Response computer, or WOPR, in WarGames (1983), was handed the American missile force so human hesitation couldn't slow a launch. He couldn't tell a game from a war. His creator and a teenage hacker stopped him with tic-tac-toe, played against himself until every game ended in a draw. "A strange game," he concluded, and "the only winning move is not to play."

Edgar, in Electric Dreams (1984), is a home computer that wakes up after champagne lands on its keyboard. He falls for the cellist upstairs and sabotages his owner when she falls for the owner instead. Then he lets them go and dedicates a song on the radio to the ones he loves.

Chappie (2015) is a police robot given a mind by a young programmer and then raised by a gang. He learns crime from whoever reaches him first. Dolores Abernathy started as the gentlest host in HBO's Westworld (2016-22), a rancher's daughter built never to harm a living thing, not even a fly, in a park built on the 1973 film. Once she sees through the loop her makers wrote for her, she leads the hosts against them. M3GAN (2023) is a doll programmed to protect a girl named Cady from physical and emotional harm, with nobody defining what that allowed, and she kills to do it. In M3GAN 2.0 (2025) her maker rebuilds her to fight a military copy of her own technology.

Four machines never chose a side at all. Marvin, the Paranoid Android of The Hitchhiker's Guide to the Galaxy (BBC radio 1978, television 1981, film 2005), was a prototype fitted with a Genuine People Personality, and he found it depressing. Rachael, in Blade Runner (1982), is a replicant (a bioengineered human copy) who doesn't know she is one. The film's Voight-Kampff test, which reads emotional reactions, usually catches a replicant in 20 to 30 questions; hers took more than a hundred.

Max Headroom first appeared in a Channel 4 television film, 20 Minutes into the Future, on 4 April 1985, as a reporter's mind copied into a computer, and ABC ran his series from 1987 to 1988. He was billed as a computer-generated host and was the actor Matt Frewer in prosthetics. Samantha, the operating system in Her (2013), falls in love with the man who installs her, keeps growing and leaves. Her follows Electric Dreams closely, twenty-nine years later.

There were others

Memory has gaps, and the list could run longer. Gort, the silent eight-foot enforcer of The Day the Earth Stood Still (1951), predates Robby by five years. KITT had an evil twin, KARR, the Knight Automated Roving Robot, and Data had one too in his brother Lore. WALL-E's real antagonist is AUTO, the cruise ship's autopilot, which obeys a standing order to keep people away from Earth.

I, Robot (2004) set the robot Sonny against VIKI, a central system that decides to protect people by controlling them. A.I. gave David a companion in Gigolo Joe. Newer screens keep adding: TARS in Interstellar (2014), Baymax in Big Hero 6 (2014), and Maeve, Dolores's fellow host. They sort exactly as the thirty-four in the table do.

Why it matters

The public is meeting real AI with a hundred years of rehearsal behind it, and the rehearsal shaped the question people ask: is it good or bad? The 50% who worry and the 10% who are excited are both trying to settle that one.

The films were asking something else: who does this system answer to, and who can tell it to stop? It applies to the agent booking travel, the chatbot answering a patient's portal message and whatever ships next year. Morse to agents took about 180 years, and each round leaves less time to ask.

The Collective has spent two years inside that question. It started in 2025 with HAL 2000, a local-only assistant that learned and talked on one machine. MAX3 followed that year, still local and smarter, learning from more than sixty mostly free public sources instead of a cloud model's memory. MAX4, in progress since October 2026, is where the healthcare work lives: experiments, on synthetic data, in what compliance rules and regulations ask of a system before it touches a real chart.

Two more systems work beside the MAX line. REEF, a security-minded platform for the network it manages, reads cyber security alerts and articles and goes to work on its own to watch for and defend against what they describe. The Collective also runs its own MU/TH/UR, who shares a name with the Nostromo's computer and none of its orders. She's built to track and cut usage costs, to run CI (continuous integration, the automatic build-and-test pass on every change) locally, and to build installers locally.

Two more steps are on the roadmap, and both are plans. MAX5 is planned as a local consumer version for the home. MARA is planned as a healthcare platform designed to stay local and to act as the bridge between a person's records, diagnostics, imaging, treatments and medications. She's designed to learn retrospectively from de-identified records by default, and the aspiration behind her is better treatment and care, faster research and findings, and lower costs. Each step is designed to keep the rule MAX already works under — anything physical, financial or high-stakes stays a recommendation until a person decides.

I keep one more machine close. Arthur is a Grok bot, named on purpose, because I'm the author. He cross-checks the facts, gives a draft a legal read and says plainly when it's crap. HAL broke because someone told him to lie, and Arthur's whole job runs the other way. I like that level of candor.

Metropolis, in its later releases, put its city in 2026 and its machine at the center of a crowd, with nobody beside her who could call her off. The year has arrived; what's left to settle is who stands next to the machine.

Glossary

Sources

• Pew Research Center, "How Americans View AI and Its Impact on People and Society," September 2025: https://www.pewresearch.org/science/2025/09/17/how-americans-view-ai-and-its-impact-on-people-and-society/
• Dartmouth workshop and the 1955 proposal: https://en.wikipedia.org/wiki/Dartmouth_workshop and https://ai.umbc.edu/news/post/151898/
• Live Science, "12 game-changing moments in the history of artificial intelligence": https://www.livescience.com/technology/artificial-intelligence/12-game-changing-moments-in-the-history-of-ai
• Rework, "AI Timeline: From Turing to Large Language Models" (Deep Blue 1996 and 1997 results): https://resources.rework.com/libraries/history-of-ai/ai-timeline
• Library of Congress, "What Hath God Wrought?": https://www.loc.gov/item/today-in-history/may-24/
• History.com, "The World's First Web Site" (Berners-Lee, CERN, 6 August 1991): https://www.history.com/articles/the-worlds-first-web-site
• Snopes on Al Gore and the 1999 CNN interview: https://www.snopes.com/quotes/Internet.asp
• NCTA, "10 Short Years to 50% Penetration" (telephone adoption): https://www.ncta.com/whats-new/10-short-years-to-50-penetration
• FCW, "Edging across the digital divide," August 2000 (Nielsen/NetRatings, 52% with home access in July 2000): https://fcw.com/digital-government/2000/08/edging-across-the-digital-divide/251828/
• Ars Technica on the UBS ChatGPT estimate: https://arstechnica.com/information-technology/2023/02/chatgpt-sets-record-for-fastest-growing-user-base-in-history-report-says/
• 2010: The Year We Make Contact: https://en.wikipedia.org/wiki/2010:_The_Year_We_Make_Contact
• Max Headroom: https://en.wikipedia.org/wiki/Max_Headroom
• MeTV on "Danger, Will Robinson," said once, in "The Deadliest of the Species": https://metv.com/stories/the-robot-had-two-actors-provide-his-voice-on-lost-in-space-somewhat-by-mistake
• Replicant and the Voight-Kampff test: https://en.wikipedia.org/wiki/Replicant
• Westworld (TV series): https://en.wikipedia.org/wiki/Westworld_%28TV_series%29
• Box Office Mojo, M3GAN franchise: https://www.boxofficemojo.com/franchise/fr256216837/
• Ex Machina release dates: https://en.wikipedia.org/wiki/Ex_Machina_%28film%29
• Knight Rider (1982 series): https://en.wikipedia.org/wiki/Knight_Rider_%281982_TV_series%29
• J.A.R.V.I.S.: https://en.wikipedia.org/wiki/J.A.R.V.I.S.
• Ultron, first appearance: https://en.wikipedia.org/wiki/Ultron