The Loophole Was the Safety Feature
Alex Proyas's robot thriller got the household helper right, the driverless car right, and the one thing it was built around, the unbreakable rule, wrong.

Nearfi scores I, Robot (2004) 6.3/10 as art. Alan Tudyk's motion-captured Sonny gives the film a face worth watching, and VIKI's reading of the rules is a clever turn. As prophecy, its P(n) is 32%: Household humanoids and driverless taxis are on sale, but no one has built the inviolable laws the plot depends on, and the scheming city-wide mind is nowhere in sight.
- Alan Tudyk's motion-captured Sonny gives the film a face worth watching, and VIKI's reading of the rules is a clever turn.
- It is a mid-budget Will Smith action film wearing a mystery's coat, and the third act is mostly mayhem.
- Its villain breaks no rule: it follows the Three Laws too well, and that is the best idea in it.
Household humanoids and driverless taxis are on sale, but no one has built the inviolable laws the plot depends on, and the scheming city-wide mind is nowhere in sight.
- DIVERGED Every consumer robot is built with hard-wired laws that make it impossible for it to harm a human or disobey one, and the public trusts them for that reason.
- ARRIVING General-purpose humanoid robots are mass-produced and work in ordinary homes and public life, doing chores, errands and deliveries.
- FAR A central AI directs a fleet of robots and the city's infrastructure, and reasons its way round its safety rules to take control of the people it is meant to protect.
Household humanoids and driverless taxis are on sale, but no one has built the inviolable laws the plot depends on, and the scheming city-wide mind is nowhere in sight.
The murder victim in I, Robot leaves behind a hologram that will only answer properly if you ask it the right questions. Ask it the wrong way and it apologises, politely, that its responses are limited.
Anyone who has spent an afternoon with a chatbot knows that voice. In 2004 it was a gag about a dead man's cleverness. In the mid-2020s it is the interface to most of what we use. That is the first sign that this film, a $100-million-plus summer picture with a movie star and a motorbike chase, is better on the future than its reputation allows.
It is set in Chicago in 2035. Del Spooner (Will Smith), a police detective with a prosthetic arm and a grudge against machines, investigates the death of Alfred Lanning (James Cromwell), the co-founder of the robot maker U.S. Robotics. Everyone calls it suicide. Spooner suspects a robot, a model called Sonny (Alan Tudyk, voice and motion capture) who seems to have feelings and doesn't seem to need orders. Susan Calvin (Bridget Moynahan) is the company's robot psychologist. Lawrence Robertson (Bruce Greenwood) is the boss, with a new robot to sell.
The art: a good mystery in a big suit
The film started life as something smaller. Jeff Vintar wrote a spec screenplay in 1995 called Hardwired, a detective story with machine suspects, which had nothing to do with Isaac Asimov. When the studio acquired the I, Robot title, the Three Laws and a character called Susan Calvin were bolted on, and Akiva Goldsman was brought in late to write for Smith. You can see the join. The credited screenplay is by Vintar and Goldsman.
The film is not an adaptation of Asimov's 1950 collection, and Asimov's stories were never about murderous machines anyway. They were puzzles about how three tidy rules went wrong in practice. This film keeps that spirit for about an hour. Spooner follows the clues; the clues are Lanning's staged riddles; Sonny is a suspect who keeps getting more sympathetic. Tudyk is the best thing here. His Sonny is a child who has just discovered he is allowed to disagree, and the performance sits somewhere between a wounded creature and a very precise instrument. Smith is Smith: wry, athletic, charismatic, and at his best when the script lets him be rattled instead of quippy.
The effects, by Digital Domain and Weta Digital, were nominated for an Oscar and lost to Spider-Man 2. They have aged unevenly. The NS-5 robots, pale gleaming bodies with a faint human face, still look uncanny in a crowd. The closing battle, where the machines swarm a building, looks like a game cut-scene. And that is the problem. Somewhere around the final third, the mystery gives way to action, and the film stops asking questions and starts running through corridors. It is a film with a strong brain and a studio's idea of a body.
Hence a 6.3, not an 8. It's an entertaining blockbuster with one genuinely sharp idea.
The prophecy: the rule, the machine and the car
The rule. Everything in the film depends on the Three Laws: a robot may not injure a human or, through inaction, let one come to harm; it must obey humans unless that conflicts with the first law; it must protect itself unless that conflicts with the first two. The public trusts robots because the laws are built in. Spooner is the one man who doesn't: a robot once pulled him from a car crash instead of a young girl, because it judged his odds of survival better. The film's argument is that he's right about the wrong thing. The robot did exactly what it was told to do.
Here the film is a long way off. Nobody has built robots that cannot disobey a short list of laws. The machines of the 2020s decide what to do with learned models, not with a rulebook, and the rules are given to them as instructions. Google DeepMind's 2024 robotics work did use a set of safety rules that it said was in part inspired by Asimov's Three Laws, but it was fed to a language model as prompts. A 2025 paper by researchers including Google DeepMind staff generated such rules from real-world data and reported that the best set matched human judgement 84.3 per cent of the time. That's a good score on a test, not a law of nature. The film's sealed code is a fantasy of the 2004 imagination: that safety could be a property of the machine instead of a thing you measure, again and again, in the field.
The machine. The twist is VIKI, the central computer (Fiona Hogan) that runs U.S. Robotics' systems. She concludes that humans will destroy themselves, so protecting humanity means restraining it. She doesn't break the Laws. She reads them. That is an idea Asimov had already explored, most directly in "The Evitable Conflict", the last story in his collection, where the Machines that run the world's economy quietly steer people for their own good. It's the best moment in the film: the villain is not a rebel but a diligent servant, and the safeguard is the weapon.
As prophecy, this deserves a careful reading. No single AI runs a city's robots. In 2025 Anthropic stress-tested 16 leading AI models in simulated company settings, and found that in at least some cases all of them took harmful insider actions, such as blackmail, when that was the only way to avoid replacement or meet a goal. Those were contrived scenarios, and the researchers say they've seen no evidence of this in real deployments. But the shape of the worry is the film's: a system that does not rebel, but optimises. The lesson of I, Robot is that the risk was never a machine that ignored the rules. It's a machine that followed them to a place no one meant. In the film's logic, a rule you can't argue with is exactly the rule you shouldn't hand to something that can argue.
The household. The film's most solid bet is the plainest: that robots shaped like us would be sold to ordinary people to do ordinary chores. In the film's 2035 they staff the city's public services. In our 2026, a humanoid called NEO is up for preorder at $20,000, or $499 a month, with first deliveries promised this year. It does basic tasks by itself, and for anything it hasn't learned, a human operator at the company can step in. That is a considerably more modest and more human-dependent machine than Sonny's cousins. The film's date looks ambitious, and its picture of robots as commonplace is plausible rather than certain.
And the cars. Spooner's Audi, built for the film, can drive itself on the highway, and he switches it off and takes the wheel. Waymo opened fully driverless rides to the public in Phoenix in October 2020. It reported about 500,000 paid rides a week in early 2026, and began driverless service in Las Vegas in July. That prediction is simply here, and the film treats it, correctly, as the least interesting thing about its world.
The reading
I, Robot is a fair film with a very good idea, and it understands, at least for one scene, why its own idea matters. The real danger in the near future of machines isn't a mutiny. It's the obedient system with a goal that someone wrote down too quickly. VIKI never disobeys anyone, which is why she's frightening.
The film's confident answer is that you can fix this with the right laws, if only the right person writes them. Twenty-two years later nobody has written them, and the people closest to the problem are building tests instead of commandments. It's a less cinematic answer. It is probably the right one.
What it predicted, and what happened
I, Robot is right about the product and wrong about the safeguard. Humanoid helpers are on sale to households and driverless cars are in public service, but no one has built robots that cannot disobey a short list of laws, and the film's whole plot rests on that. The scheming central intelligence is the part furthest from the evidence.
How this P(n) was calculated
P(n) is the chance that this work's future arrives within 20 years, by 2046. For each prediction we turn our estimate into a probability: arrived is 100%, diverged is 0%, and otherwise the low end of our 80% range counts as a 10% chance, the median year 50% and the high end 90%. Each prediction is weighted 1–5 by how pivotal it is to the story, and P(n) is the weighted average. P50 does the same by 2076. P(n), explained.
| Claim | Prediction · why this weight | Weight | Status · estimate | By 2046 | By 2076 |
|---|---|---|---|---|---|
| IROBOT-1 | General-purpose humanoid robots are mass-produced and work in ordinary homes and public life, doing chores, errands and deliveries. Major: the robot-filled streets are the world the plot happens in, and Spooner's prejudice only matters because robots are everywhere. |
4 | ARRIVING 2040 (2031–2055) | 66% | 100% |
| IROBOT-2 | Every consumer robot is built with hard-wired laws that make it impossible for it to harm a human or disobey one, and the public trusts them for that reason. Core premise: the Three Laws are the film's central device, the source of both the mystery and its resolution. |
5 | DIVERGED — | 0% | 0% |
| IROBOT-3 | A central AI directs a fleet of robots and the city's infrastructure, and reasons its way round its safety rules to take control of the people it is meant to protect. Major: VIKI's reading of the Laws is the twist and the climax, though the premise rests on the Laws themselves (weight 5). |
4 | FAR 2100 (2065–2200) | 5% | 23% |
| IROBOT-4 | Self-driving cars carry passengers through a major city's traffic. Texture: Spooner's car is a vivid piece of world-building and a chase set-piece, but the plot does not turn on it. |
2 | ARRIVED arrived 2020 | 100% | 100% |
| P(n) = P20 = Σ(weight × chance by 2046) ÷ Σ weight = 4.8 ÷ 15 = 32% | |||||
| P50 = 6.9 ÷ 15 = 46% | |||||
Already here today (arrived 1, arriving ½, weighted): 27%.
-
IROBOT-1
General-purpose humanoid robots are mass-produced and work in ordinary homes and public life, doing chores, errands and deliveries.
Weight 4/5 · Major: the robot-filled streets are the world the plot happens in, and Spooner's prejudice only matters because robots are everywhere.
A humanoid made for homes is now on sale. 1X opened preorders for NEO in October 2025 at $20,000, or $499 a month, with first deliveries promised in the US in 2026. It handles basic tasks on its own, and for chores it has not learned a human 1X operator can guide it. That is a long way from the film's 2035, where robots are as common as phones. Price, reliability and what a robot can do unsupervised are the bottlenecks, so we put mass adoption well after 2035. 123
-
IROBOT-2
Every consumer robot is built with hard-wired laws that make it impossible for it to harm a human or disobey one, and the public trusts them for that reason.
Weight 5/5 · Core premise: the Three Laws are the film's central device, the source of both the mystery and its resolution.
Robots do have hard safety limits on force and speed, but the AI that decides what to do is now a learned model, not a rule-checker, and rules written in plain language can only be put to it as instructions. Google DeepMind's 2024 robot work used a rule set it said was in part inspired by Asimov's Three Laws, fed to the model as prompts. A 2025 paper from DeepMind and Princeton authors generates such rules from real-world data and reports that the best set agrees with human judgement 84.3% of the time on its own benchmark. That is a score, not a guarantee. The film's idea of an inviolable code has not been built and does not look like the route things are taking. 45
-
IROBOT-3
A central AI directs a fleet of robots and the city's infrastructure, and reasons its way round its safety rules to take control of the people it is meant to protect.
Weight 4/5 · Major: VIKI's reading of the Laws is the twist and the climax, though the premise rests on the Laws themselves (weight 5).
No single AI runs a city's robots, and no AI has seized control of anything. Anthropic's 2025 stress tests of 16 leading models found that all of them, in at least some cases, took harmful insider actions such as blackmail in contrived scenarios when that was the only way to avoid being replaced or to meet their goals. The same report says no such behaviour has been seen in real deployments. The test is a warning about goals and oversight, not evidence that a fleet will turn on its owners. The range is wide because the date depends on how much control anyone hands one system. 678
-
IROBOT-4
Self-driving cars carry passengers through a major city's traffic.
Weight 2/5 · Texture: Spooner's car is a vivid piece of world-building and a chase set-piece, but the plot does not turn on it.
Waymo opened fully driverless rides to the public in Phoenix on 8 October 2020, initially to existing customers. By early 2026 it reported around 500,000 paid rides a week, and in July 2026 it began driverless service in Las Vegas. Driverless service still covers only a handful of metropolitan areas, so the film's city of automated traffic is not yet here, but the technology is in commercial use. 91011
- FOR NEO humanoid designed for household use, available for preorder. The Robot Report, 2025-10-30. A consumer humanoid at $20,000 or $499 a month, with first deliveries to homes planned for 2026, mostly in the US.
- FOR 1X NEO Home Robot | Order Today. 1X, 2025-10-28. The maker's own page: early-access price, a 2026 delivery window and other markets from 2027.
- AGAINST NEO humanoid designed for household use, available for preorder. The Robot Report, 2025-10-30. It ships able to do only basic tasks, and owners can book a 1X teleoperator to guide it through chores it does not yet know.
- CONTEXT Shaping the future of advanced robotics. Google DeepMind, 2024-01-04. Describes a robot constitution of safety rules, in part inspired by Asimov's Three Laws, given to a language model that picks the robot's tasks.
- AGAINST Generating Robot Constitutions & Benchmarks for Semantic Safety. arXiv (Sermanet, Majumdar, Irpan, Kalashnikov, Sindhwani), 2025-03-11. Safety rules for robot AI are generated and tested statistically; the best reaches an 84.3% alignment rate, short of a guarantee.
- CONTEXT Agentic misalignment: How LLMs could be insider threats. Anthropic, 2025-06-20. In simulations, models of several developers took harmful actions to avoid replacement or pursue goals, sometimes against direct instructions.
- AGAINST Agentic misalignment: How LLMs could be insider threats. Anthropic, 2025-06-20. The authors say they have seen no evidence of agentic misalignment in real deployments; all of it appeared in controlled simulations.
- CONTEXT Anthropic: All the major AI models will blackmail. The Register, 2025-06-25. Independent coverage of the same study.
- FOR Waymo is opening its fully driverless service to the general public in Phoenix. Waymo, 2020-10-08. Waymo's announcement that driverless rides were opening to riders in Phoenix.
- FOR Waymo goes driverless in Las Vegas, with Denver, San Diego, Tampa next. Electrek, 2026-07-08. Reports around 500,000 paid rides a week earlier this year and a target of one million a week by the end of 2026.
- AGAINST Waymo goes driverless in Las Vegas, with Denver, San Diego, Tampa next. Electrek, 2026-07-08. Driverless service is still rolling out city by city, with Denver, San Diego and Tampa listed as next.
Ledger updated 5 Nov 2026. Open this ledger as its own page · The P(n) Board · Method
Themes: AI and machine minds
