The Number Is All You Get
A network procedural about a billionaire, an ex-spook and a government machine that sees everything got the surveillance right before the leaks did, and got the restraint wrong.
Nearfi scores Person of Interest (2011) 8.1/10 as art. By its third season it had become one of the most patient and best-plotted dramas about artificial intelligence ever made for broadcast television. As prophecy, its P(n) is 84%: Cameras that name us, contractors that join up the records and police risk scores are all here; the machine that does it all, and holds back, is not.
- By its third season it had become one of the most patient and best-plotted dramas about artificial intelligence ever made for broadcast television.
- The first season is a case-of-the-week procedural with a brilliant idea stuck inside it, and the show takes a while to notice.
- Its Machine gives out only a Social Security number, a privacy-by-design restraint that no real surveillance system has ever shown.
Cameras that name us, contractors that join up the records and police risk scores are all here; the machine that does it all, and holds back, is not.
- ARRIVING A state system fuses camera, phone and email data and automatically flags named individuals as likely to be involved in violent crime.
- NEAR A single general-purpose AI, able to reason and act across every domain at or above human level, runs continuously and is restrained by what its makers chose to build in.
- ARRIVED Cameras across a city identify people by their faces in real time and match them against watchlists.
Cameras that name us, contractors that join up the records and police risk scores are all here; the machine that does it all, and holds back, is not.
The first thing to understand about the Machine is how little it tells you.
In Person of Interest it watches everything. Cameras, phones, email, the lot. And when it has something to say, it says a nine-digit number. A Social Security number, nothing more. It might belong to the victim or the killer, and the Machine won't say which. Harold Finch (Michael Emerson), the reclusive billionaire software engineer who built it for the US government after 9/11, set it up that way on purpose. The government only wanted the numbers that meant terrorism. Everything else, the "irrelevant" murders on ordinary streets, it threw away. Finch couldn't bear that. So he hired John Reese (Jim Caviezel), a former Special Forces soldier and CIA operative, to go and find out what each number meant.
That's the whole show, and it ran on CBS for five seasons and 103 episodes, from September 2011 to June 2016. It was created by Jonathan Nolan. It's set in New York.
The art: a procedural with a god in the walls
Be fair about the start. The first season is a good premise inside a conventional frame. A number comes in, Reese and Finch work out who it is, and by the end of the hour somebody has been rescued or arrested. Joss Carter (Taraji P. Henson), an NYPD detective, and Lionel Fusco (Kevin Chapman), a reformed corrupt one, are folded into the routine. It's solidly made and sometimes dull. If it had stayed there, we'd remember it as a decent crime show with a nice hook.
It didn't stay there. The Machine turns out to be a character, then the main character. Amy Acker's Root becomes its "analog interface", the human hands it can't have. Sarah Shahi's Shaw, a former Marine and Army ISA assassin, joins the team. A rival machine called Samaritan, run by Decima Technologies through John Greer (John Nolan), arrives with none of the restraints Finch built in. The show stops being about who's in danger this week and becomes a war between two ideas of what a thinking machine should be allowed to do.
Michael Emerson is the reason it works. His Finch is fussy, limping and evasive, a man who made something extraordinary and spends every day worried about what it might do. Caviezel is deliberately blank and does it well. The score is by Ramin Djawadi, and the season-four episode "If-Then-Else", which plays out inside the Machine's simulations of a siege, earned something close to unanimous praise. It's the best argument that a network show can be formally daring.
Carter's death is the point where the show stops being safe. The late seasons also show their age: the fifth is a rushed thirteen episodes, with ratings that fell from a premiere of about 7.35 million to a finale of about 6.5 million. The finale, "Return 0", closes the war cleanly but too fast. A sixth season would have helped. That's the gap between 8 and 9.
The prophecy: three bets and a missing restraint
Person of Interest arrived in September 2011. The Snowden revelations about PRISM, which showed the NSA collecting stored internet communications from nine US technology companies, were published on 6 June 2013. The show was about twenty months ahead of the news. It never asked whether the state would watch us. It assumed it would, and asked what the people doing the watching owed everyone else.
The cameras. The Machine reads every face on every feed. In the mid-2020s this stopped being a fantasy and turned into procurement. Between October 2025 and March 2026 the Metropolitan Police ran a six-month pilot with permanent live facial recognition cameras on lampposts in Croydon. They scanned about 470,000 people, made 173 arrests and logged one false alert, according to the Met's own figures. A High Court challenge failed in April. Clearview AI, founded in New York in 2017, said in 2021 that it held more than ten billion scraped images, and the Dutch data regulator fined it €30.5 million in 2024. Nobody needs a Machine for this. A procurement office will do.
The joined-up records. The Machine's real trick is that it connects everything. The real-world version is a company. Palantir's Gotham platform, released in 2008, has been ICE's case management system since 2014, and ICE is estimated to have paid about $51.8 million for it by 2019. In May 2026 London's mayor blocked a £50 million Met contract with Palantir over a procurement breach and the risk of dependency on one supplier. The Machine is a ghost story. Gotham is a licence agreement.
The prediction. This is the one the show needed most, and the real world has been the slowest to deliver. Chicago police ran a Strategic Subject List that scored people on the likelihood of being a victim or offender in a shooting. It was shut down on 1 November 2019 when the grant money ran out, and the city's inspector general later criticised how it was run. In Europe the answer is now in statute: since February 2025 the AI Act has banned systems that predict whether a person will commit a crime based solely on profiling or personality traits. The Machine's central function is now banned across the European Union.
The restraint. Here the show misses most, and misses in an interesting way. Finch's Machine is trustworthy because one man chose to make it so. It gives out a number, and it's built to forget. Samaritan is what you get when someone makes the other choice. The series locates the decision in one workshop and one conscience. Real AI is built by large labs in competition with each other, and the choice about restraint is made over and over, by many people, under deadline. In a 2024 survey of 2,778 researchers who had published at top AI venues, the aggregate forecast put even odds on machines outperforming humans at every task by 2047. The people building that future are making Finch's choice and Greer's choice, repeatedly, in a market.
The reading
Person of Interest gets better as it goes, and its prophecy gets better with it. It understood early that surveillance wouldn't need a villain, only a contract and a brief. It understood that the dangerous part wasn't seeing, but deciding who matters. And it understood, which almost no television has, that the interesting question about a powerful AI is not whether it's good but who taught it and what they left out.
What it got wrong is the comfort. The Machine only tells you a number. Our systems tell you more, and sell it on. Somewhere a database is deciding which of us counts, and the show's quiet answer is that everyone does, to someone.
What it predicted, and what happened
Person of Interest is most prescient about the plumbing: camera networks that name people, contractors that join up state records, and surveillance run as procurement. It is further off on its machine, a single AI that predicts violence from everything it sees and restrains itself, which has no real counterpart.
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 |
|---|---|---|---|---|---|
| POI-1 | A state system fuses camera, phone and email data and automatically flags named individuals as likely to be involved in violent crime. Core premise: the Machine's numbers are the engine of every episode. |
5 | ARRIVING 2032 (2028–2045) | 91% | 100% |
| POI-2 | Cameras across a city identify people by their faces in real time and match them against watchlists. Supporting: the Machine's eyes are everywhere, but the plot turns on what it concludes, not how it sees. |
3 | ARRIVED arrived 2020 | 100% | 100% |
| POI-3 | A private contractor's data-fusion platform joins up state records and becomes the backbone of government and police analysis. Supporting: Decima and Samaritan are the show's villains, and a company is the channel for state power. |
3 | ARRIVED arrived 2014 | 100% | 100% |
| POI-4 | A single general-purpose AI, able to reason and act across every domain at or above human level, runs continuously and is restrained by what its makers chose to build in. Major: the Machine as a mind, and the Samaritan rivalry, drive the later seasons and the show's central theme. |
4 | NEAR 2045 (2032–2075) | 51% | 90% |
| P(n) = P20 = Σ(weight × chance by 2046) ÷ Σ weight = 12.6 ÷ 15 = 84% | |||||
| P50 = 14.6 ÷ 15 = 97% | |||||
Already here today (arrived 1, arriving ½, weighted): 57%.
-
POI-1
A state system fuses camera, phone and email data and automatically flags named individuals as likely to be involved in violent crime.
Weight 5/5 · Core premise: the Machine's numbers are the engine of every episode.
Pieces exist. Section 702 collection, revealed in 2013, gathers stored communications at scale, and Chicago police scored people on their likelihood of being party to a shooting until November 2019. No known system fuses everything into per-person violent-crime alerts. The EU has banned individual crime prediction based solely on profiling, and Chicago's programme was ended. The date allows for jurisdictions that have not banned it. 123
-
POI-2
Cameras across a city identify people by their faces in real time and match them against watchlists.
Weight 3/5 · Supporting: the Machine's eyes are everywhere, but the plot turns on what it concludes, not how it sees.
London's Metropolitan Police has used live facial recognition since 2016 trials, planned regular use from 2020, and ran permanent lamppost cameras in Croydon in 2025 and 2026. Clearview AI sells face search across a scraped database to police. Coverage is not city-wide, accuracy disputes continue, and the Met's own figures come from the force. 456
-
POI-3
A private contractor's data-fusion platform joins up state records and becomes the backbone of government and police analysis.
Weight 3/5 · Supporting: Decima and Samaritan are the show's villains, and a company is the channel for state power.
Palantir's Gotham has been ICE's case management platform since 2014, ingesting records from other DHS databases, and ICE was estimated to have paid about $51.8 million for it by 2019. Other states are following, though not without resistance: in May 2026 London's mayor blocked a £50 million Met contract. 789
-
POI-4
A single general-purpose AI, able to reason and act across every domain at or above human level, runs continuously and is restrained by what its makers chose to build in.
Weight 4/5 · Major: the Machine as a mind, and the Samaritan rivalry, drive the later seasons and the show's central theme.
Large AI systems now work across many domains, but none is general in the show's sense, and research shows reasoning models failing as problems grow more complex. In a 2024 survey of 2,778 researchers the aggregate forecast gave even odds of machines beating humans at every task by 2047. The show's single wise builder has no counterpart: restraint is decided by many labs under competition. The range is wide because the forecast is. 1011
- FOR PRISM (surveillance program). Wikipedia, 2013-06-06. Documents published on 6 June 2013 showed the NSA collecting stored internet communications from major US technology companies under Section 702.
- CONTEXT Advisory Concerning the Chicago Police Department's Predictive Risk Models. Chicago Office of Inspector General, 2020-01-23. The Strategic Subject List scored people's risk of being party to violence. It was decommissioned on 1 November 2019 when its federal grant ended.
- AGAINST Article 5: Prohibited AI Practices (EU AI Act). EU Artificial Intelligence Act, 2025-02-02. Prohibits assessing the risk of a person committing an offence based solely on profiling or personality traits, in force from 2 February 2025.
- FOR Met Police tout arrests, crime drop from permanent LFR camera pilot. Biometric Update, 2026-05-01. Six months of fixed cameras in Croydon scanned about 470,000 people and led to 173 arrests, with one false alert recorded.
- FOR Clearview AI. Wikipedia, 2026-11-05. A scraped face database sold to police, including the NYPD and ICE, and fined by regulators including the Dutch authority (€30.5 million, 2024).
- AGAINST Met Police tout arrests, crime drop from permanent LFR camera pilot. Biometric Update, 2026-05-01. Reports a man misidentified and detained in 2024, a disputed bias analysis, and legal challenges, one of which failed in the High Court in April 2026.
- FOR ICE extends Palantir's case management contract. Nextgov, 2019-08-21. Gotham links records and analyses telecom and location data for ICE investigators. About $51.8 million paid from 2014 to 2019.
- CONTEXT Palantir Technologies. Wikipedia, 2026-11-05. Gotham was released in 2008; later contracts include a £330 million NHS England data platform (2023) and a US Army agreement worth up to $10 billion (2025).
- AGAINST Sadiq Khan Blocks £50m Palantir AI Deal With Met Police. Novara Media, 2026-05-21. The mayor's office withheld approval over procurement breach, vendor lock-in and value for money.
- FOR Thousands of AI Authors on the Future of AI. arXiv (AI Impacts survey), 2024-01-01. 2,778 researchers gave a 10% chance by 2027 and a 50% chance by 2047 of unaided machines outperforming humans at every task.
- AGAINST The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity. arXiv (Apple researchers, NeurIPS 2025), 2025-06-07. Reasoning models show accuracy collapse beyond certain complexities, even with token budget left.
Ledger updated 5 Nov 2026. Open this ledger as its own page · The P(n) Board · Method
Themes: AI and machine minds · Surveillance and scoring
