The tech question
Were any of the passengers scanned actually on the watchlist? Without that key bit of information, how are we to judge whether the system is working or not?
British Transport Police (BTP) spent more than £320,000 putting half a million commuters through live facial recognition cameras, only for the system to identify precisely nobody it was looking for. Figures obtained by civil campaign group Liberty Investigates through Freedom of Information requests, and reported by The Guardian …
> Were any of the passengers scanned actually on the watchlist?
Statistically there would have to be least one otherwise why would you run an expensive system when you knew your targets never used the railway?
I suppose it is one, albeit expensive, way of quantifying the false positive rate.
A sensible approach would be to recruit a demographically diverse collection of volunteers who use the rail with a variety of frequencies (daily to rarely) and times (time of day, day of week) to compose a test watchlist.
If your test group carried a device etc that advertised their presence in the station, you could measure both false positives and false negatives.
you mean like the EICAR test virus used to confirm basic AV on computers
have a couple of dozen test subjects who will enter / exit the test area at random times during the test to confirm the functionality of the system and see how many times it spotted them
as on face value (excuse the pun) they don't know if the system is up to the task or no one on the watch list went through
"Were any of the passengers scanned actually on the watchlist? Without that key bit of information, how are we to judge whether the system is working or not?"
Shush now! Just be quiet. Don't ask questions - and especially not awkward ones. Snoopercams are good for you. The arseholes in the Home Office say so and they're never wrong about anything.
Ausweis bitte fraulein!
‘If you have nothing to hide, you have nothing to fear’….. angle.
Brexity, Immigrant and Burnham hating YouTube celeb lawyer BlackBeltBarrister may disagree … and based on some of his output with brain-dead Police who gain employment in the UK .. I tend to agree with his ‘not making a conclusion, just explaining the law’ skepticism here.
>Those who wish to avoid detection will find a way.
A faje beard, headgear (hat, turban, hood…) and dark glasses would probably defeat most of these systems although you might look suspicious from the get go.
A white cane or guide dog might offset that. :)
Serious criminals probably just take an Uber.
"Those who wish to avoid detection will find a way."
Which means that the bulk of people that get chased down and tased will be unwitting innocents.
There are stories of people in the US being "flocked", identified by the Flock surveillance system, as being at the site of a crime that weren't anywhere near. The filth are convinced based on the word of some clanker and can't be bothered to do any real investigating. They aren't looking for the truth, just somebody to toss in the clink to make their metrics look good. Better to arrest the wrong person to show something on a case rather than have it go nowhere.
Well, government already subsidise rail by £12bn a year, and each new rail carriage works out at about £2m a pop, so I'm not sure £320k is going to get you much.
I suspect that BTP were trialling various hypotheses here, a likely one of them is whether facial recognition could be linked to gateline ticket data to identify serial ticket dodgers and reduce the £400m of annual fair evasion. Even a 1% reduction in rail fare evasion (whether through real matches or simple deterrence) would be a three-fold return on the cost of the trial. Another likely trial would be whether the technology can match faces from on-train CCTV, for example tracing suspects after on-train criminality.
If rail fares were not so absolutely extortionate then maybe there would be less fare dodging.
I do not live in London, but there are plenty of museum / gallery exhibitions there I would like to see.
Cost of a day return from Midlands / North is so expensive that those visits do not happen - only way that is viable is off peak over multiple days & staying over in London (off peak + London hotel a lot cheaper than peak day return)
Cost of a day return from my commuter town in the southest to London is about £37 per day now if you need to travel at peak time. I've met relatives from the midlands in London for a day out (starting early in the day) and they've paid about the same for an advance ticket for a 2 hour journey as I paid for my 45 minute journey due to the way the fares work.
"Well, government already subsidise rail by £12bn a year, and each new rail carriage works out at about £2m a pop, so I'm not sure £320k is going to get you much."
The money could go to mowing down a load of homes to add more lanes to the most crowded motorways, if you like. £320,000 pounds here, another £1mn there and suddenly it adds up to real money.
I expect that many would have liked to see the funds go to having more cleaners at stations to keep platforms tidy and collect the trash that inconsiderate bastards just drop on the floor of carriages. They aren't living wage jobs, but it does employ people with few options while doing a great service.
Adding just more one lane to a motorway isn't going to fix traffic - often the only thing that improves traffic is to close lanes, my guess is that frivolous trips by those who cause traffic, occur at a reduced rate, thus resulting in less traffic; https://en.wikipedia.org/wiki/Braess's_paradox?useskin=monobook
If bastards have their garbage cleaned up for them, that would likely result in many of them dropping more garbage on the floor, making the problem worse until the cleaner gets to the carriage.
"Adding just more one lane to a motorway isn't going to fix traffic"
Agreed! If all of the jobs are downtown and all of the home are out in the suburbs (that keep moving further away), adding lanes is useless. Adding more public transportation would be far more effective. It would also help if not every company demanded that staff start at 8am and leave at 5pm. I'm not a morning person so I'd rather come in late and stay late. My mom is an early morning person so would rather start early and leave early (ok, not any more. She's retired). Couples with kids may want to get away from "normal" schedules so one is at home in the morning to see the kids off while the other will get home earlier to look after the little darlings rather than paying for childcare.
That's pointless, modulo the economic benefit(?) of employing people to build the wider roads. Traffic will just increase to fill whatever capacity is added by those extra lanes.
WTF does this have to do with PC Plod's unfit-for-purpose shiny new spy cameras?
"might explain why "seriously bad people" stay away from train stations"
Do you mean the ones that wear the standard criminal uniform of a dark hoodie and track pants? It's hard to spot a face under a pulled up hood during the day from a high angle. If they don't put the cameras up high, they get vandalized which makes installing them even more pointless.
I'm not opposed to CCTV. The problems stem from FR and other data collection.
You want to improve the train services?
Sack the CEOs and all the other C-Suite, the lot of them, and employ angry train nerds who can off-the-cuff start defending a thesis about the history of the railways they composed on the spot, and who actually use the rails and related public transport.
Don't pay them any direct financial incentives, but do promise them that if in five years' time the general consensus of the train-riding public is that things are better and look to get better still, the Government will throw Real Money into fully refurbishing their favorite historic preserved steam locomotive and a period-authentic set of carriages to run up and down the mainline at least once daily. And they'll get to ride on the footplate.
You could almost certainly attract some world-class talent for a 'decent middle-class' salary wage if you dangle the prospect of getting Duchess of Hamilton or Mallard running again in front of 'em.
> Considering the number of times a commuter is already caught on camera
That is potentially the greatest hazard posed by this technology. You could train your system to recognise the "frequent fliers", the times of travel and platforms they used (hence their potential destinations.) This easily becomes constant (24 hour) universal surveillance when you consider the number of cameras out there.
Unrecognised subjects would automatically be suspect but of course the clued up malefactor would use a mask of the face of a recognised traveller who regularly travelled to his malefactorial destination.
For those unaware, what you describe is trivial in 2026. Building profiles, linking them, and tracking them through a system is the bread and butter of Facebook et al. Google have been doing this since mid 2000's with web search. Its compute constrained, but we have plenty of that.
Face geometry is just another data point the capture of which takes a few hundred milliseconds.
I hope this is satire.
Surveillance cameras cannot do anything to end crimes - the only beneficial things that cameras can be used for is to increase the conviction rate, but as that doesn't seem to happen in practice and as the negatives always outweigh the positives, surveillance cameras should not be used.
At best cameras could move crimes to somewhere else - which isn't something that ends crime.
"Considering the number of times a commuter is already caught on camera, once more makes no difference"
Serious question here:
I totally understand the privacy campaigners' point about public surveillance. But if you have any facial recognition system looking for baddies, it's got to scan innocent people hasn't it? You can't tell if someone is wanted or not until you've scanned them.
If I'm correctly understanding your question, then the answer is obviously yes, if you want to run facial recognition to detect criminals, you have to scan plenty of noncriminals. That's the problem, well it's the biggest problem, there are more. There are many things that you can do to criminals which would work in theory against criminals, but nonetheless are not acceptable because they cause too much harm to the innocent. We have been restraining the actions of police for centuries because of this tradeoff. Unfortunately, police often see themselves as beyond reproach by definition and don't understand why a thing they think might be useful, even though this evidence helps to demonstrate that it isn't even if you didn't care about privacy, might not be acceptable to the public.
It is about proportionality. You could arrest every 10th person who goes through a ticket barrier, and then treat them to in-depth questioning or even torture in the hopes of finding something to justify arresting them. That could be done. And I predict some baddies would be found.
But it is clear that that is disproportionate. It might turn up one criminal a year but everyone would agree it was unacceptable.
Similarly - it is disproportionate to scan and attempt to identify half a million people in the search for baddies. It turns out that all 500,000 people were not wanted criminals, but they all had their privacy interfered with. IT IS NOT SOCIETY'S BUSINESS WHO I AM IF I AM NOT DOING SOMETHING ILLEGAL!!!! THAT IS WHAT A FREE SOCIETY MEANS!! THAT IS WHAT MY PARENT'S GENERATION FOUGHT AND DIED FOR!!!!!.
The article expressly stated:
"BTP says it cannot identify people who aren't on a watchlist and that it immediately deletes their biometric data."
Sounds to me like no records are kept of travelers who were not identified as being on the watchlist. I'd say that's pretty proportionate - someone suspected of being a criminal (because their face matched a specific individual the police are looking for) gets their data recorded, everybody else doesn't.
Of course, that assumes 2 things:
1. They're telling the truth that non-matches are immediately deleted.
2. The system will successfully produce a match if a previously-identified suspect is seen.
Sounds to me like no records are kept of travelers who were not identified as being on the watchlist. I'd say that's pretty proportionate - someone suspected of being a criminal (because their face matched a specific individual the police are looking for) gets their data recorded, everybody else doesn't.
Ah, well.. So I head to my nearest train station, stroll happily through it's hallowed, unstaffed halls. I pause briefly to purchase an Upper Crust baguette and bludgeon someone with it. I then board my train, tap out at the far end and go on my way. As winter is approaching, I was wearing my favourite hoodie and a scarf over my nose & mouth that has Andy Burnham's face printed on it. I leave the scarfs or masks with skulls to the lower class of ruffian.
But I wasn't on a watchlist (maybe? who knows these days?) at the start of my journey, and I probably won't be at the end. Assuming the system did manage to get a decent shot of my face, I might be added to a watchlist (or several, probably incompatible ones) so the next time I take the train, some BTP officers might be summoned to arrest me. Assuming there are any on duty and aren't busy doing something else. Or someone else entirely might be arrested because FMR.
So kinda pointless compared to just having more staff at stations, but will probably result in everyone ending up on watchlists because we're all guilty of something. Then it'll be up to TPTB to determine a suitable punishment. Get bailed for being a possible terrorist or fuel thief, get no bail for stabbing a dinghy..
Surveillance systems have been demonstrated time and time again to not delete collected data.
For example, the "discord" disservice claimed to "quickly delete verified ID's" and ended up leaking 70,000+ ID's to criminals, that weren't deleted; https://www.bbc.com/news/articles/c8jmzd972leo?ref_timestamp=1760136504517
Due to how it is guaranteed that such data will be leaked and/or abused, it's not proportionate to collect such biometric data in the first place, even if it's true that non-matches are deleted.
How the false positive rate is nonzero, which will have the result of several people who aren't criminals, having their biometric data stored, for later use and abuse, is not proportionate either.
Was there really only one false positive?
"Surveillance systems have been demonstrated time and time again to not delete collected data."
In the US, Flock has stated many times that they don't keep the data coming from their cameras beyond around 30 days. That doesn't mean their clients aren't storing it for much longer, perhaps indefinitely. The message is "We aren't keeping this forever so if you want it, store it yourself. (Here's a quick video on the easiest way to do that and manage it going forward)"
Sounds to me like no records are kept of travelers who were not identified as being on the watchlist. I'd say that's pretty proportionate...
No, it isn't.
That is my whole issue with this. It is nobody's business who I am, or what I am doing, unless I am doing something illegal. And scanning half a million people to find nothing is, to me, obviously disproportionate. No one should be scanned just "in case they might be a criminal". We went through this many years ago with "stop and search" - maybe you weren't around for that but many of us were. Who I am, what I am doing, where I am going, is nobody's business - particularly not the state's.
It is a general principle of a free society. But just in case you don't care because it is happening to "other people" just think what will start happening to you when someone who looks a bit like you does something illegal. Immediately you will have to allow an extra 30 minutes to get every train. You will have to tell every policeman you walk past in the street where you are going and why and produce the letter you have from the Chief Constable which says "this guy's OK - he just happens to look a bit like MAJOR CRIMINAL X", ...
"it is disproportionate to scan and attempt to identify half a million people in the search for baddies"
But where is the disproportionate harm to the innocent in that example? Facial scanning happens with no harm or detriment to the person scanned, they won't even know it is happening.
Remember that the facial scanning isn't about identifying each person scanned. It's only about seeing if they are on a "wanted" database.
You forgot to add "and building a database of every face that gets scanned and when/where that happened". Why bother with snoopercams unless they get (ab)used for that?
The cops will eventually say they might as well do that in case all that yummy data comes in handy later on. They may not say that now. But who would believe them? Really?
Besides, the end goal is to monitor everything and everyone, just as it's always been. That's becoming much more likely since the costs of snoopercams and image processing are getting cheaper and cheaper. Coming soon: AI will do the drudge work rather than minimum wage drones from Crapita, G4S, etc staring at CCTV.
The cops kept quiet about the epic fail of the station snoopercams. And when the truth emerged about this exercise in futility, instead of abandoning it as a blatant waste of everyone's time and money, what did BTP do? They extended it. Bastards.
You want fingerprints to be leaked to criminals too?
If fingerprints are merely collected at mass, distributed between hundreds of stations, there is a 100% chance that there will be dozens of leaks (let alone if copies of all known fingerprints are distributed to stations).
Any competent criminal has the capability to prepare gloves that modify the resulting scanned fingerprints, so such scanner wouldn't be a concern to them - only to non-criminals.
Focusing resources on primarily attacking non-criminals, is not how criminals are caught.
They should have used AI. Then it would have made up a few matches.
Anyway, this is one useful building block to the uibiquitous dystopian surveillance state. If it actually worked. Now both people who want it to work, and people who would like to have some privacy left, will have to pay close attention as to why nobody got matched. Even though they've a police watchlist to match against, it's far too easy to extend this to any sort of political opponent. Already UK police managed to pay house visits over innocuous social media posts and then "innocuously comment" that the visitee had a lot of "brexity" books in his bookcase. Welcome to the surveillance state rat race.
On a philosophical note, the transport police exists to keep travelers safe, not to be a convenient funnel to feed their dragnet surveilance toy with. So this is official abuse.
The cameras used AI facial recognition, which happens to be faster and more reliable at facial recognition than humans, although with a false positive rate that is still far too unacceptable.
Yes, if LLM's were used (which isn't even AI), the false positive rate would be exponentially higher (but you just gave them the idea now).
Yes, the biggest criminals here are the transport police, who have the habit of attacking travelers via surveillance cameras, instead of keeping them safe.
According to the article, the trial lasted 6 months and used 100 hours of police time - thats circa two people for one day a week. from having run in-the-field trials of video technologies, I suspect the frequency of trials would have lower and the individual trials longer, so probably 2~3 people for 3~4 days once a month for 5~6 trials across the period. Suspect the cost figure includes equipment, system time, installation/teardown, expert consultancy, office/"desk" and expense costs.
>"it's even more insulting to waste almost 100 hours of officers' time and £320,000 of public money when it produces such meagre results"
If you don't know how the technology will perform, you firstly run a "benchtop trial" then a field trial. I would thus say the findings are consistent with ths. Hence the question is having got these results is deciding on what to do next. I call that a result.
Interestingly, I note "BTP says it cannot identify people who aren't on a watchlist and that it immediately deletes their biometric data. " which would seem to indicate a simple trial of running a scan one day - say in rush hour, and then repeating the scan on the following day and see how many "regular" commuters it picks up can't be performed and would also provide information on the performance of the tech.
Finally, I would not regard CCTV scanning "intrusive identity checks during their commute"; to be intrusive people would need to be stopped and made to either pose for the camera or be asked to produce Id. ie. have their commute interrupted.