'I was misidentified as shoplifter by facial recognition tech'
bbc.com
bbc.com
Being evicted from a store is necessarily a public act. Telling someone wrongly that they are a thief, in public, seems like libel. Since most people go to stores in their neighbourhood, this could easily happen in front of your friends, neighbours, parents of other kids at your school, fellow members of social clubs or worship clubs, etc.
If this happened to me, I'd be going and seeing if I could get any lawyers to take this on a no-win no fee basis.
It seems like businesses are increasingly failing to think through what their process will look like to real human beings.
If people are going to treat these things like they are forensic evidence and not lead generation, then they shouldn't be legal. Some precedent that people need to check these things before acting on them would be very good.
human oversight policies provide a false sense of security in
adopting algorithms and enable vendors and agencies to shirk
accountability for algorithmic harms.
Having worked on systems that required human oversight, I've seen first-hand, and analyzed data on, how people click the "LGTM" button without reflecting. We even noticed how when we improved our predictor with expert input, so that it performed better according to experts who worked with it, it would perform worse when we benchmarked on the previously "manually checked" data set. Analysis showed that the manual checks had let through so many of the algorithm's faults, that when the algorithm improved, it got a worse score.
See also https://predictive-optimization.cs.princeton.edu/ :
We present seven flaws of predictive optimization. Our aim is to outline a set of
objections inherent to predictive optimization that cannot be easily fixed using a
design or engineering change. Taken together, these critical flaws undermine the
legitimacy of applications of predictive optimization.
1. Good predictions may not lead to good decisions
2. It's hard to measure what we truly care about
3. The training data rarely matches the deployment setting
4. Social outcomes aren’t accurately predictable, with or without machine learning
5. Disparate performance between groups can’t be fixed by algorithmic interventions
6. Providing adequate contestability undercuts putative efficiency benefits
7. Predictive optimization doesn't account for strategic behaviorIt should be a given that these systems don't produce anything actionable, they are a guide for a user to look more closely, of course, that requires someone in the chain stops and says "this isn't usable evidence".
In the case of store workers, it's not obvious what other checking they could do. The only reasonable use I can see is for them to keep an eye on the person that has been flagged. But I expect that the entire purpose of this tech is to reduce the number of store guards they need to pay, so they are likely not to have the manpower. If the policy is 'don't let them in the store', I don't see how a store worker can avoid making false accusations.
Any piece of software that puts out data that is acted upon should have clear disclaimers attached. Calling it "intelligent" is irresponsible.
If a human of animal behaving the same were to be describe as intelligent, than so are those systems.
> If a human of animal behaving the same were to be describe as intelligent, than so are those systems.
I cannot parse this sentence. Behaving the same as what? What bearing do claims of human or animal intelligence have on me stating that generating liability by trusting probabilistic analysis, and actively harming people, by calling software "intelligent", is not responsible behavior?
Require that the system intentionally produce false positives (selected at random) at approximately the same rate as the true positives. This means that the users of the system will be aware that any indication has only a 50% chance of being "real", and therefore can be used as an indicator to increase caution but NOT on it's own as a signal of guilt.
I think it could be an actually pretty good solution for this. A bit like Chaos Monkey but for human systems. Introducing error often enough that operators don't assume it never happens.
Without the intentional false positives they could simply throw out 1 individual with no need to keep an eye on anyone at all. That's enen MORE efficient, but it utterly ruins the lives of anyone falsely identified.
Are most kinds of forensic evidence actually any better?
Certainly, more transparency than "we bought some facial recognition software, no, you can't see the code, that's a trade secret".
That's precisely my anecdotal experience.
Subtlety and nuance are easily ignored where their absence appears to provide easy answers and the avoidance of responsibility for their outcomes.
Similarly to "It Is Difficult to Get a Man to Understand Something When His Salary Depends Upon His Not Understanding It".
That's because the step from "potential leads" to "flawless machine" is a relatively small leap easily hidden by throwing out bad stats and case studies about people caught red handed. Even if you say the machine is 80% accurate, a lot of people will happily accept harming the 20% on the chance they might be in the 80%. (Though, I'm willing to bet the false identification percentages are both not public and not kept up to date, if they are even known).
Slander.
Libel is published, Slander is spoken. If they put a picture of her on a sheet that said "shoplifter" and posted in on the wall (or the net), it'd be libel.
But, I don't think you'd win a case based on Slander either:
* you'd still need to show it was published to a third party (though as you say, maybe you had friends present) * They can show that they reasonably believed it to be true (the system told them so). Their wording then becomes incredibly important - did they say "you're a shoplifter" or "you've been identified by our shoplifter recognition system"
They might also try to argue that stopping shoplifting is in the public interest and that this was a statement of opinion related tot hat.
I agree though, if this happened to me, I'd definitely be sending a letter or two to dissuade them from making a similar mistake in future (I think I'd also be trying my luck with Facewatch given that their system disseminated that "information")
Although, given that this is third party tech, maybe Facewatch counts as the publisher.
I think the better legal tactic would be defamation as it doesn’t require intent.
If the system got gamed (people with a history of false positives seeking to be misidentified), fine, it would encourage the technology to improve and eliminate false positives.
I remember being taken aback by the amount of street CCTV cameras back in the early '00s. Coming from Canada for a visit that feeling of being observed nearly constantly was really uncomfortable and bizarre. For Brits it looked like a norm (or they didn't show it). Also the parallels with 1984 were uncanny, especially given that Orwell was English.
http://soerenkern.com/web/wp-content/uploads/2014/06/uk-met1...
I'm really glad the EU just completely banned predictive policing.
Almost nobody worries about these transport safety cameras because we don’t fear or tolerate overpolicing in the same way.
Of course there should be safety cameras on public transport!
And if a US tourist was on a night bus in London they too would be happy there were cameras. A nation of drivers has a very different calibration, I suspect.
In the end the CCTV doesn't actually prevent anything from happening ...
The question of whether it prevents things from happening is a question of statistics.
I suspect the data doesn't support your claim, and that certain categories of crime have indeed been deterred. But neither of us can know without that statistical analysis.
Yes, the tory government have been trying to stack the governance of the BBC with their own people, but it doesn’t always go the way they expect.
Australia wants that title!
Bunnings here also uses (used?) facial recognition technology.
Also concert venues and sporting events.
https://www.sbs.com.au/news/article/bunnings-denies-reintrod...
I'd be willing to bet that supermarkets are next, if they aren't already.
The vast, vast majority of CCTV systems in the UK, that are part of this “culture”, are just store and private CCTV systems that police can get access to with a warrant, by going door to door. The same as in almost all western countries.
https://en.wikipedia.org/wiki/Mass_surveillance_in_the_Unite...
There’s almost no police CCTV network, and there’s no reality to the fantasy element of police being able to see you wherever you go in the UK in real time (with the exception of a few crowded high crime areas where tourists get robbed, ironically, and a few high violence and terrorism risk areas, public transport facilities and live road traffic cameras).
There are some council-owned networks in high crime areas which are pretty contentious, and then there are these police facial recognition vans which are clearly offensive bullshit overreach.
In particular a lot of visitors misunderstand the prevalence of cameras pointed at our roads, which are not live feed cameras at all but speed-triggered safety cameras.
Basically it’s important to understand that the vast majority of cameras in the UK are people and businesses legally protecting their own shit, just like in the USA.
One key cultural difference is ANPR. In America the general perception is that privacy includes where you go in your car, because you treat it as a portable living room. There’s never been that presumption here. ANPR systems probably can track cars in metropolitan areas and on some motorways, but live traffic cameras are unusual in most of the country.
There is definitely a slightly less self-centred lean to the balance of individual rights compared to community safety compared to North America. But this is also part of why we don’t have to put up with anywhere near the same risk of violence or crime, even in areas that are significantly more densely populated.
We also have pretty good laws governing this stuff.
In short, Hot Fuzz ain’t real, and I think North Americans like to project this idea of a surveillance culture onto the UK without considering that they are likely under the exact same sorts of surveillance in the same sorts of places, without as many of the same legal protections.
(And with plainly discriminatory surveillance systems like Shotspotter to boot)
Do you think retailers in the USA and Canada are not using the systems described in the article? Because you would be wrong if you do.
https://www.dailymail.co.uk/sciencetech/article-12592563/wal...
- your system thinks I am someone who stole something but I am not that someone. (Mis-identification)
- your system thinks I am someone who stole something, and I am that someone but I did not stole anything ever. (Mistake in the database)
- your system thinks I am some who stole something, and I did, but I have already been properly punished for it. (Rehabilitation)
If you just bar people from the shop automatically on a positive identification you are going to make the above mistakes on a massive scale. And because of the nature of these system if they made a mistake once with someone it is possible they will make the mistake again and again, so you better handle the exceptions in a gracefull and sensible maner. And shops will absolutely not do the right thing on their own. Because it is just so much cheaper to hassle the few who has fallen off from the happy path than build in the proper safeguards, so shops on their own will just decide to “eat” the losses caused by persecuting innocents. And that is the part which is not right. That is where laws should come in and mandate proper procedural safeguards.
What are the proper safeguards? I don’t know. I have ideas on how that could go, but this comment is already getting very long. What I am saying is that I think the correct solution is not “no cctv with facial detection allowed” as many here seems to propose, but it is also not letting shops persecute innocent victims of the above exceptions. There is a third way with regulated and safeguarded procedures where I think we should be looking.
The straw that broke my camel's back was learning of the wide spread practice of police asking crime victims "what celebrity does the perpetrator look the most alike?" And then the police would put that celebrity's photo into their FR system and start harassing people in their city that look like the celebrity. So, mind numbingly stupid, I got angry at clients doing this, got barred from talking to clients, and soon there after quit the industry. The industry believes it is "not their problem".
I also guarantee this is not how sales and marketing are pitching it.
I’m aware of that. What I’m proposing is that we should make that a statutory requirement. If you are operating a camera system with automatic face detection it must be operated in a certain manner.
How many grocery stores will tell you what's security system they use when they already assume your a thief?
And if we are already proposing new laws to provide standing to sue why would we not also mandate some procedure so people can work out these edge cases among themselves?
This has been an issue with tenancy in my country. A tenant goes to the tenancy tribunal for any reason - even if it's the landlord's fault, and then all property managers, who collectively control most of the rental market, blacklist that person and suddenly they can't live anywhere.
With suing, the parties can already work it out among themselves if they both want to avoid going to court. I don't think it needs a mandated procedure. No-win-no-fee lawyers exist to serve poor people who have a clearly winnable case.
After talking to them and requesting the store manager, it turns out they had a photo of a shoplifter that looked very similar to me. Eventually we were let back in the store and it was all OK.
So, it’s not an entirely new problem? Although with facial recognition I guess there is less recourse; a manager is not as likely to believe they got it wrong if their computer tells them otherwise.
I don't think this status quo should be empowered further by error prone tech
we can limit it to those where the amount of individual stealing is relevant. not the long tail but the fat creamers.
He is talking about the ruling class with power over all of us, so they should have more oversight than any normal person.
Like there is no need to rob a store if the clerk cant open the vault.
The game is rigged at the level of who gets to stand for election, particularly in safe-seat districts that get decided in the primary. Very few people are able to get elected without help from their local political machine getting out the vote - enough to provide plausible deniability, but not enough to create an effective opposition.
Remember the monopsony of Sheldon Silver? He used his position to control which politicians would receive contributions for their campaigns, creating a network of people dependent on him for political survival. It's that kind of power broking that's the problem. The clear and present danger of entrenched power is why politicians must not be allowed a shred of privacy.
The (psychologically) correct way to think of this is as a colleague making a claim and telling you how certain they are. But this tech is not sold as a colleague, it's sold as a machine that is better than humans. It's not a perfect super cop but that's how it is marketed and why people buy it.
It prevents privacy issues and doesn't stop research.
Exceptions could be made in gov buildings perhaps due to national security but the idea of unlimited private companies taking my photo, or using video for gait analysis, facial analysis etc is abhorrent.
I'm not against tech and love what AI can do but private companies using my images for facial recognition is a step too far.
Next step: every big company/store requires a written authorization to enter. Back to square one.
It's easily countered by creating a law that prevents discriminating on the basis of opting into tracking, AI and whatnot.
Problem solved.
(I do think the idea isn't a good one and am not defending it.)
It will essentially kill it immediately but still allow research.
How different is it from a police officer stopping someone in the street based on some photo on his most wanted cheat sheet? Or a shopkeeper denying entry to a customer based on the likeness to a list of mugshots held under the counter. Those are faillible processes too.
Where face recognition is really useful is where it can match faces that would have otherwise fooled the human eye, just because it focuses on specific traits and ignores others like hairstyle.
Whereas the people in this article are going to get challenged, if not outright refused or arrested, in an increasing number of stores as this tech spreads.
There's a big difference. The cop knows that the resemblance could be coincidental and will (usually) do further investigation before actually harming the suspect. If an automated system fingers someone, the assumption is that the system is correct and innocent people caught up will suffer harm.
I'm not sure about the computer program though. It telling a third person (i.e. the person from the store) would be libel, but since it is the result of a calculation, I'm not sure if it's considered libel.
Technically, it’s probably saying “there’s an 80% chance this person matches a thief” and the store is interpreting that as “ban them.”
I don’t know UK slander laws, but in the US, it has to be incorrect to be slander. So this probably wouldn’t pass the test.
In the UK it would be for the defendant (I.e. security guard/store) to prove that the claim was actually true. So in this case they would be stuffed.
The person should definitely contact a solicitor.
In any case, the article is not about an arrest. Making someone leave is not an arrest, it's the opposite (an arrest is detaining someone).
"Within less than a minute, I'm approached by a store worker who comes up to me and says, 'You're a thief, you need to leave the store'."
Knowing that one in 40 people so flagged aren't guilty of a crime means they are deliberately imposing upon those so flagged. The choice to use a faulty tech isn't itself a mistake its a choice. EG if you exposed a million people to this tech daily you would falsely arrest 11,000 people.
Same result https://www.youtube.com/watch?v=R5Cw043lPCE ?
Unfortunately it doesn’t mean anything of the sort. This tech doesn’t prevent typical street crime like muggings or assault. And the chance of getting hurt or killed by a police officer whose dispatcher just identified you as a criminal is much higher than what the random folks who skipped bail that are in this system would do to you (ie nothing).
Then there’s the “low” false positive rates. They say only one of 33,000 people scanned is falsely identified. Okay, that’s pretty high. If they scan one percent of the UK population with these devices each day, that’d be 20 false positives a day, and 20 more potentially violent encounters with police a day. Compared to what’s likely a much smaller number of correct identifications of actual wanted individuals, most of whom could probably be found by much cheaper, less problematic methods.
This should apply even to past criminals.
Such laws protect against racism and other forms of discrimination, and they help integrate convicts back into society, at the cost of "commercial freedom".
In jurisdictions with these laws (I know at least one), it's impossible to create, for example, bars only for men or only for women or refuse service to people we don't like (also, because they belong to some group, incl. morally reprehensible groups). There are some hacks around this and exceptions, but generally, it works and solves more problems, than it creates. Private gatherings or clubs are likely ok (nondiscrimination laws should still apply).
UK and US are outliers here with their "can refuse service to anyone" approach.
I’ve only known examples where you’re free to reject anyone for any or no reason as long as it’s not because they’re one of a very small list of protected groups.
138 misdemeanor law: Whoever, professionally engaged in the provision of services, demands and collects a payment higher than the applicable fee for the service, or intentionally, without a justified reason, refuses to provide the service to which he is obliged, shall be subject to a fine.
Some forms of propagation of fascism, communism and nazizm is prohibited in Poland, but I guess it probably would be illegal to refuse service to someone who otherwise declares support for some reprehensible ideas, and does that privately, so it doesn't fall under hate or other laws.
On a humid day in Bethnal Green, in east London, we joined the police as they positioned a modified white van on the high street.
Cameras attached to its roof captured thousands of images of people's faces.
If they matched people on a police watchlist, officers would speak to them and potentially arrest them.Or what about this? A man covered his face when walking past one of these vans in a public street. He protested and was then fined £90 for swearing (swearing is not illegal but The Pubic Order Act is another law that is often abused by police but that's tangential)
https://www.independent.co.uk/news/uk/crime/facial-recogniti...
Of course, all this assume US police have consequences for breaking the rules. So this is unfortunately moot.
Simply going out looking for people of interest in east London is perfectly fine, we call it "police work" and have done it for bloody ages.
Maybe instead we could issue officers with a deck of cards with 52 faces on them and they can use that as a comparison point? Seems pretty inefficient to me but im only a tax paying pleb what would i know.
Police surveillance should be inefficient. People do not exist to be arbitrarily inspected at scale; it should be an expensive, manual process and you should be forced to choose carefully whom you target, by necessity keeping that list small and limited to those for whom you can articulate cause.
If you're taking money from the public and spending it you should be legally bound to use it in the most efficient and effective way possible.
Snark aside, surveys are quite literally the least efficient means of gathering a sample. Scanning random people on the street when a known criminal is likely to lay low and stay inside. I can't imagine expensive surveillance cameras is cheaper than paying an investigator to track down a suspect.
Let's ignore the privacy implications as not all countries have an expectation of privacy outside windowless closed private spaces. How is this meaningfully different from a general stop and frisk policy? Sure, this adds a non-physical filter before the physical police interaction but it's still a widespread warrantless search. How does this work with the presumption of innocence?
If there was some guarantee all that data wasn't going into some government sanctioned back end, we can start to compare it to police out on the street.
>However, fundamentally, being arrested for looking like someone else is not a new phenomenon.
Was always wrong and still is wrong.
But if you want a modern difference: companies these days (even minimum wage) are really strict about arrests. Even a false arrest can cause complications in a background check, so there is more potential (albeit indirect) slanderous effects from mistaking people nowadays.
I don’t understand this comment. The police sometimes have to arrest people based on descriptions and other fallible data points. It’s unfortunate when the wrong person is arrested (and false arrests should obviously be minimized), but it’s not reasonable to expect police to get this right absolutely 100% of the time. If someone is a fugitive then they are very unlikely to reveal any information that identifies them with certainty.
Also, police in the US use the same technology: https://www.nyc.gov/assets/nypd/downloads/pdf/public_informa...
It's always been wrong and still is wrong, but the reasons changed over the years. Americans have traditionally been adverse to mass surveillance and I don't think that changes much today. The government just does it in sneaker ways nowadays.
What might matter is the actions taken - but would they be any different if a human said “oh look that’s the well known shoplifter, keep her out”
Humans may also be disciplined for not attending to an instruction from the infallible machine, whereas there are no stats or electronic records of human judgement of "is that the lady who shoplifted the other day? Hmmm, maybe... Maybe's not really good enough for how little I'm paid".
How did they know she shoplifted in the first place?
Does the tech also identify shop lifting, or do they just search a database of people convicted of shoplifting (in which case they could have just checked this person's ID and realized it was mistaken identity).
Is being caught shoplifting a credible reason to deny someone access to basic goods indefinitely? What if all supermarkets in a town use the same facial rec service?
Even putting a human being in the loop is trouble prone because people aren't on average good enough at this.
It would be more useful to require a free membership to shop there and validate allowed users via bluetooth and ban users devices and phone numbers.
She didn't.
If random store employees start walking up to people and asking them to show ID, is that an acceptable state for society?
But note that this is the UK. Different issues from the States (where this wouldn't ever pass. We have weak social security specifically becsuse citizens rejected a national ID).
Should shoplifting once mean that a person is banned from all stores that use this service for life?
Also: I assume that this service wouldn't just impact shoplifters, but anyone that a store saw fit to ban for any reason. What about people who experience mental health crises? I had a friend who had psychotic breaks while her issues were still undiagnosed and untreated
I know that this ultimately depends on the store's own policies, the facial recognition tech just provides information. but frankly I don't want to live in a world where making such a minor mistake in your youth could basically ruin your life
I mean, it depends, right? Barred from buying any groceries for stealing a candy bar is obviously wrong… but if you walk out the door with a five-figure watch you didn’t pay for, I’m happy for you to be banned from Rolex shops for life, yeah.
Unfortunately it didn't stick. If you wear a mask these days, you only stand out more.
It's very important to ponder what the PRC is pioneering today because we're about three years behind doing exactly the same things, and for the same reasons. It's only going to be a brief time before things we're horrified about, and vitriolically condemn totalitarian states for, are things we're going to fully normalize within our own culture.
Or just implant RFID into every citizen at birth. It's for your safety, after all - if you get lost or abducted, that's how they'll find you.
And for anyone that complains about tech being used this way; you're too late, it's already happened, when you apply for a bank account, mortgage, any kind of insurance...the list goes on and they've been profiling people for decades and decades.
It is impossible to prevent facial recognition being used...so laws need to be updated to ensure that the public have fair rights around this tech. But since when did laws or public opinion/understanding ever keep pace with technology?
Given the UK population of 67 million, that’s about 2000 false matches.
$100 says they've taken the numerator of 'people we detained where it turned out we dun goofed' over the denominator of total people scanned.
I think this sentence sums it up perfectly...
> "It felt intrusive… I was treated guilty until proven innocent," he says.
I don't want to live in this kind of society.
Under UK slander law it would be for the store to prove that the claim was true, not for her to prove that it was not.
In any case, contacting a solicitor does not hurt.
https://www.ftc.gov/news-events/news/press-releases/2023/12/...
Do we do that? Do we want to do that?
Given current controversies around Her, I'm sure some techie will see Minority Report and view it as a dream to pursue rather than a cautionary tale.
The BBC should understand that much facial recognition software struggles with non-white faces, like Mr. Thompson's. "Family resemblance" my ass.
If they’re standing there with a van and immediately stopping you the only thing you have to do is pull out your ID?
TFA says that it has a failure rate of 1:33,000. That's a "do not ship" rating for almost anything else.
> immediately stopping you the only thing you have to do is pull out your ID?
Someone I know was detained on the side of the M25 for an hour sorting things out after being pulled over.
He presented the police with his ID and they decided it was a fake. His name was almost identical to someone who was wanted - his middle name and date-of-birth were different.
The Police said that was common on fake IDs - just change a few small bits of information - and that they'd have to take him to the station where he could sort it.
The only reason he didn't get taken in in the end is because the description of the wanted person noted that he had tattoos on his chest. On the side of the M25, at night, the only thing that stopped my mate being hauled into a London police station was taking his top off.
Anecdotes don't make data, but the idea that a copper will simply accept ID despite a system saying "this is your guy" is incredibly naive and suggests you've not had to interact with them much.
Or the opposite is true and this has been enough to solve all my police problems? I’ll admit it is not in the UK, but I doubt they have a significantly less professional police force.
> TFA says that it has a failure rate of 1:33,000. That's a "do not ship" rating for almost anything else.
For anything where failure means death, sure. For situations where failure means a minor inconvenience, maybe not so much.
To me, the only relevant point of comparison here is the rate of misidentification by officers while _not_ relying on the face id tech. Because that’s the alternative, not having no arrests or searches at all.
Ah, that explains a point that I'd didn't bother to pull you up on.
Carrying ID is not a routine thing here in the way that it is in some other countries (there have previously been attempts to introduce a national ID but they were staunchly opposed). So, it's not a given that you'll have ID on you to show them.
Drivers probably have their driving license in their wallet, but even that's not guaranteed (because you don't have to have it on you when driving).
The "quality" of police varies by force (and, of course, by officer).
The Met, though, have had some pretty serious issues with misconduct (including sexual assault and murder) and are still working through the processes of identifying personnel who shouldn't be in uniform at all (the Met themselves found there were hundreds of officers who should have been sacked previously).
They're working to fix things (or claim to be), but you probably don't want a force that's been described as "institutionally racist, sexist and homophobic" to be entrusted with something like this.
> For anything where failure means death, sure. For situations where failure means a minor inconvenience, maybe not so much.
I would still say the failure rate is too high given that the outcome of interactions with the Police varies quite significantly (par.
> To me, the only relevant point of comparison here is the rate of misidentification by officers while _not_ relying on the face id tech.
I'd also be interested to know this. But, I don't think it'll go quite the way you expect.
I'd expect there'd be _fewer_ overall stops: coppers simply won't (mis)recognise as broad a range of people. If they're only stopping people they recognise (or based on stuff that's been radioed through), their success rate is probably better
> institutionally racist, sexist and homophobic
I’m not sure those things go together very well. If they’re bored they’ll just pull ‘random’ people off the street. They might not be at direct risk of arrest, but it’d still waste your time.
That’s half of the reason I need to show my ID all the time :/
Edit: I’m starting to think you might be right, and there’s no direct analogue because police don’t just sit in the middle of the street hoping they’ll see some suspect if they cannot instantly compare with 40k wanted faces. No officer would be able to remember them all.
Question becomes more if it’s worth to have some people misidentified to catch the bad guys.
I find statements like this interesting, from a risk analysis point of view. In the UK there are breast cancer screening programmes and the risk of a radiation-induced cancer for a woman attending full field digital mammographic screening is between one in 50,000 to 100,000. That puts the rate of inducing cancer vs finding cancer at 1 in 400 to 1 in 800 (because the majority of scan results find no cancer). Should we "not ship" breast cancer screening?
But the error count is much higher once an someone is actually flagged. One in 40 alerts so far this year has been a false positive."
I think it's a good idea to make it clear what you mean by "failure rate".
Why I find it hard to believe is that I am imagining having access to a face photo of everyone in the world, and trying to pick a specific close friend from those photos. Given infinite time to find the photo of that one person, I am convinced that I would find tons of false positives that might be that person, and that I would never even be sure I found the correct photo.
I am guessing that there is some serious sampling bias leading to such a low false positive rate (assuming the number is not complete fiction).
According to government (referring to "live facial recognition"): "All deployments are targeted, intelligence-led, time-bound, and geographically limited." https://homeofficemedia.blog.gov.uk/2023/10/29/police-use-of...
which makes that 1 in 40 number completely irrelevant when discussing false positives from mass deployment in grocery stores or other non-targeted spying.
It's an absolutely disastrous false positive rate when the consequences to those who are falsely identified are more than trivial.
But, hey: Progress.
1. If any one ensnare another, putting a ban upon him, but he can not prove it, then he that ensnared him shall be put to death.
2. If any one bring an accusation against a man, and the accused go to the river and leap into the river, if he sink in the river his accuser shall take possession of his house. But if the river prove that the accused is not guilty, and he escape unhurt, then he who had brought the accusation shall be put to death, while he who leaped into the river shall take possession of the house that had belonged to his accuser.
3. If any one bring an accusation of any crime before the elders, and does not prove what he has charged, he shall, if it be a capital offense charged, be put to death.
A 97.5% accuracy rate seems… really really high!
Obviously facial recognition shouldn’t be the only bit of evidence. But that’s remarkably accurate.
Another way to look at it is for the worst case person what is their false positive rate? It's going to be 100%: Joe Blow, who did nothing wrong to no one, will go from place to place getting kicked out for the rest of his (/the system's) life. The failures aren't random (or at least not entirely)-- some people will get a heap of error in their lap.
So while getting punted 1:10000 times you go somewhere unfairly isn't too bad, getting always punted is awful. The unequal distribution of errors creates bad incentives: the system is almost risk-less for those of us who aren't the victims of false hits, so why should we demand improving or banning it?
eg: 33,000 people walk past cameras but only two are flagged as matching a person of interest on the watch list.
One of those is a mistake, the other was correctly identified.
From the Police PoV that's one mistake in 33,000 .. but whose to say the remaining 32,998 people were correctly identified?
Perhaps they were all criminals that weren't identified, perhaps they were all not criminals but not correctly matched.
Perhaps 2,000 were criminals and on the watch list but not identified.
Machines are going to come up with false positives, but so will human beings who are asked to be on the lookout for some kid of some redacted race in a hoody.
That said, yes, it would be nice if there was a way for my car to automatically scan car plates around, linked to my phone so that when an amber alert comes in and I explicitly choose to enable the scanner at that point, it would report matches for that particular number automatically.
But we all know that's never going to happen. Well, the automatic plate scanners in every car will probably happen eventually - it's too easy technologically for someone to not come up with a bright idea to mandate them "for safety". But they will be enabled and effectively controlled by law enforcement 24/7.
John Lewis,
Co-op,
Tesco,
Sainsbury's,
Waitrose,
Next,
Marks & Spencer,
Boots
Primark
John Lewis,
Co-op,
Tesco,
Sainsbury's,
Waitrose,
Next,
Marks & Spencer,
Boots
Primark
I dislike shoplifting and petty crime but I also dislike this form of cheap, petty intimidation.
I don’t want to live in this dystopia.
(Though people told me that was exceptional and I was just unlucky)
Perhaps it’s already covered under the European convention on human rights, but I don’t know if that still applies.
HN today...
In the story I also noticed that people flagged are not treated as innocent until their identity has been truly established.
What kind of society do we want? Was the money spent on facial recognition systems better spent on “prevention”? Probably not as this money would not have gone to the police so we can’t have that! /s
1. Okay well maybe the technology was wrong.
2. Or maybe the shoplifter is lying?
3. Surely "my word against yours" in the case of "human vs. computer" is solved by believing the human?
4. Isn't the whole point of this technology to catch humans who lie?
5. Then surely, if it's a shoplifter, there is video footage and that will clear up the issue.
6. But what if there isn't footage. What if it's just an AI that's detected someone as a shoplifter based on their face?
7. The technology for this is really mature, now I'm inclined to believe the computer.
8. But what about the guy? (Yay, gender assumption based on the word "shoplifter")
9. Well, in order to believe him I would want to understand why the AI was wrong.
10. Oh, the whole thing with AI in many contexts is that YOU CAN'T KNOW THIS.
11. This is why people care about understanding how AI makes decisions and have serious reactions to technology that just "magics" a decision out of thin air, even when it's the right one this time.
12. But what about the guy?
Well, I've just newly understood some things about the social impacts of pervasive AI. Time to read the article.
If we surveil people, it's because we're just keeping the public safe.
By "here" do you mean in the UK?
Somehow the UK is consistently the first non-China nation where this authoritarian surveillance crap takes hold.
Other countries make pushes towards it, but UK appears to be consistently leading.
For the government and police, it's a cheap option to deal with the problem.
The abuse is a very real problem, I know a few who work in shops, all of whom have been verbally abused and physically threatened at various times. One of them was not long ago punched in the face in front of witnesses. Tesco, ever caring, had her working the next day, just a few hours later (incident happened on late shift, next shift was early morning, "are you sure you can't come in?, it'll be hard to find someone to cover") and seemed to put as more effort into investigating what she might have done wrong as they put into helping the police gather evidence to investigate the abusive individual, so she was glad of the CCTV. She got out off Tesco as quick as she could. She has had notification from the police that the individual involved is being prosecuted, but for a number of other offences also so it is unlikely she'll need to testify or otherwise be involved further.
The staff want the CCTV because they don't trust the managers and higher to look after their interests otherwise.