13 ms·
Face recognition police tools 'staggeringly inaccurate'
- camel_gopher 9y agoThe Nest doorbell is fairly good at this, but it has a relatively narrow field of view and 1080p resolution. Most of the CCTVs out there have a very wide field of view, which makes the number of pixels per face low.
- jjxw 9y agoAs another commenter pointed out, it also doesn't have to sift through thousands of faces. Chances are the doorbell is only exposed to a handful of faces a day which greatly reduces the volume of false positives regardless of the field of view and resolution.
- bsenftner 9y agoModern FR systems can compare tens of millions of faces per second, per core, on a $99 Intel Compute Stick. The thing that sucks about Nest and similar "security cameras" is their wide angles - that makes the distance of recognition far too close. To detect a face, FR needs between 28 and 54 pixels of head height, but to recognize with accuracy is requires a minimum of 120 pixels of head height. A Nest camera will not produce a head that large until it is 3 feet from the camera. I'd prefer to make recognition at 30 feet or more, preferably when a bad actor is not yet inside the property.
- jfasi 9y ago> Police facial recognition cameras have been trialled at events such as football matches, festivals and parades... High-definition cameras detect all the faces in a crowd and compare them with existing police photographs, such as mugshots from previous arrests. > "When we first deployed and we were learning how to use it... some of the digital images we used weren't of sufficient quality," said Deputy Chief Constable Richard Lewis. "Because of the poor quality, it was identifying people wrongly. They weren't able to get the detail from the picture." > Information Commissioner Elizabeth Denham said the issue had become a "priority" for her office. When you roll out facial recognition technology at an event with ten thousand people and expect your magical facial recognition system to pick out just the bad guys, you quickly discover that even a system that's 99.99 percent accurate has a 63 percent change of giving you at least one false positive in that crowd. In reality these systems are probably closer to 95 percent accurate, in which case you're looking at vastly more false positives. Like hundreds per event. This is the exact same statement as "breast cancer screening is staggeringly inaccurate." Do you know why breast annual cancer screenings aren't recommended for women under 40? Because breast cancer is so rare in that population that the high false positive rate outweighs the benefit of early detection. The correct thing to do is only screen only if there is already a suspicion of cancer. Similarly, pervasive surveillance where you (statistically) consider everyone a possible criminal is a waste of time. Anyone who's ever taken a probability course will tell you that unless your facial recognition tech is fantastically accurate you'll waste all your police officers' valuable time sorting through false positives. This is not an image quality issue. It's not a data quality issue. It's not a deployment issue. It's a Bayesian statistics issue. Unless your tech has superhuman 99.999% accuracy, it doesn't matter how accurate your facial recognition is if you're looking for a needle in a haystack. I'm sure the facial recognition is just as wonderfully accurate as breast cancer screening, it's just being applied to a population with a stupidly rare prior. This is a statistics 101, freshman undergrad-level mistake, and frankly it makes me mad as hell that authorities are stupid enough to make it and boneheaded enough to endanger their entire population's civil liberties in the process. For more info, check Wikipedia's page on the aptly-named prosecutor's fallacy: https://en.wikipedia.org/wiki/Prosecutor%27s_fallacy https://en.wikipedia.org/wiki/Prosecutor%27s_fallacy
- tzs 9y agoThe base rate fallacy [1] and the false positive paradox [2] seem to be much more on point for using face recognition for screening seem to be much more on point than the prosecutor's fallacy. [1] https://en.wikipedia.org/wiki/Base_rate_fallacy https://en.wikipedia.org/wiki/Base_rate_fallacy [2] https://en.wikipedia.org/wiki/False_positive_paradox https://en.wikipedia.org/wiki/False_positive_paradox
- thaumasiotes 9y agoWhat exactly are you trying to correct? Read your own link. https://en.wikipedia.org/wiki/False_positive_paradox https://en.wikipedia.org/wiki/False_positive_paradox says: > concluding that a positive test result probably indicates a positive subject, even though population incidence is below the false positive rate, is a "base rate fallacy". Whereas https://en.wikipedia.org/wiki/Prosecutor%27s_fallacy https://en.wikipedia.org/wiki/Prosecutor%27s_fallacy says: > At its heart, the fallacy involves assuming that the prior probability of a random match [i.e. the odds of a positive test result] is equal to the probability that the defendant is innocent [i.e. the odds of a positive subject]. For instance, if a perpetrator is known to have the same blood type as a defendant and 10% of the population share that blood type, then to argue on that basis alone that the probability of the defendant being guilty is 90% makes the prosecutor's fallacy Those are the same mistake. The prosecutor's fallacy is a base rate fallacy.
- tzs 9y agoFalse positive paradox: "The false positive paradox is a statistical result where false positive tests are more probable than true positive tests, occurring when the overall population has a low incidence of a condition and the incidence rate is lower than the false positive rate" Prosecutor's fallacy: "The prosecutor's fallacy is a fallacy of statistical reasoning, typically used by the prosecution to argue for the guilt of a defendant during a criminal trial". The false positive paradox is a result. To get to a prosecutor's fallacy or base rate fallacy from it, you have to use the result in a way that involves certain fallacious reasoning. There's nothing necessarily wrong with using a screening test that the false positive paradox applies to, as long as you recognize its limitations. When you are looking for a needle in a haystack (which is typically when you hit the false positive paradox) it gives you a smaller haystack to search for the needle, but that can still be a big improvement over not using it as long as it has a low false negative rate.
- stochastic_monk 9y agoPolice add yet another massively inaccurate tool to their arsenal which they can easily put innocent people behind bars. For example, bite mark “evidence” has been long disproven as having no basis in fact, but that doesn’t stop prosecutors from using it.
- ChristianBundy 9y agoMaybe I'm misremembering, but wasn't bite mark evidence used to catch Ted Bundy?
- dbasedweeb 9y agoNot really, and frankly as with many prolific serial killers the active ingredient was police incompetence rather than rare genius of the killer. https://en.m.wikipedia.org/wiki/Ted_Bundy#Arrest_and_first_trial https://en.m.wikipedia.org/wiki/Ted_Bundy#Arrest_and_first_t...
- JustSomeNobody 9y agoNot catch per se, but it was used in the trial.
- losteric 9y agoAlternatively, politicians continue cutting police budgets, forcing them to replace people with underperforming computers.
- jessaustin 9y agoWhere on earth is this happening? I would like to move there. In my county/state/nation, the federal government makes up new funding programs every year to induce local departments to hire ever more assholes to pry ever more into the lives of their subjects.
- derefr 9y agoBite mark evidence is fine... if what you’re trying to prove is whether the bite came from a human or an animal. It doesn’t do much more than that. There are a lot of forensic analyses like that: you can’t really use them to pinpoint a suspect; but they’re great at proving that a whole class of suspects couldn’t have been responsible. The real problem with police is the thought process that leads to playing Guess Who with suspects. Only one person of interest has no alibi, and can’t be excluded? They must have done it then, even if we have literally no proof positive that they did, and have no idea what the evidence in favour of all the people we haven’t looked at yet would be.
- jimnotgym 9y agoSo the UK police, under huge budgetary pressures (there are 21,000 less police than 7 years ago at the start of "the cuts" ) were duped into investing in an expensive technology that promised to save them resources in catching criminals. The pressure to get results overrode common-sense and concerns about civil-liberties. A bit like the DNA "fingerprint" craze which is now unraveling with a number of convictions being challenged. If it had been treated from the beginning as corroborating evidence, rather than The Answer, this would never have happened I am amazed every day at work that everyone expects technology to be a "silver-bullet", and are outraged if they still have to do anything themselves
- godelski 9y agoI know a lot of people here are talking about the tech and how of course it makes too many false positives. But are we not going to talk about IF this type of thing is even okay? I for one don't think so. Too Orwellian.
- salty_biscuits 9y agoI don't really see that it is that bad compared to what happens now. You have people doing it currently with all their subjective biases (such as race and class). Maybe this just makes it cheaper and maybe more effective as the tech matures. Surely accuracy concerns can be addressed by combining multiple data sources (e.g. looks like someone dodgy, is somewhere strange, doing something strange, just the way a police officer would rationalize investigating further).
- jdc 9y agoBecause we don't want people who can track the entire population at once. It's too much power.
- dreamfactored 9y agoThe fundamental principle of presumption of innocence means you shouldn't go around spying on everyone looking for signs of criminality. And the idea that the tech is more dispassionate and therefore fairer than humans is a step on the road to hell. I guess you never had an anomalous credit or insurance rating yet. (Notably GDPR has a clause for precisely the ability to review automated decision making where it can have a material effect.)
- zeth___ 9y agoAnd you think the algorithms made by racists aren't racist? This isn't a sorting algorithm where there is a simple provable solution, this is putting heuristics in a computer.
- godelski 9y ago> And you think the algorithms made by racists aren't racist? I'd hedge that a little. Because I think there are a lot of people who are just naive and think that using data that is historically racist will result in racist outcomes. Not saying that the person creating the algo is racist, but for such a harsh accusation you need enough substantial evidence. That, or we dilute the term.
- trophycase 9y agoLet's not let the ineffectiveness of this technology distract us from the fact that it is unethical. If it worked properly it would still be horrible.
- iblo66 9y agoWhy is it unethical? Because it allows us to catch criminals?
- bigiain 9y agoPerhaps because it allows "them" to catch whoever they feel like, and claim "oh, the computer said they were a criminal". I'm astounded anybody who lives in a country where police can and do get away with shooting and killing people for faulty taillights - could possibly think this is a good idea...
- iblo66 9y ago> Perhaps because it allows "them" to catch whoever they feel like, and claim "oh, the computer said they were a criminal". No because they have to prove it's the actual criminal, otherwise they have to release the guy.
- Clubber 9y agoResisting arrest is a typical workaround for that. The antidote is, "well why were you arresting him?" This dilutes the antidote with, "the computer said he was a criminal." Also, we make way too many things illegal in the US and the punishments are often harsh for a country founded on "freedom." I wish more people would shed this archaic delusion.
- iblo66 9y agoWhy would you resist arrest even if you're innocent?
- petermcneeley 9y agoPerhaps humans will reapply for this job: http://www.rcmp-grc.gc.ca/en/gazette/never-forget-a-face http://www.rcmp-grc.gc.ca/en/gazette/never-forget-a-face
- thisacctforreal 9y agoPeople with rare abilities have a chance to refuse or whistleblow unethical work. Automated facial recognition systems will listen to whatever they’re told without question, and can watch all cameras at all times. Add gait recognition and you won’t be able to go anywhere within CCTV without it being known.
- AstralStorm 9y agoNot with current failure rates where they are unable to generalise glasses, facial hair, bad lighting, different clothing etc.
- Maybestring 9y agoThere's an easy fix here. Criminalize looking like a criminal.
- bigiain 9y ago"It's a crime to be broke in America! And it's a crime to be Black in America!" -- Spearhead/Michael Franti 1994.
- rqs 9y agoStill remember few weeks ago or so people are talking about China's few jaywalk shame thing? I wonder how many people got their face, address and ID publicly displayed there for false positives, and it maybe even effects their "Social Credit Score". I believe had this kind of bold to blindly implement such system which had such false positive rates clearly showing how incredibly responsible our government really is.
- nayuki 9y agoI'm curious about how the China facial recognition experiment works too. Giving the generous assumption that the matches are correct, I wonder if they are using mobile phone tracking or RFID card reading to supplement the visual recognition system.
- rqs 9y agoI didn't found mention about supplement systems, so I don't actually know. Only thing I do found is mentions about the correct rate, some over 90%, some over 99%, none 100%.
- softbuilder 9y agoI once helped build a fingerprint processing system. One of the things I learned is that police technicians essentially photoshop fingerprints before searching to ensure the best possible match. It took a lot of the shine off of the process and made it clear that as high-tech as things seemed, there were a lot of organic factors at play that were hard or impossible for a computer to sort out by itself. I expect facial recognition has similar constraints. Edit: Conjugation
- Taniwha 9y agoWe've had a lot of press this week about the police here in New Zealand rolling out similar stuff - really we need to have a wide ranging public discussion on the issues around this stuff before it's done, sadly the police have largely been doing this under the radar. My take is that - yes we should have this stuff, but with the following caveats: 1) it only get loaded with the faces of people for whom there is an active warrant out there - ie you need a judge to sign something 2) all data collected that is not relevant should be discarded asap (including false positives the moment you know they are false) 3) you don't deploy something like this until the false positive rate is appropriately low, for everyone - it has to be able to deal equally to people of different races, genders, haircuts, with and without beards, makeup, hats, helmets, zinc oxide sunscreen, green St Paddy's day faces, actual smurfs etc etc all the diversity of normal street life
- flashman 9y agoEven if you get the false positive rate low, the low prevalence of criminals versus innocents will lead to a large number of false positives. Let's say one person in 1000 has an outstanding warrant. Let's say that when the camera sees this person, it has a 99% chance of recognising them correctly. Let's also say that it has only a 1% chance of recognising an innocent person as a suspect. Under these generous conditions, 10 out of every 11 hits will be false positives. Put another way, if you run the hypothetical system on a football match with 60,000 people in the crowd, you'll find 59 or 60 of the ones with outstanding warrants, and several hundred without. This is just the same math that was used to show how ineffective PRISM mass surveillance would be: https://bayesianbiologist.com/2013/06/06/how-likely-is-the-nsa-prism-program-to-catch-a-terrorist/ https://bayesianbiologist.com/2013/06/06/how-likely-is-the-n...
- a1369209993 9y agoI think this may be a problem of terminology. When I (and, I think, others) say "low false positive rate" what we're actually talking about is the (more useful for approximation) ratio "fraction of negatives classified positive" over "fraction of population that is positive". You have described a false positive 'rate' of one thousand percent (10 to 1). A genuinely low false positive 'rate' of 1 to 10 would require a actual-rate around 0.01% chance of recognising an innocent person as a suspect. It might help if there was widely known term for 'rate's, rather than using wrong-term-in-scare-quotes. Edit: Also, it's increasingly impossible to have a low false positive 'rate' for one-in-a-million or one-in-a-billion events, and not-technically-lying about how good your detectors are is probably a significant secondary factor in why this sort of thing gets so much flak for "staggering inaccuracy".
- IloveHN84 9y agoComputer vision is hard. Face recognition, autonomous driving, whatever works with dynamic environments will terribly fail for at least another 5 years. Recognition and classification work fine when working with static environments (pipe checking, train rail track checking, etc), but people and cars are hard to be tracked.. too many variables (light,. Day/night transition, skin tone)
- bsenftner 9y agoYes, it is very hard. That is why the leaders in the industry are not recent startups, rarely have anyone under 40 on their staff, and have been working in the field for 15 to 20 years. This is not casual technology, it is intended as a public safety technology, and the industry takes that seriously.