7 ms·

If the basis of banning facial recognition technology is its poor accuracy, will facial recognition technology be unbanned if it is 99% accurate for everyone?

colechristensen 6y ago
If you assume the best then poor accuracy is just the easiest justification not necessarily the only, the best, or the most important. I’ll take it. There are narrow use cases where i think facial recognition in law enforcement would be a good thing, but it is ultimately too much power and too easily abused to trust legal systems to adhere to right usage. Banning is the correct action when misuse is as bad and correct use as complex as they are.
raverbashing 6y ago
> if it is 99% accurate for everyone? Think 99.99% and even then 99% accuracy in a city setting means (more or less) that you'll get the wrong person 1 in 100 times. If you're scanning a crowd with this, you can imagine what can happen
badrabbit 6y ago
Evem 99.999% is not good enough be enough because you will have at least one person that is guaranteed a false arrest and prosecution (which 90%+ prosecuted take a plea deal). There is inaccuracy with other methods as well but you have humans being held accountable. When it comes to justice,mistakes are tolerable so long as adequate compensatiom exists but when the mistake is systemic it becomes intolerable due to the preexisting guarantee of a mistake as opposed to a specific human making an error as a matter of chance. This is all exacerbated by other systemic cruelties of the US justice system where even an arrest and release for no cause means days if not weeks of imprisonment and if charged most people accept a plea bargain deal regardless of actual guilt. It's better to let actual guilty criminals get away than explicitly and systemically accept even one innocent person being punished incorrectly, because among other reasons, the justice system has legitimacy because its goal is to administer justice, accepting any amount of injustice invalidates that legitimacy and authority. A good analogy would be a chef tolerating fecal matter in their food. Yes, there is always some small chance of that happening, but no one accepts food from s chef that explicitly treats toilet water and claims 99% of the fecal matter is gone and only one in 100 people will get sick from it.
- winrid 6y ago
  99.99% is probably better than actual cops right? Also there is the idea that such an infastructure would scare criminals into not doing incriminating things.
  - badrabbit 6y ago
    Thats what I said, better than humans is not good enough. Humans can make mistakes but they can be held accountable and their mistake is not an accepted outcome of the system. To put it differently, facial recongition would be compared with explicitly accepting only low IQ people for cop jobs (which some PDs do!) as opposed to hiring the best people and expecting them to have 100% accuracy. Their mistake is a matter of judgement and negligence. Facial recongition's mistakes are a matter of the system accepting the mistake. Technical people often focus on the pure metrics but as in engineering,why faults happen is important. The metric should never be the goal. To put it differently, explicitly accepting injustice is incompatible with mine and I believe society's understanding of justice. This is the same reason why ML shouldn't be used to replace human judges regardless of improvements in outcome. The authority of judges,cops and legislators is legitimate because the people have a contract with them where the people's view of justice is implemented as law in return the people subject themselves to these laws. Facial recogntion and ML judges violate that contract and as a result the outcome is illegitimate,100% of convicts become wrongly convicted because as a people the views on justice of their society was nor enforced by a member of their society that is elected or appointed by the people but by a blackbox logic that only looks at the outcome as a metric. A defense attorney can claim eyewitness testimony is inaccurate for example and cross examine witnesses but that attorney cannot interrogate a ML presenting as evidence a blurry picture with a 99% accuracy match.
    - winrid 6y ago
      It seems a lot of arguments like this are based on immediate conviction. I'm totally not for that. I think your identity should be verified using other means before being arrested when possible.
raxxorrax 6y ago
I hope not. I am ok with it as long as we don't plaster the whole public space with video cameras. That would be more than unfortunate. I doubt most places would gain anything from it. I am not keen of my state knowing my whereabouts either and think camera deployment would create more problems than it solves. Countries where it has been deployed don't have impressive advantages to show for and all the privacy disadvantages. So why should we even consider it? What problem should it solve? My opinion, if you are still scared of terrorists a psychiatrist might be more useful than a camera.
- CaptArmchair 6y ago
  Facial recognition isn't that different from fingerprints or DNA traces. It's a comparison against a database of biometric data. It used to be that law enforcement relied on fingerprints, eye-witnesses or a video tape backed by a judicial and legal system that - more or less - severely limited how and when that information could be collected and used e.g. within the context of discrete on-going criminal investigations. This balance has been shifting entirely over the few short years. Now, video camera's, fingerprinting and so on are collected pre-emptively; on literally everyone. Undeniably, it's easier to move forward on an investigation if you already have a lot of the data at hand. The argument put forward by proponents is that speeding up investigations saves lives. But that comes with a ton of hard problems. First, searching accurately through a vast mountain of diffuse information is a hard problem. Law enforcement has outsourced that part to private companies, which opens up a can of worms in terms of confidentiality and privacy. Second, searching through such a database becomes a black box as far as law enforcement is concerned: just upload a picture or a video fragment and it will easily yield any result. The fallacy being that convenience lulls LE into assuming that the results are accurate enough at face value. No need for verification or critical thinking (Much like you'd accept the results yielded by Google Search at face value). Third, it's a convenient way to offshore responsibility towards the public. Law enforcement didn't make the wrong assertion if the wrong person ends up getting prosecuted and convicted: the data was just "not good enough". The fallacy here is that data gets treated as a commodity, which it is anything but. Fourth, there's a difference between what's morally right and what's legal. The latter changes depending on how defines public governance. A database may be a great idea, until control is ceded to someone who uses that data against the public. There's actually a historic precedent here: the 1943 bombing of the Amsterdam Civil Registry office. Following the 1940 German invasion of the Netherlands, all Dutch Jews had to carry a mandatory identity card and there whereabouts were recorded centrally at the Civil Registry. This was easy as, prior to the war, people's religious denomination had been recorded. The Dutch Resistance understood the importance and the downsides of a centralised record containing private information falling into the wrong hands. And so, they ended up attacking and demolishing the Civil Registry office containing that record. Albeit half-successful. https://en.wikipedia.org/wiki/1943_bombing_of_the_Amsterdam_civil_registry_office https://en.wikipedia.org/wiki/1943_bombing_of_the_Amsterdam_... Obviously, I don't advocate attacking datacenters - that would be an effort in futility anyway - but your assertion that investing in mental health support would be more helpful rings very much true.
Veen 6y ago
99% accurate would still lead to a huge number of false positives because most people aren't criminals. The base rate of non-criminals is so much larger than the number of criminals that there would be hundreds of thousands or millions of false positives, depending on how widely facial recognition was deployed. There would be many times more false positives than accurate identifications.
- pbhjpbhj 6y ago
  It's a tool, not the only tool. If you have a population of 100M, even if 1M match, then you add geographic area, skin colour, hair colour/style, height, known criminality, ... suddenly you're down to enough people you can investigate. Sure, if your police/justice system is not competent you have a problem, but you had that in the first place.

colechristensen 6y ago

If you assume the best then poor accuracy is just the easiest justification not necessarily the only, the best, or the most important. I’ll take it. There are narrow use cases where i think facial recognition in law enforcement would be a good thing, but it is ultimately too much power and too easily abused to trust legal systems to adhere to right usage. Banning is the correct action when misuse is as bad and correct use as complex as they are.

raverbashing 6y ago

> if it is 99% accurate for everyone? Think 99.99% and even then 99% accuracy in a city setting means (more or less) that you'll get the wrong person 1 in 100 times. If you're scanning a crowd with this, you can imagine what can happen

badrabbit 6y ago

Evem 99.999% is not good enough be enough because you will have at least one person that is guaranteed a false arrest and prosecution (which 90%+ prosecuted take a plea deal). There is inaccuracy with other methods as well but you have humans being held accountable. When it comes to justice,mistakes are tolerable so long as adequate compensatiom exists but when the mistake is systemic it becomes intolerable due to the preexisting guarantee of a mistake as opposed to a specific human making an error as a matter of chance. This is all exacerbated by other systemic cruelties of the US justice system where even an arrest and release for no cause means days if not weeks of imprisonment and if charged most people accept a plea bargain deal regardless of actual guilt. It's better to let actual guilty criminals get away than explicitly and systemically accept even one innocent person being punished incorrectly, because among other reasons, the justice system has legitimacy because its goal is to administer justice, accepting any amount of injustice invalidates that legitimacy and authority. A good analogy would be a chef tolerating fecal matter in their food. Yes, there is always some small chance of that happening, but no one accepts food from s chef that explicitly treats toilet water and claims 99% of the fecal matter is gone and only one in 100 people will get sick from it.

winrid 6y ago
99.99% is probably better than actual cops right? Also there is the idea that such an infastructure would scare criminals into not doing incriminating things.
- badrabbit 6y ago
  Thats what I said, better than humans is not good enough. Humans can make mistakes but they can be held accountable and their mistake is not an accepted outcome of the system. To put it differently, facial recongition would be compared with explicitly accepting only low IQ people for cop jobs (which some PDs do!) as opposed to hiring the best people and expecting them to have 100% accuracy. Their mistake is a matter of judgement and negligence. Facial recongition's mistakes are a matter of the system accepting the mistake. Technical people often focus on the pure metrics but as in engineering,why faults happen is important. The metric should never be the goal. To put it differently, explicitly accepting injustice is incompatible with mine and I believe society's understanding of justice. This is the same reason why ML shouldn't be used to replace human judges regardless of improvements in outcome. The authority of judges,cops and legislators is legitimate because the people have a contract with them where the people's view of justice is implemented as law in return the people subject themselves to these laws. Facial recogntion and ML judges violate that contract and as a result the outcome is illegitimate,100% of convicts become wrongly convicted because as a people the views on justice of their society was nor enforced by a member of their society that is elected or appointed by the people but by a blackbox logic that only looks at the outcome as a metric. A defense attorney can claim eyewitness testimony is inaccurate for example and cross examine witnesses but that attorney cannot interrogate a ML presenting as evidence a blurry picture with a 99% accuracy match.
  - winrid 6y ago
    It seems a lot of arguments like this are based on immediate conviction. I'm totally not for that. I think your identity should be verified using other means before being arrested when possible.

raxxorrax 6y ago

I hope not. I am ok with it as long as we don't plaster the whole public space with video cameras. That would be more than unfortunate. I doubt most places would gain anything from it. I am not keen of my state knowing my whereabouts either and think camera deployment would create more problems than it solves. Countries where it has been deployed don't have impressive advantages to show for and all the privacy disadvantages. So why should we even consider it? What problem should it solve? My opinion, if you are still scared of terrorists a psychiatrist might be more useful than a camera.

CaptArmchair 6y ago
Facial recognition isn't that different from fingerprints or DNA traces. It's a comparison against a database of biometric data. It used to be that law enforcement relied on fingerprints, eye-witnesses or a video tape backed by a judicial and legal system that - more or less - severely limited how and when that information could be collected and used e.g. within the context of discrete on-going criminal investigations. This balance has been shifting entirely over the few short years. Now, video camera's, fingerprinting and so on are collected pre-emptively; on literally everyone. Undeniably, it's easier to move forward on an investigation if you already have a lot of the data at hand. The argument put forward by proponents is that speeding up investigations saves lives. But that comes with a ton of hard problems. First, searching accurately through a vast mountain of diffuse information is a hard problem. Law enforcement has outsourced that part to private companies, which opens up a can of worms in terms of confidentiality and privacy. Second, searching through such a database becomes a black box as far as law enforcement is concerned: just upload a picture or a video fragment and it will easily yield any result. The fallacy being that convenience lulls LE into assuming that the results are accurate enough at face value. No need for verification or critical thinking (Much like you'd accept the results yielded by Google Search at face value). Third, it's a convenient way to offshore responsibility towards the public. Law enforcement didn't make the wrong assertion if the wrong person ends up getting prosecuted and convicted: the data was just "not good enough". The fallacy here is that data gets treated as a commodity, which it is anything but. Fourth, there's a difference between what's morally right and what's legal. The latter changes depending on how defines public governance. A database may be a great idea, until control is ceded to someone who uses that data against the public. There's actually a historic precedent here: the 1943 bombing of the Amsterdam Civil Registry office. Following the 1940 German invasion of the Netherlands, all Dutch Jews had to carry a mandatory identity card and there whereabouts were recorded centrally at the Civil Registry. This was easy as, prior to the war, people's religious denomination had been recorded. The Dutch Resistance understood the importance and the downsides of a centralised record containing private information falling into the wrong hands. And so, they ended up attacking and demolishing the Civil Registry office containing that record. Albeit half-successful. https://en.wikipedia.org/wiki/1943_bombing_of_the_Amsterdam_civil_registry_office https://en.wikipedia.org/wiki/1943_bombing_of_the_Amsterdam_... Obviously, I don't advocate attacking datacenters - that would be an effort in futility anyway - but your assertion that investing in mental health support would be more helpful rings very much true.

Veen 6y ago

99% accurate would still lead to a huge number of false positives because most people aren't criminals. The base rate of non-criminals is so much larger than the number of criminals that there would be hundreds of thousands or millions of false positives, depending on how widely facial recognition was deployed. There would be many times more false positives than accurate identifications.

pbhjpbhj 6y ago
It's a tool, not the only tool. If you have a population of 100M, even if 1M match, then you add geographic area, skin colour, hair colour/style, height, known criminality, ... suddenly you're down to enough people you can investigate. Sure, if your police/justice system is not competent you have a problem, but you had that in the first place.