5 ms·

Trying to ban FR seems to be a solving the privacy problem in the wrong way. Instead of banning FR, what should be fixed is the misidentification problem or the

by matz1 7y ago

Trying to ban FR seems to be a solving the privacy problem in the wrong way. Instead of banning FR, what should be fixed is the misidentification problem or the issue that come out when the FR become public.

martin_a 7y ago
This problem needs to be tackled from several angles, I think. Omnipresent cameras, linked to FR systems, are just as big a problem as misidentification.
- matz1 7y ago
  Yes i agree but instead of trying to stop/hide the information, we should instead embrace it and fix whatever issue that come out from that.
- otterley 7y ago
  Can you explain the problem in more detail?
  - martin_a 7y ago
    The problem with omnipresent cameras which are nowadays able, through facial recognition systems, to track any number of people fully automated? Which problem do you _not_ see with that?
    - otterley 7y ago
      The one making the claim is the one burdened with the justification.
      - martin_a 7y ago
        Ok, I will give you two examples, but you will need to use your brain to try to adapt the problems to other situations. Can you do that? Great! Example 1: So, let's say you're on a festival where cameras and facial recognition systems are used "for safety purposes" and everybody is feeling really safe because of that. Now some crime happens, like sexual assault or whatever. Nobody really saw the culprit and there's no usable evidence on the victim or crime scene, but cameras filmed him when he flew from the scene. The images are not the best. It was dark, he ran quite fast and he was also wearing a black hoodie. Facial recognition systems worked hard on every frame of the video material and are somewhat sure (90% probability) that you are the sexual offender. You also happen to have a black hoodie with you and you were not with your friends when the assault happened. Obviously you will tell the police that you were on your way to the toilet and that's why you split from your friends for half an hour, just as the crime happened. But what reasons do they have to believe you? The software solution is quite sure of two things: You are the suspect (90% probability) and the system found nobody else that matched the video frames. So, it can only be you, right? Well, seems like you could go to jail for some time before the real culprit is found. Or maybe you will be convicted for something that you have not done. That is a real possibility when law enforcement starts replacing "real" police work with automated solutions which are prone to errors. Will those systems even say that they are only 90% sure? Or will there be pressure on the manufacturers of those systems to only deal in absolutes and have great hit rates? Example 2: You're on this march against right-wing politicians. Police is using cameras to watch the march and is also using facial recognition technology to save who attended the march, in case something happens. Your face is detected and linked with your personal information because a reference picture of you is already present in the passport database, which the police obviously has access to. Police says that the data will be deleted after the march, but somehow it's saved for testing purposes in a MongoDB hosted on AWS and never got deleted after all. Elections come and go and now those right-wing politicians your protested against are in control. Or some hackers stumble over the database, copy it and sell it to whomever is willing to pay for that. Or some corrupt members of the police get payed a nice bribe for a copy of that list and they don't really care about "that data". Nothing of that really sounds unlikely in this day and age. In any case you are now on a list of "political enemies" because you were on that march. Nobody knows what will happen with you from here on. Maybe you will just lose your job when working in a public institution, because there is somebody more "loyal". Or you simply don't get promoted anymore, although you are doing really great work. But maybe you end up on some kind of bounty list for some ultraradical people, who knows. Just look over to Hong Kong. Would you really feel well when the Chinese government knew that you were attending the protests against the government? From here on you might think about more examples where facial recognition could go really wrong for you, although you did nothing wrong.
        otterley 7y ago
        > Ok, I will give you two examples, but you will need to use your brain to try to adapt the problems to other situations. Can you do that? Great! There's no need to be uncivil or rude. Let's take Example 1. It would indeed be problematic if we convicted people based solely on the algorithmic conclusion of a machine. But thankfully, we still have jury trials in the U.S.; it requires a unanimous verdict to convict someone; and it requires a level of proof "beyond a reasonable doubt." It's important to understand that a machine is not a witness. It can't be cross-examined. So whatever result it produces is evidence. How strongly that evidence is believed is up to the jury; and a good defense attorney is going to do whatever it takes to attack the machine (using expert witnesses and so forth) to introduce reasonable doubt if its conclusion is incorrect. The defense attorney is also going to introduce whatever other exculpatory facts and witnesses they can find. In your example, if a sexual assault occurs, the victim is likely to testify. Without the victim's testimony identifying the defendant, and without any corroborating DNA evidence, the defendant is likely to walk, even with the machine saying it's 90%+ probability match. In any event, I think what's needed here is some additional law forbidding the sole use of FR data as evidence used to convict a defendant. That, I think, would help assuage most people's concerns -- even though I think it unlikely that a conviction would result this way anyway. A really important fact that you're not considering is that FR has a strong potential to prevent false identification. Many studies have shown that eyewitness identifications, including line-ups, are extremely unreliable; and yet, we've been using them for decades to convict people. And we've released innocent convicts many, many times, sometimes decades after imprisonment, when it turned out the witness was lying or simply had a faulty memory. And in your example, if you in fact went to the toilet during the crime, the cameras would be able to corroborate that. OTOH, if a machine strongly disagreed with the eyewitness, I have little doubt that the defendant would be dismissed. Example 2 is a bit trickier. You don't need FR technology to have a police state: North Korea has had one for nearly 70 years without one. East Germany had one for over 40 years (see: Stasi). There are cruder methods than using FR technology; spying is a very old game. So, FR can help enhance control for regimes that are already inclined to preventing freedom by making it much more economical than having a spy network, but it's by no means necessary. Besides, you can trivially foil FR by wearing a mask. I don't know how to prevent a police state, other than to employ violent resistance, and even that doesn't always work. I don't think anybody knows. But if you want a prevent a police state, I think banning FR isn't really going to help. At least so far, history shows that many police states resulted from democracies or republics ceding control under some sort of "necessity."

by matz1 7y ago

martin_a 7y ago

This problem needs to be tackled from several angles, I think. Omnipresent cameras, linked to FR systems, are just as big a problem as misidentification.

matz1 7y ago
Yes i agree but instead of trying to stop/hide the information, we should instead embrace it and fix whatever issue that come out from that.
otterley 7y ago
Can you explain the problem in more detail?
- martin_a 7y ago
  The problem with omnipresent cameras which are nowadays able, through facial recognition systems, to track any number of people fully automated? Which problem do you _not_ see with that?
  - otterley 7y ago
    The one making the claim is the one burdened with the justification.
    - martin_a 7y ago
      Ok, I will give you two examples, but you will need to use your brain to try to adapt the problems to other situations. Can you do that? Great! Example 1: So, let's say you're on a festival where cameras and facial recognition systems are used "for safety purposes" and everybody is feeling really safe because of that. Now some crime happens, like sexual assault or whatever. Nobody really saw the culprit and there's no usable evidence on the victim or crime scene, but cameras filmed him when he flew from the scene. The images are not the best. It was dark, he ran quite fast and he was also wearing a black hoodie. Facial recognition systems worked hard on every frame of the video material and are somewhat sure (90% probability) that you are the sexual offender. You also happen to have a black hoodie with you and you were not with your friends when the assault happened. Obviously you will tell the police that you were on your way to the toilet and that's why you split from your friends for half an hour, just as the crime happened. But what reasons do they have to believe you? The software solution is quite sure of two things: You are the suspect (90% probability) and the system found nobody else that matched the video frames. So, it can only be you, right? Well, seems like you could go to jail for some time before the real culprit is found. Or maybe you will be convicted for something that you have not done. That is a real possibility when law enforcement starts replacing "real" police work with automated solutions which are prone to errors. Will those systems even say that they are only 90% sure? Or will there be pressure on the manufacturers of those systems to only deal in absolutes and have great hit rates? Example 2: You're on this march against right-wing politicians. Police is using cameras to watch the march and is also using facial recognition technology to save who attended the march, in case something happens. Your face is detected and linked with your personal information because a reference picture of you is already present in the passport database, which the police obviously has access to. Police says that the data will be deleted after the march, but somehow it's saved for testing purposes in a MongoDB hosted on AWS and never got deleted after all. Elections come and go and now those right-wing politicians your protested against are in control. Or some hackers stumble over the database, copy it and sell it to whomever is willing to pay for that. Or some corrupt members of the police get payed a nice bribe for a copy of that list and they don't really care about "that data". Nothing of that really sounds unlikely in this day and age. In any case you are now on a list of "political enemies" because you were on that march. Nobody knows what will happen with you from here on. Maybe you will just lose your job when working in a public institution, because there is somebody more "loyal". Or you simply don't get promoted anymore, although you are doing really great work. But maybe you end up on some kind of bounty list for some ultraradical people, who knows. Just look over to Hong Kong. Would you really feel well when the Chinese government knew that you were attending the protests against the government? From here on you might think about more examples where facial recognition could go really wrong for you, although you did nothing wrong.
      - otterley 7y ago
        > Ok, I will give you two examples, but you will need to use your brain to try to adapt the problems to other situations. Can you do that? Great! There's no need to be uncivil or rude. Let's take Example 1. It would indeed be problematic if we convicted people based solely on the algorithmic conclusion of a machine. But thankfully, we still have jury trials in the U.S.; it requires a unanimous verdict to convict someone; and it requires a level of proof "beyond a reasonable doubt." It's important to understand that a machine is not a witness. It can't be cross-examined. So whatever result it produces is evidence. How strongly that evidence is believed is up to the jury; and a good defense attorney is going to do whatever it takes to attack the machine (using expert witnesses and so forth) to introduce reasonable doubt if its conclusion is incorrect. The defense attorney is also going to introduce whatever other exculpatory facts and witnesses they can find. In your example, if a sexual assault occurs, the victim is likely to testify. Without the victim's testimony identifying the defendant, and without any corroborating DNA evidence, the defendant is likely to walk, even with the machine saying it's 90%+ probability match. In any event, I think what's needed here is some additional law forbidding the sole use of FR data as evidence used to convict a defendant. That, I think, would help assuage most people's concerns -- even though I think it unlikely that a conviction would result this way anyway. A really important fact that you're not considering is that FR has a strong potential to prevent false identification. Many studies have shown that eyewitness identifications, including line-ups, are extremely unreliable; and yet, we've been using them for decades to convict people. And we've released innocent convicts many, many times, sometimes decades after imprisonment, when it turned out the witness was lying or simply had a faulty memory. And in your example, if you in fact went to the toilet during the crime, the cameras would be able to corroborate that. OTOH, if a machine strongly disagreed with the eyewitness, I have little doubt that the defendant would be dismissed. Example 2 is a bit trickier. You don't need FR technology to have a police state: North Korea has had one for nearly 70 years without one. East Germany had one for over 40 years (see: Stasi). There are cruder methods than using FR technology; spying is a very old game. So, FR can help enhance control for regimes that are already inclined to preventing freedom by making it much more economical than having a spy network, but it's by no means necessary. Besides, you can trivially foil FR by wearing a mask. I don't know how to prevent a police state, other than to employ violent resistance, and even that doesn't always work. I don't think anybody knows. But if you want a prevent a police state, I think banning FR isn't really going to help. At least so far, history shows that many police states resulted from democracies or republics ceding control under some sort of "necessity."