5 ms·

ignoring issues related to scale, I wonder how the tech's accuracy compares to london's team of human "super-recognizers."[1] 1. https://www.newyorker.com/maga

by tribby 8y ago

ignoring issues related to scale, I wonder how the tech's accuracy compares to london's team of human "super-recognizers."[1] 1. https://www.newyorker.com/magazine/2016/08/22/londons-super-recognizer-police-force https://www.newyorker.com/magazine/2016/08/22/londons-super-...

drasticmeasures 8y ago
>Ignoring scale, I wonder how the tech's accuracy compares to london's team of human "super-recognizers." You can't ignore scale when it is the game-changer of automated facial recognition. The tech compares very poorly to super-recognizers, with an absurdly high number of false positives, putting the wrong innocent people under an aggressive suspicion that they won't be able to know about and dispute.
- tribby 8y ago
  in a hypothetical you can ignore whatever you want :) infinite super-recognizers vs. infinitely available facial recognition is what I'm wondering about here. > The tech compares very poorly to super-recognizers, with an absurdly high number of false positives, putting the wrong innocent people under aggressive suspicion. well that's not good!
  - drasticmeasures 8y ago
    >well that's not good! The whole concept is a nightmare for our civil freedoms even if the tech was perfect -- precisely because of realistically being able to scale -- and must be protested until the whole idea is let go of. >infinite super-recognizers vs. infinitely available facial recognition is what I'm wondering about here. You keep editing your comments to remove contentious remarks. I can only reply to the one I've read.
    - drusepth 8y ago
      Why doesn't such a system scale? What about it makes it undesirable even if the tech was perfect?
    - tribby 8y ago
      apologies for the edits; in attempting better clarity I am achieving the opposite. > The whole concept isn't good even if the tech was perfect -- precisely because of scale -- and must be protested until the whole idea is let go of. I don't disagree with this. my questions are not political in nature, I was simply curious if the tech had approached or surpassed the accuracy of super-recognizers.
- AlphaSite 8y ago
  False positives are fine. It is basically a bloom filter. False negatives are very bad, but false positives are fine.
  - jobigoud 8y ago
    It depends on what is your opinion on the old question of do you prefer an innocent in prison or a free criminal.
    - simsla 8y ago
      If murder is bad because you're taking away the life that might've been lived by a third party, then one could argue that false incarceration is similarly grave as murder. That being said, I don't agree with your premise. Facial search/recognition is just the first step of the investigation, identifying a pool of "possible leads". This is very similar to creating a list of people who were in the neighbourhood. Lots of false positives initially, but still a good place to start looking.
  - nmstoker 8y ago
    As the other respondent says, it depends on your view of accidentally involving the innocent in police activity. In an ideal world, someone wrongly identified would just be questioned, they'd identify that there was an error and there's no physical impact just a brief disruption. That is the approach which genuinely could be beneficial to society. However, I am somewhat concerned that many police officers won't keep the fact that it's fairly likely to be wrong in the forefront of their minds. Combine that with an eagerness amongst some police to start applying "justice" immediately (rough use of handcuffing, pushing to the ground etc), at best treating people like hospital trolleys and at worst inflicting injury, and it's a recipe for escalation and injustice in scenarios involving completely innocent people (and you could argue it shouldn't happen with those who turn out to be guilty too, but that's a distraction!)
    - simsla 8y ago
      It just needs to be framed as a search problem (many results), rather than recognition (one result). E.g. "Here are the 200 people the system thinks kinda look like that picture you gave me. The one you're looking for might be in here. Then again, they might not. You take it from here. Happy detectiving. "
gaius 8y ago
I strongly suspect that “super recognisers” don’t exist and it’s just a cover story concocted at a time that they believed facial recognition worked a lot better than it does. You know, like carrots and radar.
fenwick67 8y ago
Sounds like a convenient cover for parallel construction more than anything, TBH.

by tribby 8y ago

drasticmeasures 8y ago

>Ignoring scale, I wonder how the tech's accuracy compares to london's team of human "super-recognizers." You can't ignore scale when it is the game-changer of automated facial recognition. The tech compares very poorly to super-recognizers, with an absurdly high number of false positives, putting the wrong innocent people under an aggressive suspicion that they won't be able to know about and dispute.

tribby 8y ago
in a hypothetical you can ignore whatever you want :) infinite super-recognizers vs. infinitely available facial recognition is what I'm wondering about here. > The tech compares very poorly to super-recognizers, with an absurdly high number of false positives, putting the wrong innocent people under aggressive suspicion. well that's not good!
- drasticmeasures 8y ago
  >well that's not good! The whole concept is a nightmare for our civil freedoms even if the tech was perfect -- precisely because of realistically being able to scale -- and must be protested until the whole idea is let go of. >infinite super-recognizers vs. infinitely available facial recognition is what I'm wondering about here. You keep editing your comments to remove contentious remarks. I can only reply to the one I've read.
  - drusepth 8y ago
    Why doesn't such a system scale? What about it makes it undesirable even if the tech was perfect?
  - tribby 8y ago
    apologies for the edits; in attempting better clarity I am achieving the opposite. > The whole concept isn't good even if the tech was perfect -- precisely because of scale -- and must be protested until the whole idea is let go of. I don't disagree with this. my questions are not political in nature, I was simply curious if the tech had approached or surpassed the accuracy of super-recognizers.
AlphaSite 8y ago
False positives are fine. It is basically a bloom filter. False negatives are very bad, but false positives are fine.
- jobigoud 8y ago
  It depends on what is your opinion on the old question of do you prefer an innocent in prison or a free criminal.
  - simsla 8y ago
    If murder is bad because you're taking away the life that might've been lived by a third party, then one could argue that false incarceration is similarly grave as murder. That being said, I don't agree with your premise. Facial search/recognition is just the first step of the investigation, identifying a pool of "possible leads". This is very similar to creating a list of people who were in the neighbourhood. Lots of false positives initially, but still a good place to start looking.
- nmstoker 8y ago
  As the other respondent says, it depends on your view of accidentally involving the innocent in police activity. In an ideal world, someone wrongly identified would just be questioned, they'd identify that there was an error and there's no physical impact just a brief disruption. That is the approach which genuinely could be beneficial to society. However, I am somewhat concerned that many police officers won't keep the fact that it's fairly likely to be wrong in the forefront of their minds. Combine that with an eagerness amongst some police to start applying "justice" immediately (rough use of handcuffing, pushing to the ground etc), at best treating people like hospital trolleys and at worst inflicting injury, and it's a recipe for escalation and injustice in scenarios involving completely innocent people (and you could argue it shouldn't happen with those who turn out to be guilty too, but that's a distraction!)
  - simsla 8y ago
    It just needs to be framed as a search problem (many results), rather than recognition (one result). E.g. "Here are the 200 people the system thinks kinda look like that picture you gave me. The one you're looking for might be in here. Then again, they might not. You take it from here. Happy detectiving. "

gaius 8y ago

I strongly suspect that “super recognisers” don’t exist and it’s just a cover story concocted at a time that they believed facial recognition worked a lot better than it does. You know, like carrots and radar.

fenwick67 8y ago

Sounds like a convenient cover for parallel construction more than anything, TBH.