3 ms·

Genuine question, are the ethics of "anonymous facial recognition", where the system keeps track of a particular face, does not remember the face, and does not

Genuine question, are the ethics of "anonymous facial recognition", where the system keeps track of a particular face, does not remember the face, and does not correlate the face to a identity , comparable to a social-credit like system?

godelski 2y ago
> anonymous facial recognition I don't have an answer to the specific question, but I have experience working on single and multi-camera person tracking systems and I can tell you that they highly focus on the faces for recognition[0]. There is no such thing as anonymous facial recognition or even person tracking. Data creeps in through things that can be later extracted, though you can make this process more difficult and that's about the best you can do. For a non-technical intuition of this you might notice that you can often recognize a friend walking away from you (no visible face) down the street. Even if you don't know why you can recognize them, it is down to subtle things like the way they walk, dress, hair style, etc. It might be harder to identify people this way and more prone to mistakes, but even when less accurate you could significantly narrow down possible candidates (even more with correlating information like location and time). [0] This was a surprising thing to me because I suspected clothing would be a stronger indicator especially when face isn't always visible but this could also happen for other reasons and can be highly influenced in pretraining. Even saw it with blurred faces in testing, but depended on how these were blurred. Gate tracking also seemed to be picked up but I never confirmed.
- samarthr1 2y ago
  I see. The reason I ask is that india has a bit of a littering problem at bus stands, and the fines for littering are hardly enforced due to lack of budget. I was hoping to fix the issue by applying a small fine, all the time. To do it, I was hoping I could build a sort of person-hood tracking system that says, see, that fellow who threw a packet of chips is now standing at platform 10, near the concession stand, go catch him, easy 100₹. By consistently enforcing penalties, we can change the behavior of people. It is worse than useless if it is inconsistently applied, for those caught will be likely to reoffend, considering it plain chance that they got caught.
  - godelski 2y ago
    Yeah I think this is kinda the danger of these types of systems, and why I don't want to work on them anymore. While I think there are a lot of good uses to this tech, it also has the power to do much more harm and the way I've seen it used even in democratic societies makes me uncomfortable about authoritarian overreach. I could not see a way this can be used in a privacy preserving manner, even for things as simple as counting number of patrons in a store[0]. Right now, I don't feel comfortable furthering this technology in the US, not to mention how I saw it being used elsewhere[1]. I'm sure if you take some time thinking about how the tech could be abused you would find a lot of ways. But it is up to you to decide if it is worth it after doing that exercise (your mind might even change later). I think this is one of the difficult things about creating things. It is easy to get caught up in the benefits of a technology and how it can be used for good. But technology isn't good or bad, it is a tool. Like any tool, you need to consider how it will be used, and that means how it will be abused. My undergrad was in physics, and this lesson is often taught when you get to atomics (or maybe because my advisor worked at Los Alamos and with many of the people who built the atomic bombs). Physics tends to not hide the skeletons in the closet because atomic weapons. But I think CS still does and maybe we haven't caught up to how influential our work is in modern day. How software can (and does) kill people. Or maybe it is the abstraction of software, working at a high level it is easy to feel disconnected. Either way, I think we need to tread more carefully and just make sure we're concerned about ethics. It is a personal decision where the lines are drawn, but they can't be ignored. After all, the road to hell is paved with good intentions, right? Unfortunately evil isn't only created by evil men, but often good men with good intentions. FWIW, I'm all for the intent. But I question if there are other ways to solve these issues. But that may require a more complex and nuanced approach[2]. But if enforcement is already a problem, then I think there are bigger problems at play that this technical solution might make worse. Unfortunately as we advance the world gets more complex haha [0] You can get pretty good if you do upfront processing to blur people, but that first means you need to detect people lol. So this would need to be done on device (as in, the camera itself. Which is possible btw). [1] I saw a lot of videos from India (as well as the whole subcontinent and South East Asia), where the customers were trying to prevent theft (good usage), but there were some other things that had me concerned and were being hinted at. I also think one big difference is that I appeared to be the only person on my team that would look at data we were processing. Not to snoop or be invasive, but because this is the best way to solve data processing problems; you need to check your assumptions, learn what is creating errors, find things you didn't consider, and all that. To be good at ML you can't just pipe in data and expect results, and you shouldn't treat the algorithms like black boxes. While they aren't transparent ("white boxes"), that doesn't mean they're completely opaque either (and we're learning more every day). [2] I think one key piece of information is that we need to ask why things like littering and theft is less common in richer neighborhoods. Or at least by their residents. Personally, my belief is that this is more to do with a personal safety net. It's easy for ethics to be eroded when you're worried about your next meal or being able to afford rent. These are obviously harder issues to resolve, but I think they are worth addressing. Because I think lifting up the quality of life of the lowest class of people benefits everyone. A rising tide lifts all ships, and I do believe it even lifts the biggest ships more.

godelski 2y ago

> anonymous facial recognition I don't have an answer to the specific question, but I have experience working on single and multi-camera person tracking systems and I can tell you that they highly focus on the faces for recognition[0]. There is no such thing as anonymous facial recognition or even person tracking. Data creeps in through things that can be later extracted, though you can make this process more difficult and that's about the best you can do. For a non-technical intuition of this you might notice that you can often recognize a friend walking away from you (no visible face) down the street. Even if you don't know why you can recognize them, it is down to subtle things like the way they walk, dress, hair style, etc. It might be harder to identify people this way and more prone to mistakes, but even when less accurate you could significantly narrow down possible candidates (even more with correlating information like location and time). [0] This was a surprising thing to me because I suspected clothing would be a stronger indicator especially when face isn't always visible but this could also happen for other reasons and can be highly influenced in pretraining. Even saw it with blurred faces in testing, but depended on how these were blurred. Gate tracking also seemed to be picked up but I never confirmed.

samarthr1 2y ago
I see. The reason I ask is that india has a bit of a littering problem at bus stands, and the fines for littering are hardly enforced due to lack of budget. I was hoping to fix the issue by applying a small fine, all the time. To do it, I was hoping I could build a sort of person-hood tracking system that says, see, that fellow who threw a packet of chips is now standing at platform 10, near the concession stand, go catch him, easy 100₹. By consistently enforcing penalties, we can change the behavior of people. It is worse than useless if it is inconsistently applied, for those caught will be likely to reoffend, considering it plain chance that they got caught.
- godelski 2y ago
  Yeah I think this is kinda the danger of these types of systems, and why I don't want to work on them anymore. While I think there are a lot of good uses to this tech, it also has the power to do much more harm and the way I've seen it used even in democratic societies makes me uncomfortable about authoritarian overreach. I could not see a way this can be used in a privacy preserving manner, even for things as simple as counting number of patrons in a store[0]. Right now, I don't feel comfortable furthering this technology in the US, not to mention how I saw it being used elsewhere[1]. I'm sure if you take some time thinking about how the tech could be abused you would find a lot of ways. But it is up to you to decide if it is worth it after doing that exercise (your mind might even change later). I think this is one of the difficult things about creating things. It is easy to get caught up in the benefits of a technology and how it can be used for good. But technology isn't good or bad, it is a tool. Like any tool, you need to consider how it will be used, and that means how it will be abused. My undergrad was in physics, and this lesson is often taught when you get to atomics (or maybe because my advisor worked at Los Alamos and with many of the people who built the atomic bombs). Physics tends to not hide the skeletons in the closet because atomic weapons. But I think CS still does and maybe we haven't caught up to how influential our work is in modern day. How software can (and does) kill people. Or maybe it is the abstraction of software, working at a high level it is easy to feel disconnected. Either way, I think we need to tread more carefully and just make sure we're concerned about ethics. It is a personal decision where the lines are drawn, but they can't be ignored. After all, the road to hell is paved with good intentions, right? Unfortunately evil isn't only created by evil men, but often good men with good intentions. FWIW, I'm all for the intent. But I question if there are other ways to solve these issues. But that may require a more complex and nuanced approach[2]. But if enforcement is already a problem, then I think there are bigger problems at play that this technical solution might make worse. Unfortunately as we advance the world gets more complex haha [0] You can get pretty good if you do upfront processing to blur people, but that first means you need to detect people lol. So this would need to be done on device (as in, the camera itself. Which is possible btw). [1] I saw a lot of videos from India (as well as the whole subcontinent and South East Asia), where the customers were trying to prevent theft (good usage), but there were some other things that had me concerned and were being hinted at. I also think one big difference is that I appeared to be the only person on my team that would look at data we were processing. Not to snoop or be invasive, but because this is the best way to solve data processing problems; you need to check your assumptions, learn what is creating errors, find things you didn't consider, and all that. To be good at ML you can't just pipe in data and expect results, and you shouldn't treat the algorithms like black boxes. While they aren't transparent ("white boxes"), that doesn't mean they're completely opaque either (and we're learning more every day). [2] I think one key piece of information is that we need to ask why things like littering and theft is less common in richer neighborhoods. Or at least by their residents. Personally, my belief is that this is more to do with a personal safety net. It's easy for ethics to be eroded when you're worried about your next meal or being able to afford rent. These are obviously harder issues to resolve, but I think they are worth addressing. Because I think lifting up the quality of life of the lowest class of people benefits everyone. A rising tide lifts all ships, and I do believe it even lifts the biggest ships more.