6 ms·

Usually these systems are implemented via a generic face detector that first runs on the whole image, to locate all the faces that are present. After the faces

Usually these systems are implemented via a generic face detector that first runs on the whole image, to locate all the faces that are present. After the faces are located, they’re matched against a set of candidates (which I would guess is an enumeration if all your Nth order friends [who don’t opt out of this], to reduce computational complexity). This isn’t the only possible way to do it, but I believe it’s the most common and effective approach. So it’s not at all unlikely that when running the identifier, they simply don’t match against people who say no to this. They’d never be recognized at all, and the system would just identify their face as an “unknown” identity. I cannot say for sure whether or not this is how they implemented it, but it seems likely.

bjt2n3904 9y ago
Anecdotal, but years ago I uploaded a picture of a friend. He's a bit of a hermit, and doesn't have social media accounts. The picture scrolled by several times that week, and Facebook asked me to tag who this was. (This was perhaps... 2010?) I uploaded perhaps twenty or so photos, but that was the only one they asked about. I haven't seen this for a while, but I used to get my friends pictures scrolling by. (Not even my own!) "Tag Becky in this photo? Yes/No". Clicking yes confirms what Facebook already knows. Clicking No tells them... what? That they were right, and it is Becky, but I don't want to tag her? What does not responding to the prompt mean?
- bitL 9y ago
  If you want to fight it, tag them as wrong persons to mess up their training dataset. I am sure they would love that! :) Maybe even start a Facebook group "False positives" where people would do this for fun and to "stick it to the man" lol
  - bjt2n3904 9y ago
    I'm sure they're on the lookout for that, as well. If they have enough confidence in their model, I'm sure they'd raise an eyebrow at something that only has 30% confidence, say "thanks", and then simply flag my account to not be trusted to train their models.
    - bitL 9y ago
      That still sounds like win. If it gets momentum, they won't be able to recognize any new additions due to them being not trusted. Physique changes significantly roughly every 10 years as well, so if all they have are outdated photos... OK, Deep Reinforcement Learning might be used on reconstructed 3D models from younger photos to simulate aging process, but it might not be very accurate even if looking very realistic.
    - badwolf 9y ago
      Similarly to "suggested edits" for pages/businesses. You have an accuracy score as an editor based on how many of your suggested edits get accepted/approved. I'd be willing to bet my hat that they have a similar scoring for approving suggested tags - don't click them/ignore them complete for a few days and you'll probably start seeing that suggesting a lot less.
  - craigsmansion 9y ago
    If you want to fight it: don't use Facebook! No matter how much an individual thinks they're resisting or messing with the system, they're not. The only way to win is to not participate.
    - TremendousJudge 9y ago
      if other people still upload pictures of you to Facebook it's the same thing
      - gaius 9y ago
        Just as, if you’re ever in an e-mail chain that includes just 1 person on GMail.
    - bitL 9y ago
      I don't use Facebook myself for a few years already, which is paradoxical as they were headhunting me recently to lead one of their ML teams and didn't mind that I quit them long time ago :D But maybe they were just harvesting my own "outlier" training dataset - who knows if I was filmed while onsite and now my gait is analyzed to smallest details including reconstructed skeleton and muscle attachments, my appearance is predicted for the next 30 years under various aging assumptions, maybe later used to fabricate realistically faked video etc. These things are now possible and not that difficult anymore.
      - jacquesm 9y ago
        I don't understand this. Why would you want to interview with a company whose core product you do not support?
        bitL 9y ago
        Why not? You keep up to date, experiencing world-class hiring process and its requirements (IMO FB's hiring process is the best I've ever experienced, Google's feels one level down), you meet interesting people you might later work together on something cool either there or at another company, you visit interesting places, get some inspirations for your own businesses, interviews might be also a lot of fun (one interview I had at FB was an hour of one-joke-after-another while discussing/inventing architecture of some scalable system; imagine Cryptotinder which is a joke on its own) etc.
  - John_KZ 9y ago
    It doesn't work that way. They can see that you tag wrong too many people, and you're obviously an outlier, so your data is rejected. The only way you can "win" is by deleting facebook and pushing for better data protection laws.
    - bitL 9y ago
      You won't do it alone, but if most of your friends start doing that, either you mess up their training dataset by introducing a lot of noise, or you all will get flagged as outliers. Both situations are win comparing to learned helplessness one might fall into when realizing the ways how this data can be misused.
      - John_KZ 9y ago
        Right now the systems in public domain reach 97% accuracy. Facebook doesn't need you, even if you get a million people to try to "trick" the system, you won't do anything. The verification is to fine-tune it. By the way, you forget that Facebook can simply exclude almost all those who know about this. What are you going to do? Do you think nobody will notice the 10M user fan page themed "Fool facebooks AI"? It just knows an extra thing about you, that you don't want to be tracked, so it'll hide news about tracking from your feed.
        bitL 9y ago
        We already know Uber is cherry-picking users it suspects as working for competitors, so precedent is set for what you are suggesting. It's doable obviously, and if it's doable, companies have already either implemented it or having somebody working on it.
    - gaius 9y ago
      by deleting facebook They still have your photos in their training set
      - John_KZ 9y ago
        You can't undo the past but you can safeguard the future. Besides, if facebook goes bankrupt with heavy legislation demanding the destruction of user data and not allowing their exchange as assets, they will eventually be deleted, along with their training set and everything else. Your biometrics aren't coming back, but at least only the government will have them.
kilotaras 9y ago
> After the faces are located, they’re matched against a set of candidates (which I would guess is an enumeration if all your Nth order friends [who don’t opt out of this], to reduce computational complexity. Face recognition is usually done by first transforming the face into vector space and than looking for the closest vector from known, e.g. openface[1] (which is based on google 2015 paper "FaceNet: A Unified Embedding for Face Recognition and Clustering"). Facebook itself opensourced[2] a library for efficient search of nearest neighbors in such spaces. It doesn't seem likely that generating set of Nth order friends (which grow very quickly with N) would be better than just having a global set or a split of global set by some characteristics (e.g. gender, race). [1] https://cmusatyalab.github.io/openface/ https://cmusatyalab.github.io/openface/ [2] https://github.com/facebookresearch/faiss https://github.com/facebookresearch/faiss
- wpietri 9y ago
  Wouldn't the betterness of using friend networks being reduced false positives? Facebook has 2 billion MAU, which means they have a lot of faces. Filtering down to the 50k or so that are close to you means it's much easier to find the right person, rather than a similar stranger.
  - sangnoir 9y ago
    > Wouldn't the betterness of using friend networks being reduced false positives? They are not limiting to friend networks, otherwise they would not catch (out of network) strangers using your pics. One of the use cases touted is "Help protect you from strangers using your photo"
    - tekknik 9y ago
      Also finding images you’re in but may have been taken by someone not on your friends list (think events). Eventually probably recommending friends from these images.

bjt2n3904 9y ago

Anecdotal, but years ago I uploaded a picture of a friend. He's a bit of a hermit, and doesn't have social media accounts. The picture scrolled by several times that week, and Facebook asked me to tag who this was. (This was perhaps... 2010?) I uploaded perhaps twenty or so photos, but that was the only one they asked about. I haven't seen this for a while, but I used to get my friends pictures scrolling by. (Not even my own!) "Tag Becky in this photo? Yes/No". Clicking yes confirms what Facebook already knows. Clicking No tells them... what? That they were right, and it is Becky, but I don't want to tag her? What does not responding to the prompt mean?

bitL 9y ago
If you want to fight it, tag them as wrong persons to mess up their training dataset. I am sure they would love that! :) Maybe even start a Facebook group "False positives" where people would do this for fun and to "stick it to the man" lol
- bjt2n3904 9y ago
  I'm sure they're on the lookout for that, as well. If they have enough confidence in their model, I'm sure they'd raise an eyebrow at something that only has 30% confidence, say "thanks", and then simply flag my account to not be trusted to train their models.
  - bitL 9y ago
    That still sounds like win. If it gets momentum, they won't be able to recognize any new additions due to them being not trusted. Physique changes significantly roughly every 10 years as well, so if all they have are outdated photos... OK, Deep Reinforcement Learning might be used on reconstructed 3D models from younger photos to simulate aging process, but it might not be very accurate even if looking very realistic.
  - badwolf 9y ago
    Similarly to "suggested edits" for pages/businesses. You have an accuracy score as an editor based on how many of your suggested edits get accepted/approved. I'd be willing to bet my hat that they have a similar scoring for approving suggested tags - don't click them/ignore them complete for a few days and you'll probably start seeing that suggesting a lot less.
- craigsmansion 9y ago
  If you want to fight it: don't use Facebook! No matter how much an individual thinks they're resisting or messing with the system, they're not. The only way to win is to not participate.
  - TremendousJudge 9y ago
    if other people still upload pictures of you to Facebook it's the same thing
    - gaius 9y ago
      Just as, if you’re ever in an e-mail chain that includes just 1 person on GMail.
  - bitL 9y ago
    I don't use Facebook myself for a few years already, which is paradoxical as they were headhunting me recently to lead one of their ML teams and didn't mind that I quit them long time ago :D But maybe they were just harvesting my own "outlier" training dataset - who knows if I was filmed while onsite and now my gait is analyzed to smallest details including reconstructed skeleton and muscle attachments, my appearance is predicted for the next 30 years under various aging assumptions, maybe later used to fabricate realistically faked video etc. These things are now possible and not that difficult anymore.
    - jacquesm 9y ago
      I don't understand this. Why would you want to interview with a company whose core product you do not support?
      - bitL 9y ago
        Why not? You keep up to date, experiencing world-class hiring process and its requirements (IMO FB's hiring process is the best I've ever experienced, Google's feels one level down), you meet interesting people you might later work together on something cool either there or at another company, you visit interesting places, get some inspirations for your own businesses, interviews might be also a lot of fun (one interview I had at FB was an hour of one-joke-after-another while discussing/inventing architecture of some scalable system; imagine Cryptotinder which is a joke on its own) etc.
- John_KZ 9y ago
  It doesn't work that way. They can see that you tag wrong too many people, and you're obviously an outlier, so your data is rejected. The only way you can "win" is by deleting facebook and pushing for better data protection laws.
  - bitL 9y ago
    You won't do it alone, but if most of your friends start doing that, either you mess up their training dataset by introducing a lot of noise, or you all will get flagged as outliers. Both situations are win comparing to learned helplessness one might fall into when realizing the ways how this data can be misused.
    - John_KZ 9y ago
      Right now the systems in public domain reach 97% accuracy. Facebook doesn't need you, even if you get a million people to try to "trick" the system, you won't do anything. The verification is to fine-tune it. By the way, you forget that Facebook can simply exclude almost all those who know about this. What are you going to do? Do you think nobody will notice the 10M user fan page themed "Fool facebooks AI"? It just knows an extra thing about you, that you don't want to be tracked, so it'll hide news about tracking from your feed.
      - bitL 9y ago
        We already know Uber is cherry-picking users it suspects as working for competitors, so precedent is set for what you are suggesting. It's doable obviously, and if it's doable, companies have already either implemented it or having somebody working on it.
  - gaius 9y ago
    by deleting facebook They still have your photos in their training set
    - John_KZ 9y ago
      You can't undo the past but you can safeguard the future. Besides, if facebook goes bankrupt with heavy legislation demanding the destruction of user data and not allowing their exchange as assets, they will eventually be deleted, along with their training set and everything else. Your biometrics aren't coming back, but at least only the government will have them.

kilotaras 9y ago

> After the faces are located, they’re matched against a set of candidates (which I would guess is an enumeration if all your Nth order friends [who don’t opt out of this], to reduce computational complexity. Face recognition is usually done by first transforming the face into vector space and than looking for the closest vector from known, e.g. openface[1] (which is based on google 2015 paper "FaceNet: A Unified Embedding for Face Recognition and Clustering"). Facebook itself opensourced[2] a library for efficient search of nearest neighbors in such spaces. It doesn't seem likely that generating set of Nth order friends (which grow very quickly with N) would be better than just having a global set or a split of global set by some characteristics (e.g. gender, race). [1] https://cmusatyalab.github.io/openface/ https://cmusatyalab.github.io/openface/ [2] https://github.com/facebookresearch/faiss https://github.com/facebookresearch/faiss

wpietri 9y ago
Wouldn't the betterness of using friend networks being reduced false positives? Facebook has 2 billion MAU, which means they have a lot of faces. Filtering down to the 50k or so that are close to you means it's much easier to find the right person, rather than a similar stranger.
- sangnoir 9y ago
  > Wouldn't the betterness of using friend networks being reduced false positives? They are not limiting to friend networks, otherwise they would not catch (out of network) strangers using your pics. One of the use cases touted is "Help protect you from strangers using your photo"
  - tekknik 9y ago
    Also finding images you’re in but may have been taken by someone not on your friends list (think events). Eventually probably recommending friends from these images.