Worst idea ever. Worst privacy violation ever. This would would be a boon to stalkers, perverts, creeps, serial killers, and I imagine 99% of people would never opt into it.
You would opt into it implicitly just by having your pictures on Facebook. Not necessarily just by having an account, mind you; your friends might upload and tag pictures of you.
Yes, it would be a huge privacy violation. Facebook just likely wouldn't care.
Instead of revealing the individual's identity, FB could send them an automated msg pointing back to your profile with the option to connect (or at least chat). This would leave the decision (privacy protection) in the hands of the recipient.
Does anyone know if fb face recognition can actually work for all 800 million users or does its tagging do something like nearest neighbor to your friends which I imagine is a simpler problem.
I'd probably argue that this is not only an unsolved research problem, but an unsolvable one.
I mean, do we have any evidence whatsoever that given a corpus of photos of almost a billion people, there's any way to pick out photos of a given person without a very high probability of (many) false positives?
I've seen probably five or six people over the years that in certain photos I wouldn't be able to distinguish from myself, especially in real world situations with crappy lighting, poor focus, etc. It's rare, sure, but with a billion people, "rare" means you'll "only" see, what, a thousand cases? I think there's just too much ambiguity in photos (and looks in general), not that we necessarily lack the right algorithms or anything like that.
However, I do believe that the meta-data that Facebook has on all of us is probably enough to get damn close in most cases, even if people don't have friend connections. Especially if they've got some location data...
Besides, for the purposes of what this article is talking about, I'd guess that if someone looks similar enough to the person you saw on the subway to falsely trigger a face match, you'd still probably be interested in following up on that "missed connection".
Completely agreed. Even out of a million people, this is probably impossible. I'm quite sure humans couldn't do it.
To test humans:
Take city of million people. Take photo of one person.
Proceed to ask every person in said city "who is this?" Most likely answer is "no idea", so only consider labeled answers.
You will obviously get multiple answers for the same photo due to look-alikes. I suspect the look-alike problem is so great that even the best training optimization could not achieve >= 50% system labeling accuracy for a picture of a single person with no other context.
You may be able to achieve 50+% labeling accuracy when you present many photos of the unknown individual to those million people, especially if that individual is pictured with other individuals.
It's not nearly good enough for this. While I don't know what they're doing, it doesn't matter -- the state of the art in face recognition is far lower than what this requires.
I would also assume that Facebook's recognition performance is heavily dependent on the fact that everyone has a limited friend circle, because then the requirement is only that you can be distinguished among the few hundred or thousand people you've ever been photographed with, not the 100s of millions of users you haven't.
While not directly related (nor the method they are probably using), here's a project that describes how using others in your photographs can help recognition: http://lear.inrialpes.fr/pubs/2008/MV08/
So, what if once you upload the "missed connection" photo, Facebook tags it with GPS and filters possible matches against your current location? That would narrow it down quite a bit. I assume they have enough GPS data from other users' photos and status updates to at least place them generally within a state or city at a given time.
Without social network context (you are much more likely to meet a friend of a friend than someone living in a country half way around the world), that is impossible with current-day research.
I'd argue that given facebook's dataset of low-resolution photos in crappy lighting conditions, it could actually be 100% impossible. That is, amongst 800 million people, there are bound to be people who look exactly the same as me with available data.
Yes, that would seem to be the obvious reaction. But that's just because you're a "normal", someone that has, on occasion, interacted romantically in a fairly standard and direct manner with members of the opposite sex.
Now, imagine that you're a forever-alone that spends most of his free time (well, that minuscule portion of free time that's not consumed by LoL, SWTOR, Skyrim, Minecraft, etc. [1]) on Reddit complaining about being "friend-zoned" and fantasizing about that magical girl that looks like a fashion model but "gets you", who likes you just for being a "nice guy" that's "smart", breaking from her natural tendency to go for "bad-boys" because your awesomeness was so obvious to her despite your awkward creepiness and social ineptitude. It's a very different world, one where the idea of semi-passively stalking randoms on the subway without any fear of direct rejection seems a lot more appealing...
[1] I say this as someone that wastes an awful lot of free time on Reddit, as well as playing LoL, SWTOR, Skyrim, Minecraft, and many more, including some Facebook games that I'm far too embarrassed about to cop to on HN (even if I do work in the field and have the ostensible excuse of "market research")...
yeah, server just melted. Up now. Here's text in case it goes down again:
- - -
You know those Craigslist missed connections posts? The ones where you’re looking across the subway platform at someone, and you wish you could get their number, but you can’t really shout across the tracks?
Well, I was in New York City this weekend and I had a sudden inspiration. There’s a (hypothetical) solution to the “missed connections” problem. And it’s sitting around in plain sight.
Here’s how it works. Many of you are probably aware that Facebook has been steadily developing its facial recognition algorithm. It’s quite good at this point. It can scan any of your photos and tag you automatically, if you’ve enabled the feature. That means Facebook’s algorithm can recognize your unique face. It stands to reason, therefore, that Facebook’s algorithm could scan any photo from any of its 800 million members, and recognize you if you’re in it.
So, let’s take that one logical step further. Imagine you take a photo of someone and upload it to Facebook’s servers. Facebook can analyze the photo and tag that person’s face.
I bet you see where I’m going with this.
You’re back on the subway platform — you see someone across the way. You raise your phone and snap a quick picture via the Facebook app. The train comes. While you’re pulling away, the app has displayed that person’s profile. You click “friend.”
And just like that, missed connections are a thing of the past.
Interesting, eh? Also, creepy. Obviously, if Facebook enabled this feature it would cause a firestorm of privacy concerns. But Facebook hasn’t exactly shown much concern for privacy in the past. And everyone does have a public profile, anyway… and subways are public areas. Given that the tech exists right now, I bet we’ll see this within a couple years.
"It stands to reason, therefore, that Facebook’s algorithm could scan any photo from any of its 800 million members, and recognize you if you’re in it. So, let’s take that one logical step further. Imagine you take a photo of someone and upload it to Facebook’s servers. Facebook can analyze the photo and tag that person’s face."
This could work for a mutual friend at a party. But if the face recognizer is given a low-quality subway photo and has 7 million people to choose from (NYC residents) of equal probability, the chance of current technology getting an accurate match is nil.
The author may be a bit confused how Facebook's technology works. Yes, it recognizes faces, but it also eliminates ~99.9999% of members as possibilties by leveraging the social graph.
It might still work if it could narrow down the possibilities to, say, a hundred photos or less. The user should be able to go the "last mile" and pick out the one they're looking for.
I'd be curious to see if Facebook's facial recognition technology is this good.
Also, lots of people out in public (especially in NYC) aren't necessarily residents of the given area, which potentially makes this much harder.
19 comments
[ 1266 ms ] story [ 117 ms ] threadYes, it would be a huge privacy violation. Facebook just likely wouldn't care.
Technology stops for no man and no laws. It may be time to start rethinking what we mean by privacy.
I mean, do we have any evidence whatsoever that given a corpus of photos of almost a billion people, there's any way to pick out photos of a given person without a very high probability of (many) false positives?
I've seen probably five or six people over the years that in certain photos I wouldn't be able to distinguish from myself, especially in real world situations with crappy lighting, poor focus, etc. It's rare, sure, but with a billion people, "rare" means you'll "only" see, what, a thousand cases? I think there's just too much ambiguity in photos (and looks in general), not that we necessarily lack the right algorithms or anything like that.
However, I do believe that the meta-data that Facebook has on all of us is probably enough to get damn close in most cases, even if people don't have friend connections. Especially if they've got some location data...
Besides, for the purposes of what this article is talking about, I'd guess that if someone looks similar enough to the person you saw on the subway to falsely trigger a face match, you'd still probably be interested in following up on that "missed connection".
To test humans: Take city of million people. Take photo of one person. Proceed to ask every person in said city "who is this?" Most likely answer is "no idea", so only consider labeled answers.
You will obviously get multiple answers for the same photo due to look-alikes. I suspect the look-alike problem is so great that even the best training optimization could not achieve >= 50% system labeling accuracy for a picture of a single person with no other context.
You may be able to achieve 50+% labeling accuracy when you present many photos of the unknown individual to those million people, especially if that individual is pictured with other individuals.
I would also assume that Facebook's recognition performance is heavily dependent on the fact that everyone has a limited friend circle, because then the requirement is only that you can be distinguished among the few hundred or thousand people you've ever been photographed with, not the 100s of millions of users you haven't.
While not directly related (nor the method they are probably using), here's a project that describes how using others in your photographs can help recognition: http://lear.inrialpes.fr/pubs/2008/MV08/
I'd argue that given facebook's dataset of low-resolution photos in crappy lighting conditions, it could actually be 100% impossible. That is, amongst 800 million people, there are bound to be people who look exactly the same as me with available data.
Now, imagine that you're a forever-alone that spends most of his free time (well, that minuscule portion of free time that's not consumed by LoL, SWTOR, Skyrim, Minecraft, etc. [1]) on Reddit complaining about being "friend-zoned" and fantasizing about that magical girl that looks like a fashion model but "gets you", who likes you just for being a "nice guy" that's "smart", breaking from her natural tendency to go for "bad-boys" because your awesomeness was so obvious to her despite your awkward creepiness and social ineptitude. It's a very different world, one where the idea of semi-passively stalking randoms on the subway without any fear of direct rejection seems a lot more appealing...
[1] I say this as someone that wastes an awful lot of free time on Reddit, as well as playing LoL, SWTOR, Skyrim, Minecraft, and many more, including some Facebook games that I'm far too embarrassed about to cop to on HN (even if I do work in the field and have the ostensible excuse of "market research")...
- - -
You know those Craigslist missed connections posts? The ones where you’re looking across the subway platform at someone, and you wish you could get their number, but you can’t really shout across the tracks?
Well, I was in New York City this weekend and I had a sudden inspiration. There’s a (hypothetical) solution to the “missed connections” problem. And it’s sitting around in plain sight.
Here’s how it works. Many of you are probably aware that Facebook has been steadily developing its facial recognition algorithm. It’s quite good at this point. It can scan any of your photos and tag you automatically, if you’ve enabled the feature. That means Facebook’s algorithm can recognize your unique face. It stands to reason, therefore, that Facebook’s algorithm could scan any photo from any of its 800 million members, and recognize you if you’re in it.
So, let’s take that one logical step further. Imagine you take a photo of someone and upload it to Facebook’s servers. Facebook can analyze the photo and tag that person’s face.
I bet you see where I’m going with this.
You’re back on the subway platform — you see someone across the way. You raise your phone and snap a quick picture via the Facebook app. The train comes. While you’re pulling away, the app has displayed that person’s profile. You click “friend.”
And just like that, missed connections are a thing of the past.
Interesting, eh? Also, creepy. Obviously, if Facebook enabled this feature it would cause a firestorm of privacy concerns. But Facebook hasn’t exactly shown much concern for privacy in the past. And everyone does have a public profile, anyway… and subways are public areas. Given that the tech exists right now, I bet we’ll see this within a couple years.
The implications are rather large.
Dan
This could work for a mutual friend at a party. But if the face recognizer is given a low-quality subway photo and has 7 million people to choose from (NYC residents) of equal probability, the chance of current technology getting an accurate match is nil.
The author may be a bit confused how Facebook's technology works. Yes, it recognizes faces, but it also eliminates ~99.9999% of members as possibilties by leveraging the social graph.
I'd be curious to see if Facebook's facial recognition technology is this good.
Also, lots of people out in public (especially in NYC) aren't necessarily residents of the given area, which potentially makes this much harder.