“What’s that?” she asked.
“The ugliest T-shirt in the world,” he said, and kissed her cheek.
“The Bollards will be disappointed,” she said, coming in and closing the door. “I thought they’d had me sleeping in that.”
“So ugly that digital cameras forget they’ve seen it.”
“Cameras can see it. The surveillance cameras can all see it, but then they forget they’ve seen it.”
“Why?”
“Because their architecture tells them to forget it, and anyone who’s wearing it as well. They forget the figure wearing the ugly T-shirt. Forget the head atop it, the legs below, feet, arms, hands. It compels erasure. That which the camera sees, bearing the sigil, it deletes from the recalled image. Though only if you ask it to show you the image. So there’s no suspicious busy-ness to be noticed. If you ask for June 7, camera 53, it retrieves what it saw. In the act of retrieval, the sigil, and the human form bearing it, cease to be represented. By virtue of deep architecture. Gentlemen’s agreement."
That sounds more realistic than one might expect. Most printers will apparently reject printing of patterns containing the EURion pattern [1], commonly found on paper cash. Not a far stretch that some facial recognition tech might contain deliberate hidden patterns in the future.
I like the license plate dress featured in the article. A simpler point to bear in mind that even after the immediate pandemic emergency is over many people will need to continue to wear a mask in public. Doing errands/outdoor exercise/volunteer work in public while masked is an enjoyable increase of freedom and privacy.
The juggalos are evolving. They have superior technology. We thought we'd finally cracked the code, but it was too late: the enemy was already within the gates. Our hopes were doused in a shower of Faygo, and the machine lay sizzling its last breath in the putrid foam.
As with ad blocking, no one really cares until it becomes widespread. CV doesn't account for dazzle because no one's really using it; there are more immediate problems to solve of higher utility.
I'm certainly no expert in CV, but having worked in government/law enforcement tech previously (pre-Snowden days) I have full confidence that if people started actually using CV Dazzle to evade detection, they would work on breaking through it. Generally speaking those systems only receive features that are actually problems the users are facing. Once they identify the problem they can throw nearly unlimited amounts of money at it, and they will win.
This is why I'm focusing my efforts on political solutions to protect our privacy, rather than technical solutions. That's not to say I don't think technical solutions are important, because they absolutely are, but even things like encryption could be made illegal politically if they want to. Thus politics is the most important arena for privacy advocates IMHO.
And the issue now is that as much as you might blend in to the computer, if you're going out and about with CV dazzle on you're going to attract a lot of scrutiny from actual people.
The argument that we need political solutions originally came from a time when technology people weren't paying attention to politics at all, and of course then you get this:
But it feels like it has gone the other way now, to the point that people dismiss potentially helpful technical solutions because "what really matters is politics" even though in reality you need both.
If the government bans privacy technologies then it's hard to use them, but if nobody develops or uses them to begin with then you don't have them either.
And for some people the government in question isn't ours, it's a foreign authoritarian government where "policy solutions" are just not going to happen and the only option available is technology. It's a lot easier to develop that technology in less authoritarian countries, which means we have a responsibility to do it, because if they have to do it the cost of a mistake is a lot higher.
Agree with you completely. If I take my previous statement further to "don't even bother holing up as insurgents because the government will win the battle" then it goes to a place I don't agree with (I think there's plenty of evidence that the "little guy" can win against big governments. Afghanistan is a great example).
So I really should re-phrase my position to clarify that I meant in the United States, but also even in the US I do think the technological solutions are important and should be developed. The important point I want to stress is that I don't think we should ever disregard the political war because of technological confidence in our ability to thwart the authoritarians. Both politics and technology are both critical pieces. If we lose on either front, we lose.
when they're generated digitally, it's surprisingly hard to get good robustness against adversarial examples, and people have been working on it for a while. of course you can train to stop any specific input, but then new ones can be created for the newly-trained network.
there are techniques that can prevent this, but you pay a real price in overall performance, which may or may not be worth it.
of course it's harder to make physical objects work, but if somebody did manage to make a method for physical adversarial attacks that worked reliably, I would guess you'd run into the same problems. it's not at all clear that the endgame is that we can easily stop these attacks.
long term storage overcomes the deficiencies in todays software
As soon as they update the software, they run it over the stored video and investigate the differences. When they identify particular evasion techniques, they can flag up all instances and see what was done under their cover.
>Clothes change much slower than software updates,
I change my clothes at least once a day and add to my wardrobe on a ~weekly to ~monthly basis. Having been party to large system upgrades at even a reasonably competent company, I'd be shocked to find out that any surveillance system anywhere is matching that velocity.
> The issue with adversarial examples is defending against a wide class of examples, not a few specific images.
So who says it has to be a few specific images? They could auto-generate a different selection to choose from for each purchaser and then print to order. Every individual article of clothing could be unique.
Getting strong William Gibson/Zero History vibes from that article.
If you haven’t read the book (written over a decade ago), a major plot point revolves around a T-shirt that renders the wearer invisible to CCTV cameras.
You might be able to take your car out for a walk if the adaptive cruise will latch onto and follow your shirt plate! It could be like having a leash-trained 3500 lbs steel dog - just hope it doesn't see any squirrels. Heel, Tesla!
Honestly, I'd wear that dress easily, even for regular outings, not just for hiding from cameras. Those hoodies, though, maybe I'm an old fart but they're a bit too "abstract". Sadly, those seem like the most effective option right now. I read the Adversarial Patch article [0] and it pretty much demonstrated similar patterns.
Trouble is, if these get widespread enough - they won't look out of place but the moment they become widespread is the moment updates get pushed to make them obsolete. It's an unescapable situation.
> Trouble is, if these get widespread enough - they won't look out of place but the moment they become widespread is the moment updates get pushed to make them obsolete. It's an unescapable situation.
Did you just say recurring revenue (for the clothing makers)?
Related thought, on hiding from cameras and overzealous public identification systems...
Most western nations are now, temporarily, accepting face coverings as daily attire in the age of COVID-19. Plenty of asian and muslim nations did previously, for cultural and health reasons.
Culturally acceptable anonymity, courtesy of the pandemic. Maybe this isn't temporary if the fear of future viral spreads takes root in culture.
Sadly, it's definitely much easier to push out a software update than it is to replace your clothing. But that game of cat and mouse won't start until the clothing becomes more commonplace, which is very doubtful.
The mostly black backpack makes me wonder if a vantablack-covered tarp poncho would work. Since you should just be a black, featureless blob to a camera there's not much for the NN to latch onto other than outline.
The backpack works because it's not just black and it actually includes some adversarial input ("fake" license plates). A Vantablack poncho wouldn't protect you more than a regular poncho with no distinguishing patterns. The image recognition wouldn't really latch on to creases in your clothes to perform recognition. Your face would still be visible and there's nothing about solid black that further confuses the AI behind the system, there are far better adversarial examples for that.
It's not to disguise the wearer, it's to feed junk data into any license plate reader that happens to look at you. I don't think it's really useful, but the idea is that if the systems get filled up with enough junk data, it won't be worth it to maintain the systems if they have to sort through a bunch of nonsense every time they want to search for something.
How do you distinguish between ones that don't match because they're invalid or ones that don't match because the algorithm didn't properly process the image?
It seems like you could cause some real confusion if you lived near one of these cameras by putting a screen in your window that cycled through random plates. Especially if you run your own ALPR on the street and only use plates that have passed by before.
I wonder what the rules are about displaying license plates on your car. No doubt you're required to display the one issued to the vehicle, but what happens if you also cover the car in pictures of different ones?
I think they're hysterically funny, though I'm having trouble imagining a use case for the leggings.
Clubbing? Throw on a long coat to cover it until you get to the club, then wander around making provocative ironic social commentary without saying a word.
That's all that comes to mind (and isn't an option currently, due to the pandemic).
> For me at least, that first link has some NSFW content.
> Specifically, it shows a pair of leggings on a woman designed to look like a naked man titled "David's Marble Legs":
But how is that NSFW? This is one of the most famous sculptures in western art. Everybody has seen it. I had a photo of the David sculpture in the cover of a history book when I was 11, given by our school. I find it very disturbing, even creepy, that somebody may honestly think that this image is NSFW.
imagine you happen to open this page at work while someone walks past your desk. if they say "hey, are you looking at penises at work?" you have the opportunity to explain "no, actually this is a representation of one of the greatest works of western art". more often than not, they will just keep walking and not ask for an explanation. do you really want to be the person they remember this way?
some people just don't want to open a picture of a penis at work (or be surprised by one at home), no matter how artistic. I don't find this disturbing.
What about an incredibly bright set of IR lights? Surely that would cause most cameras (at least IR ones) to massively overexpose. It might only work at night though.
To me, this is the best way to go. I've seen the hats with the UV LEDS embedded in the brim, but why not around the collar of a shirt?. The only side effect I see is UV light directly in the face. Not sure if the UV wavelengths/intensity can be low enough to blind a camera to not cause odd tan lines or worse?
Either way, blinding the camera with lights that are invisible to other people unlike these outfits would be much more effective. You can still be subversive without advertising it to anyone. If you're recruiting, then sure, where these outfits. If you actually want to not be seen by the cameras, make the cameras unable to see.
amusingly I've seen a couple of people out recently wearing anti-cv masks. I guess you wear what you've already purchased, but it's most interesting that I've seen more than one.
I seriously hope this begins to become a fashion trend in the near future. I don't see why one of the more out-there fashion designers wouldn't pick it up.
Na, just make it illegal. Governments are increasingly eroding people's freedoms. Let's remove another one, and force everyone to wear approved and regulated clothing. You know, for your own safety.
Obviously this is going to turn into one of those back-and-forth wars of more effective adversarial clothing designs versus improved detection algorithms. What I really wonder is, which one comes out on top in the end, or at least most of the time? Will designs and/or fashions get weird and effective enough that it just isn't practical to recognize people in general in public with computerized algorithms? Or will the algorithms win, with these being 95% ineffective and also resulting in drastically more attention from ordinary people and any human authorities who happen to be watching?
If I could pixelate my own skin on a whim I wouldn’t worry about the shirt. Unconfortably I think eventually we’ll become fully traceable and privacy will be a thing of the past.
Are masks of the type worn to protect against covid-19 effective against being identified on camera? Because this seems like an amazing excuse (in terms of a silver lining on a very dark cloud) for facemasks as 'anonymity wear' to become mainstream.
Would wearing some kind of full length cloak or burqa-sequence garment that obscures a lot of movement be a plausible counter? Obviously they can’t obscure everything about your gait, but I imagine they’d take away enough useful information to make it significantly harder.
Wouldn’t wearing something like this just make you much more of a target? Maybe I don’t understand these things well, but it seems if I were trying to track people I’d make the very small minority wearing something like this a target group.
Another way of thinking about it, if I’m the only person in my community wearing these weird patterns, aren’t I even more identifiable than I would be otherwise?
I was hoping for something that thwarted the processing in the CCD (chip that captures light in digital cameras AFAIK) instead of something that just thwarts the recognition algorithms.
Of course, if the camera still captures a perfect image, that image or video is always around to be processed at a later date.
CCD cameras haven't been the preferred camera sensor for more than a decade at this point. The entire industry moved to CMOS a long time ago. You might find CCDs in use today in specific scenarios like where you need global shutters, but even then, you can buy CMOS sensors with global shutters that outperform those CCDs.
This is futile. The cat is out of the bag. There is no technological counter to AI + ubiquitous sensor networks.
There are AI systems out there that can identify you by gait. Possibly already in your phone, ostensibly to help detect if the phone is carried by someone else. Straightforward to connect to the 24/7 video surveillance network that is built all around us: ring, self-driving cars/drones that double as corporate spies, video checkout, etc.
Furthermore, radars are coming to your phone in the near future: https://atap.google.com/soli. There is no way to hide from such radars picking up everything that moves: your gait, your breath, your heartbeat. Oh, and the gait, breath and heartbeat of anyone near you, whether they agreed to the 500 pages EULA surrendering their privacy or not.
I can’t disagree with anything in your post, but I would add that there always opportunities to be a pain in the ass.
Maybe not all systems can be toppled, but most can certainly be hacked, abused, slowed down, obscured, damaged, re-purposed, stolen, rebooted, copied, exposed, hidden, rendered obsolete, improved, deleted, etc.
I see your defeatism and raise you a giant middle finger!
I agree. Messing with the system can be fun even if ultimately futile. Also the fact that AI can sometimes be stupid can sometimes lead to good trading opportunities.
98 comments
[ 4.2 ms ] story [ 228 ms ] threadWilliam Gibson, Zero History (2010)
[1]: https://en.wikipedia.org/wiki/EURion_constellation
There's also the non-technical approaches, like https://en.wikipedia.org/wiki/Anti-mask_law.
This is why I'm focusing my efforts on political solutions to protect our privacy, rather than technical solutions. That's not to say I don't think technical solutions are important, because they absolutely are, but even things like encryption could be made illegal politically if they want to. Thus politics is the most important arena for privacy advocates IMHO.
Like this: https://xkcd.com/1105/
https://xkcd.com/538/
But it feels like it has gone the other way now, to the point that people dismiss potentially helpful technical solutions because "what really matters is politics" even though in reality you need both.
If the government bans privacy technologies then it's hard to use them, but if nobody develops or uses them to begin with then you don't have them either.
And for some people the government in question isn't ours, it's a foreign authoritarian government where "policy solutions" are just not going to happen and the only option available is technology. It's a lot easier to develop that technology in less authoritarian countries, which means we have a responsibility to do it, because if they have to do it the cost of a mistake is a lot higher.
So I really should re-phrase my position to clarify that I meant in the United States, but also even in the US I do think the technological solutions are important and should be developed. The important point I want to stress is that I don't think we should ever disregard the political war because of technological confidence in our ability to thwart the authoritarians. Both politics and technology are both critical pieces. If we lose on either front, we lose.
there are techniques that can prevent this, but you pay a real price in overall performance, which may or may not be worth it.
of course it's harder to make physical objects work, but if somebody did manage to make a method for physical adversarial attacks that worked reliably, I would guess you'd run into the same problems. it's not at all clear that the endgame is that we can easily stop these attacks.
Clothes change much slower than software updates, and defending against a few adversarial examples is relatively easy, so this shouldn't be a problem.
Some of these approaches are 7+ years old and still work.
As soon as they update the software, they run it over the stored video and investigate the differences. When they identify particular evasion techniques, they can flag up all instances and see what was done under their cover.
I change my clothes at least once a day and add to my wardrobe on a ~weekly to ~monthly basis. Having been party to large system upgrades at even a reasonably competent company, I'd be shocked to find out that any surveillance system anywhere is matching that velocity.
So who says it has to be a few specific images? They could auto-generate a different selection to choose from for each purchaser and then print to order. Every individual article of clothing could be unique.
If you haven’t read the book (written over a decade ago), a major plot point revolves around a T-shirt that renders the wearer invisible to CCTV cameras.
https://adversarialfashion.com/
if you read the article you'd see that they only work sometimes though.
Trouble is, if these get widespread enough - they won't look out of place but the moment they become widespread is the moment updates get pushed to make them obsolete. It's an unescapable situation.
[0]: https://arxiv.org/abs/1712.09665
Did you just say recurring revenue (for the clothing makers)?
Most western nations are now, temporarily, accepting face coverings as daily attire in the age of COVID-19. Plenty of asian and muslim nations did previously, for cultural and health reasons.
Culturally acceptable anonymity, courtesy of the pandemic. Maybe this isn't temporary if the fear of future viral spreads takes root in culture.
[1] https://thenextweb.com/neural/2020/03/10/masks-wont-make-you...
https://www.rageon.com/products/ensemble-7
And Adversarial Fashion's anti-ALPR backpacks are cool looking as well.
https://adversarialfashion.com/collections/bags/products/sta...
Sadly, it's definitely much easier to push out a software update than it is to replace your clothing. But that game of cat and mouse won't start until the clothing becomes more commonplace, which is very doubtful.
https://adversarialfashion.com/pages/diy-resources
Specifically, it shows a pair of leggings on a woman designed to look like a naked man titled "David's Marble Legs":
https://www.rageon.com/products/davids-marble-legs?refSrc=43...
Clubbing? Throw on a long coat to cover it until you get to the club, then wander around making provocative ironic social commentary without saying a word.
That's all that comes to mind (and isn't an option currently, due to the pandemic).
> Specifically, it shows a pair of leggings on a woman designed to look like a naked man titled "David's Marble Legs":
But how is that NSFW? This is one of the most famous sculptures in western art. Everybody has seen it. I had a photo of the David sculpture in the cover of a history book when I was 11, given by our school. I find it very disturbing, even creepy, that somebody may honestly think that this image is NSFW.
some people just don't want to open a picture of a penis at work (or be surprised by one at home), no matter how artistic. I don't find this disturbing.
Is anyone actually at work anymore?
Depends on the release process.
Either way, blinding the camera with lights that are invisible to other people unlike these outfits would be much more effective. You can still be subversive without advertising it to anyone. If you're recruiting, then sure, where these outfits. If you actually want to not be seen by the cameras, make the cameras unable to see.
Not to mention the safety risks of strapping IR LEDs that are that bright at other people's eye level.
Kramer buys one - or sells them? Things go sideways.
Are masks of the type worn to protect against covid-19 effective against being identified on camera? Because this seems like an amazing excuse (in terms of a silver lining on a very dark cloud) for facemasks as 'anonymity wear' to become mainstream.
Humans will forget about you, data recorded about you will be stored an reused and abused for who knows how long.
Another way of thinking about it, if I’m the only person in my community wearing these weird patterns, aren’t I even more identifiable than I would be otherwise?
I imagine for those people the ability to move around undetected would be very much worth the garish pattern.
I also question just whether the average policeman would be able to tell whether clothing was designed to be untrackable
https://en.wikipedia.org/wiki/Streisand_effect
https://www.reddit.com/r/technology/comments/fyrqi2/til_that...
Of course, if the camera still captures a perfect image, that image or video is always around to be processed at a later date.
There are AI systems out there that can identify you by gait. Possibly already in your phone, ostensibly to help detect if the phone is carried by someone else. Straightforward to connect to the 24/7 video surveillance network that is built all around us: ring, self-driving cars/drones that double as corporate spies, video checkout, etc.
Furthermore, radars are coming to your phone in the near future: https://atap.google.com/soli. There is no way to hide from such radars picking up everything that moves: your gait, your breath, your heartbeat. Oh, and the gait, breath and heartbeat of anyone near you, whether they agreed to the 500 pages EULA surrendering their privacy or not.
I can’t disagree with anything in your post, but I would add that there always opportunities to be a pain in the ass.
Maybe not all systems can be toppled, but most can certainly be hacked, abused, slowed down, obscured, damaged, re-purposed, stolen, rebooted, copied, exposed, hidden, rendered obsolete, improved, deleted, etc.
I see your defeatism and raise you a giant middle finger!
The claim Hanwang's software works is dubious at best. And they also admit with sunglasses it totally fails.
https://www.reuters.com/article/us-health-coronavirus-facial...
Privacy online is all that remains. In meatspace, I strive only to be unremarkable. In my case, just a tired old man. Nothing to see here.
My cover story, I suppose, is using a VPN service to hide my porn addiction.
Edit: I recommend Vinge's True Names.