5 ms·

So this is the future - we can't trust anything and so we trust nothing but our pre-existing biases Bugger that. This is a tech-created problem it can be tech

So this is the future - we can't trust anything and so we trust nothing but our pre-existing biases Bugger that. This is a tech-created problem it can be tech-solved. 1. Photo manipulation leaves traces on the pixels, and this must be worse. 2. cameras can hash / sign every other frame or similar - we can build a chain of hardware to image that can be followed - and of an image or video does not have a CA cert chain it should be as mis-trusted as a web site without TLS 3. Err - there must be other ways.

was_boring 7y ago
I have a friend who was working on #1 specifically. They got out of it after they realized it was just playing whack-a-mole. This is honestly one of the very few use cases of cryptography signing a video or picture that I can see. But that would mean amateur journalists or the common person might be excluded from being a trusted source with video evidence.
- lifeisstillgood 7y ago
  Why? If my iphone camera signs every 24th frame (seems a conservative estimate of speed to do hash and sign), using a combination of the apple camera CA public key and my private key in the secure enclave than we can be reasonably sure that the raw footage now on my site came from a genuine iphone and from the phone I registered (somewhere) All we are doing is trusting that the video images came from a given piece of hardware that we can trust belongs to me. Whether I can be trusted is the same age old problem we have always had But this simply makes it exponentially harder to fake or alter digitally any image (or at least look stupid when someone says "Citation Needed") - if you cannot produce the original raw footage questions will be asked why. I can imagine this being something wanted on police bodycams quite soon
keithwhor 7y ago
Riffing on your thoughts: Killer blockchain app: Couldn't we just use blockchain to protect against video deepfakes? Hash every frame as the video is recording, then create an immutable ledger of frames. All "official" videos must be able to be verified with said ledger, and the ledger itself can only be verified with vendor-specific key licensed by a central authority. Perhaps two -- say, Apple has one for iPhone and they must verify as well. This could certainly be the billion dollar opportunity blockchain fanatics have been looking for.
- chha 7y ago
  How would this prevent anything? A deepfake is essentially video which has been edited. Most newsstories have also been edited, either by processing the sound, video or adding fx and graphics. How would you verify these edits against the ledger, and how would they differ from a deepfake?
  - lifeisstillgood 7y ago
    I'm not sure about the blockchain part but I would suggest that we are just trying to prove that there exists raw footage (is an hours interview from which the one minute segment was edited) and that footage was shot on a unique identifiable camera, whose history can be traced. If the video of politician X making unguarded remarks comes from a camera owned by CNN for the past two years it is a different level of trust than it coming from one bought last week with a credit card in the name of politicians opponents new intern.
Cpoll 7y ago
> 2. cameras can hash / sign every other frame or similar - we can build a chain of hardware to image that can be followed - and of an image or video does not have a CA cert chain it should be as mis-trusted as a web site without TLS Why would this not be easily beatable? Hardware hacking aside, why not the naive approach of filming a monitor playing a deepfake?
- dTal 7y ago
  It is easily beatable. I would hazard the problem is fundamentally unsolvable, while people still have control over their machines.
- dylan-m 7y ago
  The important part there would be chain of trust. You should be able to see, given a video, "oh, that was recorded by some guy who lives in a basement; I can figure out he was never even /near/ Washington" (or "that doesn't have a signature. What is this, 2019?!") and treat it accordingly. It means you need to give up your own identity if you want to publicly verify that a video you've captured is yours, but that's kind of the point. Anonymity is great, but it isn't a source of trustworthy information. Never has been, never will be, and now we have the math to say it. So basically GPG for cameras, but with a hip startup name? Maybe any camera would ship with its own private key out of the box, and there'd need to be some kind of key rotation as well as a way to push your public keys … somewhere(s). I guess an attack here is somebody could lift a private key from someone else's camera and use that key to sign their fakes, but at least that's a reasonably tangible sort of attack that people know how to protect against. (And of course we'll never stop people from watching 24/7 news every minute of the day, so we're all doomed anyway).
  - Cpoll 7y ago
    But at most you can prove which camera shot a picture, and its original owner... Kind of, because there's nothing stopping a photo agency from signing the pictures after they're taken, in order to anonymize their contributors. So, if I want to fake a picture and make it a bit more convincing, I only need to get my hands on a few cameras from pawn shops in Washington. I just can't say "this is a Reuters photo" or "this is a banned commercial" or whatever.
Kalium 7y ago
> 2. cameras can hash / sign every other frame or similar - we can build a chain of hardware to image that can be followed - and of an image or video does not have a CA cert chain it should be as mis-trusted as a web site without TLS Sure! Also, by total coincidence, now it's possible to trace every single photograph ever taken to whoever took it. A bunch of photojournalists in Syria, Iran, Mexico, and Missouri go missing, but I'm sure that's a coincidence. Traceability is not without its drawbacks.
- slig 7y ago
  One can always copy the photo/video without those (see "analog hole").
  - llao 7y ago
    Not if the hashes are also hidden as pixel structure in the image. Something already done for preventing screenshot leaks (World of Warcraft iirc) or to trace movie leaks.
    - slig 7y ago
      Does the information hidden inside a screenshot from WoW is resilient enough to survive a capture from a cell phone camera with JPEG compression and everything in between?
      - Kalium 7y ago
        http://www.eurecom.fr/fr/publication/975/detail/a-survey-of-watermarking-algorithms-for-image-authentication http://www.eurecom.fr/fr/publication/975/detail/a-survey-of-... Not perfectly, but watermarking images in ways that survive re-encoding is far from impossible or unknown.
heymijo 7y ago
"We can't trust anything and so we trust nothing but our pre-existing biases" Oh man, you just articulated something nagging at me about deepfakes that I hadn't been able to.
djakjxnanjak 7y ago
What you’re suggesting is essentially DRM for news, where only an approved subset of sources would be blessed as non-fake. Might as well go back to the days of 3 TV channels.
- djflutt3rshy 7y ago
  People trusted the news a lot more when it was 3 TV channels, the NYT, and your local newspaper. So maybe yes, it is time to do that.
mwfunk 7y ago
I agree that it's eminently tech-solvable. Experts can still analyze the pixels and determine if anything fishy happened. I think the societal problem is more if/when people make up their minds to not trust those experts when they call out a fake and just believe it anyway because they want to believe it. An example would be, some fake video circulates of some politician saying or doing something they never said or did. Maybe someone created it to harm that politician, maybe it was a prank, doesn't matter. A day later, a gazillion experts do the analysis and come out with public evidence that the video was a fake. Great. Then someone like Alex Jones tells his millions of listeners that he knows better, and some world-class expert who he mysteriously won't name told him privately that the video was real, and that somehow all of those experts were in on a conspiracy to debunk it, even though the evidence they provided debunking the video is all out in public (if perhaps too technical for most people). And then, voila. Millions of people now decide that it was real after all, and that all of those scientists and engineers and law enforcement people who said it was fake are all in on some conspiracy to squash the truth. So you're right, it is a tech-created problem, and presumably tech solutions will always exist, but as long as you have an uneducated segment of the population that will always just believe whatever their tribe tells them, those tech solutions won't even matter. I don't know how to solve that problem, other than people becoming more educated and less geographically isolated, so that they can recognize and trust expert analysis when it's available and trustworthy. Put another way, the problem isn't that Alex Jones exists, but that there exist people who think he's credible. The very existence of his listeners is a much bigger problem than his own existence. There will always be conspiracy theorists and there will always be faked evidence to support them, but what holds society together is our collective ability to acknowledge well-documented bullshit as bullshit, regardless of what tribe someone thinks they belong to and regardless of what they want to believe.
amrrs 7y ago
Adobe just a few days back released a paper on detecting fake / cgi pics https://www.technologyreview.com/f/613781/adobes-new-ai-tool-can-spot-when-a-face-has-been-photoshopped/ https://www.technologyreview.com/f/613781/adobes-new-ai-tool...
darepublic 7y ago
In the fifties there were fake photos of UFOs.. as well as ghosts earlier than that. Whether it's an image or a video it can be faked and should not be absolute proof of anything
dogma1138 7y ago
Have you seen how they use projections in practical effects? What stopping me from casting the deepfake into the plain of a camera and getting a valid and signed footage? I don’t think you understand just how problematic it is to verify anything in the analog world. Say you sit at a bar and someone displays a news clip on the TV? How do you know it’s real? Someone video or voice calls you with a deepfake of a loved one are you going to do forensics on it before acting on what they asked you to do?

was_boring 7y ago

I have a friend who was working on #1 specifically. They got out of it after they realized it was just playing whack-a-mole. This is honestly one of the very few use cases of cryptography signing a video or picture that I can see. But that would mean amateur journalists or the common person might be excluded from being a trusted source with video evidence.

lifeisstillgood 7y ago
Why? If my iphone camera signs every 24th frame (seems a conservative estimate of speed to do hash and sign), using a combination of the apple camera CA public key and my private key in the secure enclave than we can be reasonably sure that the raw footage now on my site came from a genuine iphone and from the phone I registered (somewhere) All we are doing is trusting that the video images came from a given piece of hardware that we can trust belongs to me. Whether I can be trusted is the same age old problem we have always had But this simply makes it exponentially harder to fake or alter digitally any image (or at least look stupid when someone says "Citation Needed") - if you cannot produce the original raw footage questions will be asked why. I can imagine this being something wanted on police bodycams quite soon

keithwhor 7y ago

Riffing on your thoughts: Killer blockchain app: Couldn't we just use blockchain to protect against video deepfakes? Hash every frame as the video is recording, then create an immutable ledger of frames. All "official" videos must be able to be verified with said ledger, and the ledger itself can only be verified with vendor-specific key licensed by a central authority. Perhaps two -- say, Apple has one for iPhone and they must verify as well. This could certainly be the billion dollar opportunity blockchain fanatics have been looking for.

chha 7y ago
How would this prevent anything? A deepfake is essentially video which has been edited. Most newsstories have also been edited, either by processing the sound, video or adding fx and graphics. How would you verify these edits against the ledger, and how would they differ from a deepfake?
- lifeisstillgood 7y ago
  I'm not sure about the blockchain part but I would suggest that we are just trying to prove that there exists raw footage (is an hours interview from which the one minute segment was edited) and that footage was shot on a unique identifiable camera, whose history can be traced. If the video of politician X making unguarded remarks comes from a camera owned by CNN for the past two years it is a different level of trust than it coming from one bought last week with a credit card in the name of politicians opponents new intern.

Cpoll 7y ago

> 2. cameras can hash / sign every other frame or similar - we can build a chain of hardware to image that can be followed - and of an image or video does not have a CA cert chain it should be as mis-trusted as a web site without TLS Why would this not be easily beatable? Hardware hacking aside, why not the naive approach of filming a monitor playing a deepfake?

dTal 7y ago
It is easily beatable. I would hazard the problem is fundamentally unsolvable, while people still have control over their machines.
dylan-m 7y ago
The important part there would be chain of trust. You should be able to see, given a video, "oh, that was recorded by some guy who lives in a basement; I can figure out he was never even /near/ Washington" (or "that doesn't have a signature. What is this, 2019?!") and treat it accordingly. It means you need to give up your own identity if you want to publicly verify that a video you've captured is yours, but that's kind of the point. Anonymity is great, but it isn't a source of trustworthy information. Never has been, never will be, and now we have the math to say it. So basically GPG for cameras, but with a hip startup name? Maybe any camera would ship with its own private key out of the box, and there'd need to be some kind of key rotation as well as a way to push your public keys … somewhere(s). I guess an attack here is somebody could lift a private key from someone else's camera and use that key to sign their fakes, but at least that's a reasonably tangible sort of attack that people know how to protect against. (And of course we'll never stop people from watching 24/7 news every minute of the day, so we're all doomed anyway).
- Cpoll 7y ago
  But at most you can prove which camera shot a picture, and its original owner... Kind of, because there's nothing stopping a photo agency from signing the pictures after they're taken, in order to anonymize their contributors. So, if I want to fake a picture and make it a bit more convincing, I only need to get my hands on a few cameras from pawn shops in Washington. I just can't say "this is a Reuters photo" or "this is a banned commercial" or whatever.

Kalium 7y ago

> 2. cameras can hash / sign every other frame or similar - we can build a chain of hardware to image that can be followed - and of an image or video does not have a CA cert chain it should be as mis-trusted as a web site without TLS Sure! Also, by total coincidence, now it's possible to trace every single photograph ever taken to whoever took it. A bunch of photojournalists in Syria, Iran, Mexico, and Missouri go missing, but I'm sure that's a coincidence. Traceability is not without its drawbacks.

slig 7y ago
One can always copy the photo/video without those (see "analog hole").
- llao 7y ago
  Not if the hashes are also hidden as pixel structure in the image. Something already done for preventing screenshot leaks (World of Warcraft iirc) or to trace movie leaks.
  - slig 7y ago
    Does the information hidden inside a screenshot from WoW is resilient enough to survive a capture from a cell phone camera with JPEG compression and everything in between?
    - Kalium 7y ago
      http://www.eurecom.fr/fr/publication/975/detail/a-survey-of-watermarking-algorithms-for-image-authentication http://www.eurecom.fr/fr/publication/975/detail/a-survey-of-... Not perfectly, but watermarking images in ways that survive re-encoding is far from impossible or unknown.

heymijo 7y ago

"We can't trust anything and so we trust nothing but our pre-existing biases" Oh man, you just articulated something nagging at me about deepfakes that I hadn't been able to.

djakjxnanjak 7y ago

What you’re suggesting is essentially DRM for news, where only an approved subset of sources would be blessed as non-fake. Might as well go back to the days of 3 TV channels.

djflutt3rshy 7y ago
People trusted the news a lot more when it was 3 TV channels, the NYT, and your local newspaper. So maybe yes, it is time to do that.

mwfunk 7y ago

I agree that it's eminently tech-solvable. Experts can still analyze the pixels and determine if anything fishy happened. I think the societal problem is more if/when people make up their minds to not trust those experts when they call out a fake and just believe it anyway because they want to believe it. An example would be, some fake video circulates of some politician saying or doing something they never said or did. Maybe someone created it to harm that politician, maybe it was a prank, doesn't matter. A day later, a gazillion experts do the analysis and come out with public evidence that the video was a fake. Great. Then someone like Alex Jones tells his millions of listeners that he knows better, and some world-class expert who he mysteriously won't name told him privately that the video was real, and that somehow all of those experts were in on a conspiracy to debunk it, even though the evidence they provided debunking the video is all out in public (if perhaps too technical for most people). And then, voila. Millions of people now decide that it was real after all, and that all of those scientists and engineers and law enforcement people who said it was fake are all in on some conspiracy to squash the truth. So you're right, it is a tech-created problem, and presumably tech solutions will always exist, but as long as you have an uneducated segment of the population that will always just believe whatever their tribe tells them, those tech solutions won't even matter. I don't know how to solve that problem, other than people becoming more educated and less geographically isolated, so that they can recognize and trust expert analysis when it's available and trustworthy. Put another way, the problem isn't that Alex Jones exists, but that there exist people who think he's credible. The very existence of his listeners is a much bigger problem than his own existence. There will always be conspiracy theorists and there will always be faked evidence to support them, but what holds society together is our collective ability to acknowledge well-documented bullshit as bullshit, regardless of what tribe someone thinks they belong to and regardless of what they want to believe.

amrrs 7y ago

Adobe just a few days back released a paper on detecting fake / cgi pics https://www.technologyreview.com/f/613781/adobes-new-ai-tool-can-spot-when-a-face-has-been-photoshopped/ https://www.technologyreview.com/f/613781/adobes-new-ai-tool...

darepublic 7y ago

In the fifties there were fake photos of UFOs.. as well as ghosts earlier than that. Whether it's an image or a video it can be faked and should not be absolute proof of anything

dogma1138 7y ago

Have you seen how they use projections in practical effects? What stopping me from casting the deepfake into the plain of a camera and getting a valid and signed footage? I don’t think you understand just how problematic it is to verify anything in the analog world. Say you sit at a bar and someone displays a news clip on the TV? How do you know it’s real? Someone video or voice calls you with a deepfake of a loved one are you going to do forensics on it before acting on what they asked you to do?