Comment by keiferski
1 year ago
Can someone tell me why this idea isn’t workable and wouldn’t solve most deepfake issues?
All camera and phone manufacturers embed a code in each photo / video they produce.
All social media channels prioritize content that has these codes, and either block or de-prioritize content without them.
Result: the internet is filled with a vast amount of AI generated nonsense, but it’s mostly not treated as anything but entertainment. Any real content can be traced back to physical cameras.
The main issue I see is if the validation code is hacked at the camera level. But that is at least as preventable as say, preventing printers from counterfeiting money.
So all I would have to do to make a "legitimate" fake picture is to generate it, print it, take a signed picture of the print with a camera and then upload it on the web?
With the right setup I could probably just take a picture of the screen directly, making it even easier (and enabling it for videos too).
Presumably the camera would have to incorporate GPS or other systems to ensure that you aren’t just taking a photo of a screen.
But yes that does add a wrinkle.
For one it puts a lot of trust into those corporations which they have not earned. Corporations exist to maximize profit. They already sell advertisement to foreign hostile actors. Why wouldn't they do the same with these types of codes for enough payment?
Also it gives them a lot of power to frame anyone for anything. How do you defend yourself against a cryptographic framed "proof" that ties you to a crime? At least when no evidence is trustworthy, the courts have to take this possibility into account. That's not the case when it's only used in rare cases.
But this trust is already given to corporations that make printers to prevent counterfeiters. By your logic they would already have sold out to hostile actors.
To your second point: it’s not that my method guarantees absolutely flawless photos. It just makes it more likely to be secure.
I think placing trust in technology is misguided. Bad actors will always figure out a way to abuse the machines.
What we can do is place trust in particular institutions, and use technology to verify authenticity. Not verify that what the institution is saying is true, but verify that they really are standing by this claim.
This is challenging because no institution is going to be 100% trustworthy all of the time in perpetuity. But you can make reasonable assessments about which institutions appear more credible. Then it's a matter of policing your own biases.
Sure, but I don't see how that could be implemented in the world in a way other than how I outlined, considering that photos/videos can be created by anyone. There is no central media institution that can control all sharing of content.
It would seem to me that the institution to place your trust in would be the one that implements and verifies the coding system I discussed.
When a news organization hires a journalist to file a report, they put their reputation on the line in the hands of the journalist.
As a consumer of news, you put your trust in the institution to have a reasonable vetting process, and also a process to retract a story if it's later shown to be false.
None of this is completely foolproof. It relies on institutions taking a long term view, and people working out which individuals and institutions are worthy of their trust. This isn't like the the blockchain where you have mathematical proof of veracity that's as strong as your encryption algorithm. I don't see how that level of proof is achievable in the real world.
1 reply →