I'm more interested in how exactly you'd implement something like this.
It's not like videos viewed on tiktok display a hash for the file you're viewing; and users wouldn't look at that data anyway, especially those that would be swayed by a deep fake...
Like you said, the issue is in verification by the end-user. It is trivial to provide a digitally signed (and timestamped) file. It is also trivial to provide trusted tools to verify these files. It is immensely difficult to provide a solution user will care about; which is why more often than not the most people asks companies in the data authenticity business is "can we show a green check on screen? That would be perfect!".
And we end up with something that nobody checks beyond the "it's probably ok" phase. If the goal is to teach the masses about trusting their source, either they have a miracle solution, or it just won't work. (and all that is assuming people actually care about checking the authenticity of the stuff they see, which is not a norm as it is…)
This is the best summary I could come up with:
The White House is increasingly aware that the American public needs a way to tell that statements from President Joe Biden and related information are real in the new age of easy-to-use generative AI.
Big Tech players such as Meta, Google, Microsoft, and a range of startups have raced to release consumer-friendly AI tools, leading to a new wave of deepfakes — last month, an AI-generated robocall attempted to undermine voting efforts related to the 2024 presidential election using Biden's voice.
Yet, there is no end in sight for more sophisticated new generative-AI tools that make it easy for people with little to no technical know-how to create fake images, videos, and calls that seem authentic.
Ben Buchanan, Biden's Special Advisor for Artificial Intelligence, told Business Insider that the White House is working on a way to verify all of its official communications due to the rise in fake generative-AI content.
While last year's executive order on AI created an AI Safety Institute at the Department of Commerce tasked with creating standards for watermarking content to show provenance, the effort to verify White House communications is separate.
Ultimately, the goal is to ensure that anyone who sees a video of Biden released by the White House can immediately tell it is authentic and unaltered by a third party.
The original article contains 367 words, the summary contains 218 words. Saved 41%. I'm a bot and I'm open source!
If the White House actually makes the deep fakes, do they count as "fakes?"
Honestly I'd say that's on the way for any video or photographic evidence.
You'd need a device private key to sign with, probably internet connectivity for a timestamp from a third party.
Could have lidar included as well so you can verify that it's not pointing at a video source of something fake.
Is there a cryptographically secure version of GPS too? Not sure if that's even possible, and it's the weekend so I'm done thinking.
It's way more feasible to simply require social media sites to do the verification and display something like a blue check on verified videos.
This is actually a really good idea. Sure there will still be deepfakes out there, but at least a deepfake that claims to be from a trusted source can be removed relatively easily.
Theoretically a social media site could boost content that was verified over content that isn't, but that would require social media sites to not be bad actors, which I don't have a lot of hope in.
This is a most excellent place for technology news and articles.
