[denimalpaca]

A good way to fight this is to cryptographically sign videos/audio. Might be a bit computationally expensive compared to a text message, but for very high-profile people, having a copy of the video/audio and it's cryptographic signature will be absolute proof against fakes. We don't even need AI algorithms to learn how to detect fakes in this case, it just becomes a matter of comparing hashes.

[roywiggins]

To an extent, it literally doesn't matter. If I fabricate a video and splash it around Facebook, it won't have a signature. Maybe Facebook will put a "has a signature" icon next to videos that are signed... but remind me, how much do web users notice when a website is HTTP vs HTTPS? Anyway, if a video is fully fabricated, there will be no hash to compare it with. And, by the time anyone does, it will already have been remixed and spread across Facebook.

There will be lots of good reasons for a video not to match bit-for-bit from CSPAN- if I add commentary on top of it, or intercut it with other clips Daily Show style, or just re-encode it, the hashes won't match. False positives are the death of an indicator, and there will be far too many of those for this sort of scheme to work.

[munk-a]

When there is a binary on the internets available from various mirrors with the primary authority on it distributing a SHA of it's contents then we can use that SHA to verify the contents of the binary...

The issue here is that there is no trusted authority on the video, the people distributing the video are the content originators so we can use a signature to verify the video wasn't corrupted in transmission but not that it wasn't edited before it was signed.

(this technique would work well if the concern was that videos from some primary source were being edited by a CDN entity helping to stream that video to viewers, which is entirely separate from the article)

[sixothree]

Wouldn't the authority have to be the device manufacturers?

[munk-a]

How would they technically accomplish this? Whenever a camcorder or phone or webcam was used would the hash be transmitted to the manufacturer and registered there? Would the device sign the video signature with some secret key? (similar to that big secret number you're not allowed to show)

Assuming you got it to work sanely, how would you ensure that all manufacturers properly locked down their devices?

Assuming there's a reasonable trust framework and you're able to enforce it... How would you allow legitimate video editing without allowing people to edit video to create misleading images? How would a movie cut between characters and an exploding model in a legitimate and signed manner without allowing someone to splice a politician's speech in with terrorist training videos?

There is a lot wrong with this "Tech can fix it" approach.

[xamuel]

This would only hinder amateurs. Pros will just intercept the camera's own sensors. (A simple version of which can be done by amateurs: take a certified video of an uncertified video.) Maybe you could build GPS into the signature, which could help (I'm not sure how hard it would be to feed a camera fake GPS). But even if all these tech solutions work 100% perfectly, all you've done is weeded out laymen. The REAL powers (such as governments or multinational corporations) will just make backdoor deals with certificate authorities.

[gowld]

How does the average user know whose signatures to trust? If Facebook censors videos by signatures, people will scream censorship and ignore the signatures.