3 ms·

I do not need to even read the article to sigh and think, "can spot it _today_." Obviously for reasons both benign and not so benign, we are going to enjoy an

I do not need to even read the article to sigh and think, "can spot it _today_." Obviously for reasons both benign and not so benign, we are going to enjoy an arms race with one side ever bettering verisimilitude (for any media you like), and another devising ever more esoteric and expensive techniques to try to discern the real. Against a moving backdrop in which the innocuous evolution of technologies which make the nominal separation between the the real and un-real increasingly vague. My picture is already touched up by god knows much AI before I see it as a compressed image on my phone. My video too. That's before I even selectively apply a filter or hit the Better button. If I record myself I'm going through a tool chain which already might do all sorts of things to transmute my voice. It's only a matter of time before we see ever-more nuanced targeted application of ML to replace "compression" with "confabulation", I mean, see today's other headline https://news.ycombinator.com/item?id=32907494 https://news.ycombinator.com/item?id=32907494 Soon confabulation may be the norm. The real will retreat like a god of the gaps to face-to-face interaction exclusively. Can researchers (sic) still find specific landmarks or meta-properties today which are tells? Sure! Sometimes. For now.

alexvoda 4y ago
I think the only durable approach is to have recording hardware digitally sign any picture/audio/video with a unique per device key that can not be extracted from the chip. AFAIK, this is already being developed.
- alexvoda 4y ago
  The next step is easy verification of the signature and creating the expectation that important stuff should be signed. The UX should fell simmilar to HTTPS vs HTTP. Of course, this means no reencoding, no downscaling, and no editing in general. Other apps should be able to sign but with their own key and maybe a user key which is different than than the key of the camera. Here the UX of HTTPS offers no inspiration. We trust the tree of trust of CAs. If one private key is leaked or misused it will be eliminated in a browser update. With recording authenticity we will need a UX and a desire for people to distinguish different types of origins. I can trust that picture A is the direct output of a Nikon Z7 camera. And I can trust that picture B was edited in Photoshop by X. And these two are very different.

alexvoda 4y ago

I think the only durable approach is to have recording hardware digitally sign any picture/audio/video with a unique per device key that can not be extracted from the chip. AFAIK, this is already being developed.

alexvoda 4y ago
The next step is easy verification of the signature and creating the expectation that important stuff should be signed. The UX should fell simmilar to HTTPS vs HTTP. Of course, this means no reencoding, no downscaling, and no editing in general. Other apps should be able to sign but with their own key and maybe a user key which is different than than the key of the camera. Here the UX of HTTPS offers no inspiration. We trust the tree of trust of CAs. If one private key is leaked or misused it will be eliminated in a browser update. With recording authenticity we will need a UX and a desire for people to distinguish different types of origins. I can trust that picture A is the direct output of a Nikon Z7 camera. And I can trust that picture B was edited in Photoshop by X. And these two are very different.