Right exactly, much like p hacking with studies. Publish enough research papers and you get significant results by chance, and it gets into a headline as "Science shows..." and people believe it.
Unfortunately in this space the amount of deepfakes we can create is massive, so even a system thats 99.999% accurate will leave a lot of fakes through and grant them credibility.
I think we have to focus on an alternative path where we assume any digital content can be fake, unless its creation is provable through some verified sources or methods.
Maybe it's possible to embed some kind of digital fingerprint into images? So as to say, this image was definitely taken by this camera. Pretty sure something like this has been done.
I ended up reading that whole thing, but by digital fingerprint I meant something like embedding a cryptographic checksum in an image. Or some way of proving an image is produced by a specific camera. Itd be great for example if we could verify a security image video was real and not a fake
Unfortunately in this space the amount of deepfakes we can create is massive, so even a system thats 99.999% accurate will leave a lot of fakes through and grant them credibility.
I think we have to focus on an alternative path where we assume any digital content can be fake, unless its creation is provable through some verified sources or methods.
Maybe it's possible to embed some kind of digital fingerprint into images? So as to say, this image was definitely taken by this camera. Pretty sure something like this has been done.