Don't mark AI generated content as fake

There is growing concern around AI image, video & audio generation, and I believe that this will only become more important of a topic as time goes on. Models are only going to become more sophisticated & better able to mimic things in a way that will become indistinguishable from reality. While we have had this issue before with "photoshopping" of images, the skill required to forge a ph

oto, was time & cost prohibitive at scale. But, we've completely lowered the floor, while raising the quality of editing of images. We may still be in an uncanny valley right now, where you can sort of just "feel" that an image is generated, the flooding of these images will become harder & harder to detect.

While there may be efforts to combat this at legislative level, like the EU Artificial Intelligence Act, which say that AI generated images:

shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated

-EU AI Act Article 50(2)

I really don't think that this is the right way forward, since whatever mechanism you come up with will be forged in some way. I do however think that what we need here is not something which proves something as fake, but something which proves something as real. This is a subtle distinction, but I think it is an important one.

If you think about it, why do we not already have standards for declaring that an image is real? Like we just implicitly trust that an image on a news site is real, but there have been many times where the news has been duped into running fake stories before.

This essentially the difference between a default open system & a default closed system. If a photo, for example, cannot be proved to be generated or not, shouldn't you just mark it as fake since you cannot know if it is real or not? Let's be honest that there will always be model makers which generate images in a way that they don't add this AI generated tag, or there will be systems which modify the image to "make it real".

Instead, I would support a system where you can safely say, this photo was taken on Nick's iPhone, at this time, and potentially at this location, and you can verify this at xyz.com or example.com. I think that there could be the potential of an app on your phone that validates these signatures and validate any image you place inside. To go even further, this might even be done at the OS level to mark all content displayed to you as valid/generated/unknown.

In summary, I think that the EU legislation is fighting an uphill battle of trying to mark fake things as fake, when they instead should mark real things as real.