- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
Companies must ensure people know when they are interacting with artificially generated images, audio and text designed to look real
You must log in or register to comment.
I do wonder if they plan to draw a line somewhere in “how AI” the image needs to be. Like, if it’s a raw photo, that’s fine. If it’s enhanced by filters or something, probably not AI fair to say.
But what about the, like, magic eraser and infilling? If I use a tool to erase a bunch of people from a bridge and draw new parts of bridge that weren’t in the original photo… is that AI?
If I use AI to pull a mask and cut someone out from the background, and then put them overtop of a stock background, that uses AI to do work I could have done by hand. Probably fine. But what if I use AI to paint a new background that didn’t exist before this photo. Is that AI?
If I take a picture of me in a real place, doing a real thing, and then use AI infill to paint a different face onto my face so it’s not me anymore, but the rest of the photo including my body is real, is it AI because that person doesn’t exist?
I would say most of those materially alter the content of the picture. Something like filters or ordinary lighting settings alter how the picture appears, but don’t alter its content.
synthetic text, images, video and audio designed to look truthful must be visibly marked as AI-generated and contain a digital watermark to show the artificial origins of the content.
Great! Basically the browsers could then just highlight or filter images that are AI generated based on the digital metadata. What would be important that all image generators put those metadata tags in, and all image manipulation software leaves then in by default. And that if you paste an image in some image tool, it copies the AI label.
All this costs basically nothing and hurts nobody. It’s just adding and reading some metadata. And A LOT of the anger about not being able to tell what is real and what is not disappears. The responses are seriously silly
And of course they then need to enforce this. You’ll still have images pretending to be real which are fake, but at least you can easily and quickly filter out the marked ones. Basically Brandolini’s law, it takes the viewer much more energy to debunk bullshit he’s seeing than spreading it. So this helps at least in that. And violations can be fined, and social media will not also be encouraged to ban users posting fake images without markers.
A fundamental human right people should have is the “RIGHT TO REALITY”, to know what is real and what is fake and to not be deceived. The most extreme would be to put people into “The Matrix” or a simulation without them knowing. And what we’re seeing today is kinda like this - politicians lie and mainstream media obscures.
Yeah then some evidence of Andrew Mountbatten mutilating children comes out and they go “oh look, you can computer forensically tell the digital AI watermark has been removed! It must be fake!”
AI exists for one reason only: to help the ruling class get away with whatever the fuck they want.
I have a feeling that this will backfire. There will obviously still be a lot of unlabeled AI content made in/by foreign nations and local models, and since AI content is supposed to be labeled, people will be less cautious and more easily fooled by it.
At least you can quickly filter out posts that are tagged. It takes much more energy to refute bullshit than spreading it (Brandolini’s law).
And users or networks that are spreading untagged fake content can get fined and social media will be strongly encouraged to ban such users. That depends on the enforcement of course.
This is tautologous, if a label is needed to differentiate AI generated imagery how do you identify unlabelled generated images to enforce the rule.
The same way you detect any other fraud. It can range from analysis to whistleblowers. The point is not even to catch all fraud, it’s to catch it often enough that most people and companies will not attempt it.
There are analysis techniques that can determine if AI generated an image. They aren’t certain, but the can provide a % degree of confidence.
The problem is that AI will get better at avoiding this (even if it’s not being designed to specifically), and AI “ enhancement” also leaves similar artifacts (things like clearing up sensor-tearing or compression artifacts; iOS does a bunch of this and I think Android too)
Detection will always be an arms race; the real problem is how to we verify those techniques? How do we determine that government witnesses are testifying to verifiable methods? If we cannot do that then everything falls apart, from responsible disclosure in news articles to trials in a court of law.
EU is bi-polar. Making some of the best regulations (GDPR, AI Act…) and then suddenly fucking people over with stuff like Chat Control.
Chat control was US big techs baby. They wanted to keep scanning peoples messages but GDPR messes with that so they’ve been trying to get Brussels to give them an exception forever. The “CSAM” angle is just plausible deniability
The rules will not apply to users’ personal content
So… easy workaround then? It’s just my personal content bro, not at all paid for by the state/business interests
It’s nice to see what semi-competent adults running a region can accomplish…
glares at a giant whiny baby of a country rolling in its own shit
