- Joined
- Mar 12, 2019
I could be wrong but GIS seems to have the capability of reading enough metadata to display the resolution of every image it pulls, so if they were able to convince sites to tag their images as AI with some standard in my opinion it's feasible. Of course the big challenge of all this is actually tagging AI, but the same reason AI art is so prominent is in my opinion why it's also relatively easy to moderate; AI art allows a user to churn out images like there's no tomorrow. You could generate 50 images and upload them all at once to an image site. This is why it's so easy for it to clog up image search results but it also makes the user responsible stick out like a sore thumb. It seems to me like it's usually still a small number of users generating the vast majority of the images, which makes it less of a game of whack-a-mole than if it were more users doing smaller amounts at a time. That seems like a relatively consistent pattern of behavior so far that makes moderating this stuff have a relatively low number of maintenance points.I'm not sure it would be easy to automate that. It seems like it is probably easier for AI to obfuscate its AI origin than for an opponent AI to identify such art despite obfuscation. So malicious actors could probably stay ahead of the arms race on that if they felt like doing so. And then there will be an even larger population that is completely apathetic about tagging it.
To be clear, I don't think any of this would be 100% effective; few solutions ever are. But I do think they seem like logical next steps which will eliminate the worst of the issue and render it a much more minor nuisance.
It's already happened on Reddit. Someone posted their art and it got banned because the brainlet hivemind thought it looked too much like AI art. Someone did an AI cover of some dub voice actor singing the Welcome To the Internet song and the VA in question managed to take it down with a DMCA. Not saying it's a good thing but I think it's inevitable given the current state of the internet that heavy handed moderation starts happening.And if you have it done by humans, they'll chimp out and falsely mark stuff as AI when it isn't.