NOTE: I am NOT defending AI or it’s usage or anything like that. So just tie that horse right back up at the kitchen sink. I don’t like the whole AI thing. This post is not about AI.
I get that it was necessary at first, but now whenever there’s any hint of AI, there’s a great chorus of everyone saying the same thing ‘AI slop bla bla bla’ like a gaggle of sheep jerking their knees.
It’s not fucking helpful. It only sows even more division among the peasants, while the oligarchs take zero notice and continue on as usual. The correct way to engage with AI content you don’t like is just not at all.
Examples:
Recently I watched as a post for a community event was removed, because it had an AI image. The image was not requested or created by the organiser, it was provided to them, and had information in several languages. But the post was removed because AI slop and now there is less human connection happening. Yay. Well fucking done. You literally are helping the AI overlords now.
I’ve seen many actual legit digital artists getting hit with the cries of AI slop and even banned, when they have been making art for years before AI was even a thing. People actually can’t tell the difference half the time when they think they can, and end up hurting real artists.
Now is not the time to be attacking each other over using AI to generate some stupid meme. There are much, MUCH bigger issues at stake right now, and calling everything AI slop does nothing to help.
/rant
AI images include some special metadata in the file. It works sometimes!
Some AI providers add that metadata, yes. It’s what piefed searches for with its automatic filters. But metadata can be stripped, or not added at all in the first place.
My avatar image is AI-generated, for example. I did it using local AI models, edited it with image editing tools, converted formats, and so forth. I doubt it’s got anything the automated filters would recognize on it. If I didn’t tell you its pedigree would you know to report it? What if it actually isn’t AI generated after all, how can your report be tested for accuracy? That’s the sort of problem that OP is calling out here.
it doesnt need to catch 100% of cases
False positives are fine?
Some artist spends days working on some bit of art, posts it, gets the scarlet letters “AI” hung around their neck and now they not only have that particular post taken down they’re a “suspect” from then onward.
Yes it is fine. Posting is not a critical thing. If its falsely flagged they can submit a report to the mods to get it removed and everyone continues living.
Ah yes, mods are famously level-headed and will take the word of the person objecting to their actions. Same with the general commenter population of art communities.
well the alternative is no solution and those same people needing to make more decisions. So whats better mod review for 1 or two posts or mod/commentor review of all posts? If you cant accept false positives from this system the false positives from the community review must be unacceptable.
Yup, that’s what both I and OP are saying. This is not something that can be “solved” and this attempt at solving it is causing net negatives.
It causes far less negatives than doing nothing.
When automated detection fails then it’s up to people to manually report it and mods / admins to manually apply the label.
I was talking about screening and reporting posts and you’ve moved to profile pics for some reason?
Which is not reliable, and which therefore leads to the witch-hunting problem.
Whether an image actually is AI and whether some random person thinks it “looks” AI are not always in alignment. Especially considering that “is AI” has shades of grey too, given that lots of images are made using AI tools for only some aspects. And even “AI tool” is open to interpretation.
It was just an example of an AI-generated image that I could point to that I’m quite sure doesn’t have any special metadata attached to it. If for no other reason than I generated it years ago.