I’m thinking a community driven open source project. Maybe it could connect to PiHole somehow? Basically if anything is AI in any way, it blocks you from accessing it.
So, lets say you click a youtube video. The blocker would quick scan the URL, compare it against it’s database, and as long as the profile for that URL doesn’t come back with any flags, it lets you through. If it lets you through and it’s still AI, YOU can report it back to the database with a click of a button, and now anyone else who ever visits that URL for that youtube video, with the blocker installed ,will instead get a screen that says “The page you are attempting to access has been reported as AI.” Then you could scroll the list of reports, and choose to load the page anyways, or go back.
If you choose to load the site anyways after being warned, you can still give a report. Maybe this URL doesn’t have many reports, and maybe some are lacking or outdated. You could add your report. Then when you submit it, it would be more helpful to future visitors. Maybe the okl site had some issue and they got a report. And then 10 years later, the report has the innitial report from 10 years ago, which was true at the time. The site has since relaunched or redesigned. And whatever AI was there has been removed. Or changed. You could include that the AI reports are no longer valid. This triggers a vote the next 10 people vote on whether to remove the flags from this profile. It would require 10 community members to all vote remove. If it’s a mixed result, the flag stays. If all 10 vote remove, the flag is deleted and no longer considered to be made with AI.
And you as the user can choose which level of AI you’re willing to tolerate. Me personally, no AI please. None at all. No images made with AI. No Youtube scripts written by AI even if they’re performed and edited by real people. If even the script is AI written, no. Maybe you’re ok with a human performed AI written script. That’s how you’d set your settings, and you wouldn’t get the flag screen. I don’t want to see that video, so I’d set my settings my way, and I’d see the flag screen.
And there would be dates, but not usernames. The program knows who said what, but that information isn’t publically shown, and no account is needed to write a flag report. So you could see context of how long it’s been for things, but you can’t become famous just because you set a bunch of false flags.
Oh, I guess we should think about how to cat and mouse the bad actors.
Ok, what if instead of flagged URLs aren’t immediately flagged. They’re reviewed similarly to removing a a flag. 100 users will have to vote. 80% yes needed to create a flag. My logic is, that way you can’t have a bunch of trolls flagging every website they see. More users vote to diminish each voting power. 80% requirement instead of 100% to cancel out the votes of trolls.
I figure trolls are going to want to flag everything, but probably won’t want to de-flag anything. So there’s more leniency there, to bother less users.
Now the only question is…is this even feasible???
terranoid@lemmy.cafe 3 hours ago
Yes and no. The more you want to block it, the more technically involved it becomes.
High level, yes, if you scope it to a specific problem domain. Want to focus on YouTube videos and channels, with report functionality plus custom blocklists? Sure, that would be doable as something like a browser extension.
As soon as you add reporting, consider that idiots report bad stuff and stuff they don’t like just because, and people abuse it. You end up needing to moderate blocklists and users that can report.
If you want to go to the image level, yes but as an extension you are probably checking the hash of every single fucking image and have some sort of hash prefix map or bloom filter to match against, like Google safe browsing. This becomes a pain and the thing about AI is they generate new shit constantly. And they evade hash checks by cropping on purpose, avoid dupe detection on social media by doing that plus ocr checks.
Ever notice a word gets a letter bleeped out and people are like why did they censor that? Sometimes it’s due to ocr for dupe.image detection and they defeat it by changing the text. They also defeat your basic hash checks by flipping any pixels.
This problem of “how do I curate stuff so I only see what I like” can be a really hard problem and people are constantly trying to defeat it to make bots with high stats to use for ads.
I would figure out specifically what is frustrating you and solving the problem in a narrow space like YouTube channels, curators. Or GitHub users, repos using AI, etc. Solve a specific problem and do it well. if it’s specific enough you can come up with good ways of doing it that solve it really well, and other people might appreciate it. Then it can become a part of an anti AI suite rather than just doing everything but badly.