this post was submitted on 19 Sep 2023
74 points (88.5% liked)
Lemmy
12557 readers
5 users here now
Everything about Lemmy; bugs, gripes, praises, and advocacy.
For discussion about the lemmy.ml instance, go to [email protected].
founded 4 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Are any of the examples that your provided libre/free and open-source? I wasn't able to find any info for Google's, and Cloudflare seems to only offer theirs for free if you are already using Cloudflare's services. If not the examples that you provided, does there exist any tools that are libre/free and open-source?
AI will make this redundant
It will also make it a battle of attrition. Because now we’re not only using AI to block CSAM; Trolls are using AI to generate CSAM.
The issue is that these tools typically work by hashing the image (or a specific section of the image) and checking it against a database of known CSAM. That way you never actually need to view the file to compare it to the list. But with AI image generation, that list of known CSAM is essentially useless because trolls can just generate new images.
Even without the issue of new AI-generated images, those hash-based scanning tools aren't available to hobbyist projects like the typical Lemmy instance. If they were given to hobbyist projects, it would be really easy for an abuser to just tweak their image collection until it didn't set off the filter.
You can use CloudFlare’s CSAM scanning tool completely for free. You can’t get access to the hashes, which would allow what you are talking about.
Sure, for Lemmy instances who are Cloudflare customers. But I don't think it can be integrated with the Lemmy code by default.
No it can’t, and it shouldn’t be. It’s better to stop the CSAM before it ever makes it to any server you control rather than wait and then need to deal with it.
Bingo, that's the issue. With an endless supply of fresh content, hash checking is dead
On the other hand, if the people who want those images can satisfy their urges using AI fakes, that could mean less spreading of images of actual abuse. It might even mean less abuse happening.
However, because they're terrible people, I have to suspect that's not the case.
People who create the content and insane monsters, but a LOT of actual pedos (vs predators looking for a power play) are disgusted by their preference. I know a ton of them look to cartoons already for stimulation, so I think AI content could draw more people away from actual material. Hopefully if demand reduces there will be less creation of new real content as the potential profits fall more proportionate to the risk.