this post was submitted on 23 Dec 2023
25 points (82.1% liked)

Canada

7206 readers
354 users here now

What's going on Canada?

Communities

🍁 Meta

πŸ—ΊοΈ Provinces / Territories

πŸ™οΈ Cities / Local Communities

πŸ’ SportsHockey

Football (NFL)

  • List of All Teams: unknown

Football (CFL)

  • List of All Teams: unknown

Baseball

Basketball

Soccer

πŸ’» Universities

πŸ’΅ Finance / Shopping

πŸ—£οΈ Politics

🍁 Social and Culture

Rules

Reminder that the rules for lemmy.ca also apply here. See the sidebar on the homepage:

https://lemmy.ca

founded 3 years ago
MODERATORS
otter@lemmy.ca
25
'Slap in the face': Images of Canadian child abuse victims training AI generators (www.winnipegfreepress.com)
submitted 11 months ago by grte@lemmy.ca to c/canada@lemmy.ca
7 comments fedilink hide all child comments
 

Non-Paywall Article

top 7 comments
sorted by: hot top controversial new old
[–] baconisaveg@lemmy.ca 5 points 11 months ago

On the other hand... there's an ad in our elevator for a Little Warrior Camp, for children of sexual abuse. In the background it plays images of kids, with a disclaimer at the bottom that "these are just models, not actual victims". Seriously, that seems like the perfect use case for AI generated people. Imagine posing for a picture as a child and finding out it was used as an example of a CA victim.

[–] Grimy@lemmy.world -2 points 11 months ago* (last edited 11 months ago) (2 children)

They are currently getting rid of them. It's a database with 5 billion images, it's not feasible for someone to go through it one by one.

I'm happy it got found by a new AI filter, hopefully it can also be used to get rid of the websites that hosted them in the first place.

But a slap to the face to who? The anti AI crowd needs to grow up.

[–] swordgeek@lemmy.ca 15 points 11 months ago (1 children)

The database started out empty. They added all of the content. The filtering should have been part of the intake process, not after the fact. Image recognition has beem used to detect CP for many years now.

They could have and should gave stopped these images from getting into the dataset at all, but they didn't. Consequently, people who were victimized as children are having the exploitive images of them being used to generate new (synthetic) child porn.

[–] Grimy@lemmy.world 3 points 11 months ago

They did run filters. The group that found the new ones made a completely new stronger filter that is better at detecting it. You can't blame them for not using technology that just wasn't available at the time. They also pulled the whole dataset the moment the group alerted them to it and removed them.

[–] grte@lemmy.ca 12 points 11 months ago* (last edited 11 months ago) (2 children)

This watch dog group was able to find this content. You don't think the producer of the database should have any responsibility for the content within it? If it's not feasible for you to guarantee that the contents of your product are legal/ethical, maybe that's a problem?

[–] jadero@lemmy.ca 3 points 11 months ago

I'm not sure about guarantee. That implies perfection which is never attainable in anything. But requiring transparent evidence of due diligence is certainly doable. As are penalties for failure to meet some kind of standard.

It's past time to institute "grading standards" on large datasets. I have in mind the same kind of statistical standards that are applied in various kinds of defect and contamination analysis. For example, nobody ever guarantees that your food is free of animal feces, only that a fair and representative sample didn't find any.

[–] Grimy@lemmy.world -2 points 11 months ago

The watch dog group made a completely new filter for it.

Yes, the producers should be running all available filters and they did. This one simply wasn't available.