So there’s this AI image tool, right, called Stable Diffusion, the one that made this open-source generative image tool called Stability AI? Well, turns out they had to remove this huge dataset they were using because it had some really gnarly stuff in it. I’m talking about child sexual abuse material, or CSAM. That’s not okay.
Some real smart people over at Stanford figured this out and called out the company and some other big players in the AI game for using these massive datasets that are just chock full of images scraped from all over the internet, many of which are depicting harm to minors. This ain’t just an issue on its own, this stuff is illegal.
But here’s where it gets sticky. The dataset didn’t actually include the illegal images, just links to where the images came from on the web. And to make it even trickier, many of those links led to images that had already been taken down. So it’s like this AI was being trained using all this potentially illegal data without even knowing it!
The folks at Stability AI and the company that made the dataset, LAION, haven’t said anything yet, but there’s pressure for them to respond. Because this isn’t just about nasty images. There’s serious privacy, copyright, and safety concerns here. And it’s not just bad for the folks who might be potentially exploited or abused, it’s also bad for the reputation of AI tech overall.
This is one of those situations where it’s clear that there needs to be a big change in how this stuff is handled. The datasets and models linked to them probably need to be set aside, or at least heavily sanitized, according to the Stanford researchers who first flagged this issue.
It’s a tricky problem because we want to encourage open-source development. That’s how you get different folks from different backgrounds working on this tech. We definitely don’t want a small group of big corporations controlling everything. But putting all that power in the hands of the public creates problems too, especially if the public doesn’t know what they’re working with. The potential for abuse here is huge.