These companies should be criminally liable for training their models on child porn.
One concerning thing is that when these AI companies are pushed to crack down on this, they’ll retrain and bias their models to only recognize traditional “pornstar” body types as ‘adult’.
This is a problem when they then get used as a mass content flagging and moderation tool — it’s not hard to imagine some adult woman taking nudes (as people do) and just because they don’t have a huge bust, getting auto flagged as having illegal material. Not to mention body types that aren’t even real, like the art made in furry communities.
It’s not like there isn’t precedent, recently Discord banned a ton of people because its scanning system started throwing false positives, and that was only corrected because it made such a stink.
They should be made to retrain their models using only images of those palaeolithic venus figurines.
AI shouldn’t train on porn at all. Too many ethical issues.
Yeah we will drown in flase postives, my other worry is they will spend on this auto scaning of media call it problem solved slash bugets for other things and predators move to dumb tech or really encoded shit.
Thats if there even is enouth money/compute to scan everything…
https://tsjournal.org/index.php/jots/article/view/29
exposure to csam, through self reporting, increases the risk the viewer will seek contact with children, in case you need more arguments against csam. I’ve heard the argument that AI generated images may be a harm reduction situation. it’s not. it’s incredibly damaging to the viewer if they are seeking it out, it makes them more likely to abuse, if they are exposed to it as victims it can cause lasting damage, much of it is just remixed images of actual child abuse victims, and that revictimizes them, and causes serious damage as well. We really need to stop pretending the companies that make this software aren’t responsible for this software.
This study explores a sample of 1,546 anonymous individuals
I have a study of 23023012301230123050503450 anonymous individuals, of which 78% agreed that you smell like doo doo. Yep. True story. If you disagree, you hate science. How could you do this?
I shouldn’t be this sarcastic, the journal is called " Journal of Online Trust and Safety (JOTS) " with a name like that, how could I ever be so arrogant to not believe 1546 anonymous individuals on the internet? Shame on me, really.
Seems to be run by academics
reddit moment…
you can’t just dismiss an argument because the study is small. Sure it’s small, but it’s not easy to get large data on this subject.
The concept of csam exposure influencing behaviour is highly plausible.
The concept of csam exposure influencing behaviour is highly plausible.
The entire scientific community and publishing industry is a minefield. So is CSAM and the call for more restrictive governing of the internet.
The thing I know for sure is that there are people on this planet who have a political and a financial motive to lie and fabricate studies to justify their politics.
The thing I’m mostly skeptical on is the scientific process. Not because it can’t work, but because that entire field too, has bad incentives. “Reputation” is held up, frankly, by boomers who have trouble operating a printer, but did some important theoretical work before the internet was made public and by politics and a disinterested public that doesn’t apply scrutiny to what their “intellectual elite” is actually doing.
The internet is a KNOWN source of bad information. Those 1546 “anonymous individuals” could be one bored 4chan user, or one chinese/russian/us spy agency worker justifying the next expansion of their budget.
is highly plausible.
Is not science. It’s guesswork.
I appreciate that it is difficult to collect good information on this subject, but that’s just how it is. Tough luck. As a scientist, a proper scientist, you don’t get to publish “it’s plausible” from a source you can’t verify as fact.
Second guessing science isn’t anti-science.
It’s shit like that is why I vote for the asteroid.




