Live data from Hacker News

Facebook flooded with 'sextortion' and revenge porn, files reveal

theguardian.com

11–20 of 81 posts

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#11
post #5

Earlier quoted context omitted.

They also use image recognition to detect nudity and previously classified child porn.

No, they use Magic Number among a bunch of various tricks. You can't really call it image recognition, more like 'data-hash comparison.' To train true image recognition would require a LARGE subset of stored illegal pornography.

Why? Just use two separate datasets, one of images of adults and children, to learn age, the other of porn and non-porn, to learn what's "nudity" or "sex".

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#12
post #5

Earlier quoted context omitted.

They also use image recognition to detect nudity and previously classified child porn.

No, they use Magic Number among a bunch of various tricks. You can't really call it image recognition, more like 'data-hash comparison.' To train true image recognition would require a LARGE subset of stored illegal pornography.

What do you mean by "Magic Number" in this context?

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#13

> "54,000 potential cases... and 33 of the cases reviewed involved children." That's 0.06% of reviewed cases, which is frankly less than I would have thought. I suspect that child abusers are a more savvy lot than the average chucklehead who does something like this.

I'd imagine that the low hanging fruits are picked off by law enforcement pretty quickly.

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#14
It's disappointing, but not super surprising.

Shit like this is why I'd never want to operate a site with a large user base that can freely upload content. Sure you could probably afford the servers for awhile with minimal monetization, but could you pay for even a half-dozen policy experts to moderate the content to a point where you don't send your largest potential audiences running screaming for the hills?

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#15
post #5

Earlier quoted context omitted.

They also use image recognition to detect nudity and previously classified child porn.

No, they use Magic Number among a bunch of various tricks. You can't really call it image recognition, more like 'data-hash comparison.' To train true image recognition would require a LARGE subset of stored illegal pornography.

I mean if companies like Google train ML models on millions/billions of user's private data without looking at it, then surely it would be possible to train this in an ethical manner for ethical purposes. And via some Gov collaboration on the legal side

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#16
post #11

Earlier quoted context omitted.

No, they use Magic Number among a bunch of various tricks. You can't really call it image recognition, more like 'data-hash comparison.' To train true image recognition would require a LARGE subset of stored illegal pornography.

Why? Just use two separate datasets, one of images of adults and children, to learn age, the other of porn and non-porn, to learn what's "nudity" or "sex".

Knowledge transfer training is really hard.

If you can do this (with different categories, for instance, determining what is nudity, then determining black people, to automatically classify porn with black actors automatically), it would be quite impressive. Results have nearly 5% error on CIFAR-10, which is full of clearly distinct categories - planes versus boats.

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#17

>and the challenges for moderators, who say they are overwhelmed by the volume of work, which means they often have “just 10 seconds” to make a decision. Maybe Facebook should just hire more moderators? I know they'd love to totally relegate this to some awesome NLP algorithm but we're not there yet. Facebook makes more than enough money to bring on some more employees (and clean up their guidelines). EDIT: oops, had…

> Maybe Facebook should just hire more moderators

They are doing exactly that this year

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#18
I had a female friend last month whose ex-bf created a fake account and proceeded to set the profile and public photos and videos of them (having sex). He then went ahead and started sending friend requests to her entire family and friends. I felt so bad for her - it was awful enough to make her even have suicide thoughts.

What really surprised me was how horrible the FB system to report stuff like this is. It is basically non-existant. If this thing takes 1 month to be reviewed, that is 1 (eternal) month where the damage can be irreversible.

what is worse is that FB has posts about how they are "fighting" this without having any instructions or obvious process defined to report this. So you "report inappropriate content" and the report goes to a black hole while the perpetrator continues with their shenanigans. They dont even need to create a new acct, because they can keep going without the offending account getting blocked.

Something else is that the non-profit foundations that exist to fight this (the ones that will send legal take-down notices) are US-centric. This means that if you are from a latin american country and you face this (as was this particular case), you are screwed.

Yes obviously the lesson is not to take pics like that in the first place, but sadly a lot of the victims may not be that computer savvy or mature enough to realize what a mistake those things can be.

Re: Facebook flooded with 'sextortion' and revenge porn, files reveal

#19
post #5

Earlier quoted context omitted.

They also use image recognition to detect nudity and previously classified child porn.

No, they use Magic Number among a bunch of various tricks. You can't really call it image recognition, more like 'data-hash comparison.' To train true image recognition would require a LARGE subset of stored illegal pornography.

You are confusing image recognition with classification. They use recognition, similar to reverse image search services. Not really 'data-hash comparison', but rather 'image-hash', since it may see through re-compression and other minor modifications.
Post reply on HN