Hacker News new | ask | show | jobs
by rishabhsagar 3142 days ago
Wait, what? I submit my nude pics to FB, so that they can hash it and identify its copies online? That sounds good, but is that the only way of doing this?

Why not use facial recognition + nudity detector to on-demand find potentially compromising videos online? Maybe even send my Facebook notification that says, "Hey we found what appears to be a compromising video with you in it shared to a wide group of people, are you cool with it?"

Am I mis-understanding the concept being proposed?

2 comments

I like your idea better. For an SV darling like Facebook (and especially since this is Facebook), what the article describes sound incredibly manual and low-tech.

But what really bothers me is

> ... a community operations analyst will access the image and hash it to prevent future instances from being uploaded or shared.

Of course, this community operations analyst will not stash those photos in a flash drive somewhere, right? Because the humans FB hires are 100% trustworthy. /s

What's "SV"?
"Silicon Valley"
What facial recognition technology do you have in mind that would not face the burden of frequent false positives (nevermind false negatives) and could scale to do this efficiently on an on-demand basis?

Then there's the issue of how your proposed system requires FB to show users the porn that other users are privately viewing.

Facebook already uses facial recognition to suggest tags on photos, so I don't think false positives are that much of an issue:

https://www.facebook.com/help/122175507864081

You other points are true though, and I guess a lot of these type of photos don't always contain faces.

I'm not as heavy of a FB user as I was 5 years ago, but my impression was that false positives were at an acceptable threshold, but not nearly a solved problem. And the face recognition tech was heavily augmented by social data, such as the fact that me and a friend were both partying at the same place on the same night and that we had just 3 days ago posted and tagged photos of each other.

That secondary data does not exist reliably in the context of revenge porn videos that go viral. Nevermind that the penalty of a false positive -- which would presumably involve exposing the user to a bit of random porn -- is substantially higher than it ever is with mistagging friend photos.

>> Then there's the issue of how your proposed system requires FB to show users the porn that other users are privately viewing.

I think this can be addressed as "Show me all suspected compromising videos of me that is available to more than 10 users". The answer to that query can be produced with facial recognition and nudity detector, no?

You don't think falsely-identified users would be bothered by notifications to check random porn videos? The same userbase who freaks out (somewhat justifiably) when Facebook suggests friends/connections that the user believes (perhaps wrongly) that FB could not know independently?