We get it: You took a set of tasteful nudes, and you want your finsta followers to see them.
Last year, duping Facebook’s moderation algorithm was as simple as overlaying your skin with dots or a grid.
But when Facebook caught on, it kicked off a game of nude whack-a-mole: Users switched to other shapes and patterns instead of dots.
Now the tech giant wants to end the cat-and-mouse game for good, with a group of investigators it calls the “AI Red Team.”
Facebook’s Red Team works the same way — but instead of a bug, they’re testing for patterns of failure.
Algorithms trip up more easily than you might think. Beyond the nudepocalypse, Facebook’s hate-speech detector might malfunction when users put multiple languages in the same post.
This problem isn’t just Facebook’s — tweak a few pixels, and Google’s AI might mistake a pair of skiers for a dog.
Last week, the company created an “equity team” to weed out racial bias on the platform.
But most of the Red Team’s work will focus on another threat: building an algorithm that smokes out deepfakes as easily as dick pics.