Facebook AI Launches Its Deepfake Detection Challenge

Facebook’s AI division has been ethically producing deepfakes (manipulated videos or face swaps). The videos are part of a training data set that Facebook assembled for a competition called the Deepfake Detection Challenge that was launched yesterday. In this competition—produced in cooperation with Amazon, Microsoft, the nonprofit Partnership on AI, and academics from eight universities—researchers around the world are vying to create automated tools that can spot fraudulent media. Facebook has dedicated more than US $10 million for awards and grants.

The U.S. presidential elections in 2020 are an added incentive to get ahead of the problem, says Canton Ferrer (Facebook AI Red Team). He believes that media manipulation will become much more common over the coming year, and that the deepfakes will get much more sophisticated and believable. “We’re thinking about what will be happening a year from now,” he says. “It’s a cat-and-mouse approach.” Canton ­Ferrer’s team aims to give the cat a head start, so it will be ready to pounce.

It may seem odd that the data set compiled for Facebook’s competition is filled with unknown people doing unremarkable things. But a deepfake detector that works on those mundane videos should work equally well for videos featuring politicians. To make the Facebook challenge as realistic as possible, Canton Ferrer says his team used the most common open-source techniques to alter the videos—but he won’t name the methods, to avoid tipping off contestants. “In real life, they will not be able to ask the bad actors, ‘Can you tell me what method you used to make this deepfake?’” he says.

Why its hot: It’s interesting that Facebook is working to create this tech right after the infamous AOC versus MZ roast. Does this make up for the fact that they people make fake ads/is it going to be implemented on Facebook?

Source