The most popular and comprehensive Open Source ECM platform
Deepfakes are videos or images that have been altered in ways to make it difficult to detect that it is not authentic. Deepfakes often make celebrities and public figures say things or appear in situations that are a spoof on reality or an attempt to deceive the viewer.
Deepfakes have evolved from early days of photoshop alterations to images to now increasingly sophisticated alterations of videos and images. The problem is that it can become difficult for people to know when something is real or synthetic.
In order to combat deepfakes, Facebook, Microsoft and Amazon offered a ‘deepfake deteection challenge‘ to developers to create an algorithm that can spot when a video or image is fake or not.
The results of the competition were not that encouraging. Contestants were given 124k deepfake videos to train their algorithm with. Even with those known videos, the best algorithms were able to spot fakes on 85 percent of the time. When the algorithms were tested with a different set of 10,000 unreleased videos, the best algorithm dropped to only 65 percent accuracy.
Hany Farid, a professor at UC Berkeley, told Wired that “it’s all fine and good for helping human moderators, but it’s obviously not even close to the level of accuracy that you need. You need to make mistakes on the order of one in a billion, something like that.”
Andrew Gully, technical research manager at Google, said that “we know already how difficult it is to convince people in the face of their own biases. Detecting a deepfake video is hard enough, but that’s easy compared to how difficult it is to convince people of things they don’t want to believe.”
Mike Schroepfer, Facebook’s chief technology officer, said that “the lesson I learned the hard way over the last couple of years, is I want to be prepared in advance and not be caught flat footed. want to be really prepared for a lot of bad stuff that never happens rather than the other way around.”