At MIT Technology Review’s EmTech conference last month, deepfake expert Hao Li, whose team worked on putting the late Paul Walker (or rather, his digitally recreated self) in Furious 7 , said, “Our guess is that in two to three years, [deepfake] is going to be perfect. There will be no way to tell if it’s real or not, so we have to take a different approach.”
In an interesting coincidence, a few days after Li’s ominous proclamation, Google released an open-source dataset, containing 3,000 original manipulated videos, to aid researchers develop deepfake detection systems. The tech giant, in this pursuit, has collaborated with its incubator, Jigsaw.
The technique used in Furious 7 is only similar and not the same as deepfake, which has been used to reach many malignant ends, like fake news proliferation and fake celebrity pornography. Scarlett Johansson, a victim of deepfake porn, told The Washington Post last year, “Clearly, this doesn’t affect me as much because people assume it’s not actually me in a porno, however demeaning it is. I think it’s a useless pursuit, legally, mostly because the internet is a vast wormhole of darkness that eats itself.”
With the US Presidential election less than a year away, fake news propagation, using deepfake, is a looming threat. Google’s move, hence, is timely. “We firmly believe in supporting a thriving research community around mitigating potential harms from misuses of synthetic media, and [the] release of our deepfake dataset in the FaceForensics benchmark is an important step in that direction,” wrote Nick Dufour of Google Research and Andrew Gully of Jigsaw in the Google AI Blog ‘Contributing Data to Deepfake Detection Research’ on September 24, 2019.
To make the dataset, Google and Jigsaw worked with 28 actors, over the past year, to record hundreds of videos. “Using publicly available deepfake generation methods, we then created thousands of deepfakes from these videos,” Nick and Andrew explained, “The resulting videos, real and fake, comprise our contribution, which we created to directly support deepfake detection efforts. As part of the FaceForensics benchmark, this dataset is now available, free to the research community, for use in developing synthetic video detection methods.”
Last January, Google released a dataset of synthetic speech, in support of the Automatic Speaker Verification Spoofing And Countermeasures Challenge, to develop high-performance fake audio detectors.
