Google Makes Use Of 'Deepfakes' To Counter-Attack 'Deepfakes' Videos

Over the years, fake news has become an emerging issue for people to cope with in the Internet era. With the rise of social networks, most users are still unaware of being outwitted by fake news and misinformation, which continue to flood our cyber life every day. However, our Internet literacy could be much further to which it is expected to be, considering another intense “fake news” technology: Deepfakes. To prevent potential threats for a safer future, Google has recently collaborated with Jigsaw to unveil a large source of visual “deepfakes” data.

First and foremost, we should take a closer look at how “deepfakes” work so properly when it comes to making fake videos. In short, “deepfakes” fools us with viral fake videos though they look original due to some latest technologies. The most important technology behind “deepfakes” is nothing more than deep learning, which can collect a huge dataset of targeted individuals including existing images and videos. Once the process is done, it will combine then super-impose them onto original videos to make a fake video realistic-seeming as real as possible.

As for Google, it has found a way to counter-attack “deepfakes” by using “deepfakes” itself. The company has been working on hundreds of video recordings of consenting actors to create the first-ever dataset of such "deepfakes". Aside from that, several publicly available “deepfake” methods were used to generate thousands of “deepfakes” from videos in the dataset. For researchers who want to develop their method in synthetic video detection, this free-to-use dataset will give them a lot of help for sure.

As of now, Google will continue to improve the dataset along with the assistance of other partners to fight against “deepfakes”.