The Promise and Peril of Deepfakes

by MAi Research September 18, 2019

Technology is a wonderful thing, but it can also sometimes be terrifying. It opens so many doors, but those doors don’t always lead to good places. Consider for example the types of technology that compromise your privacy. Not just your browser history, but your face, your voice, your very identity.

This is the perilous promise of a relatively new technique for manipulating images known as “deepfakes,” which allow for the creation of new images or videos that combine more than one original. The results can often be hilarious.

Oh sure, it’s all good fun until it’s your face in the video…

 

Deepfakes have been gaining a lot of attention, especially with the recent release of the Zao deepfake face-swapping app, which quickly became the most downloaded free app in China. This means that millions of people now possess the power of deepfake technology.

This explosive popularity raises privacy concerns. The fact that apps like Zao now have a database full of user-submitted photos is obviously concerning. And as the accuracy of the technology increases and its availability widens, how long will it be until social media is flooded with increasingly realistic fake videos? Zao is currently being used for harmless social media fun, but the darker implications are anything but.

When Two AIs Go To War

One interesting aspect of deepfake technology that’s often overlooked is the artificial intelligence technique that makes it so effective: generative adversarial networks (GANs). In essence, a GAN pits two different neural networks against each other to improve the speed and effectiveness of their machine learning — in the case of deepfakes, leading to increasingly realistic composite images being generated with a speed far beyond what was possible just a few years ago.

Deepfakes are one the first applications of neural network technology to make a big splash among the public at large, and keep in mind that this technique is still in its infancy. The potential applications are as unknown as they are exhilarating — or terrifying, depending on your perspective.

What Does the Future of Deepfaking Look Like?

It’s hard to say until deepfakes become more accessible. Software such as FaceApp comes close to replicating the effect but it’s nowhere near as powerful. Since Zao is such a recent app, it’s safe to say that more and more people will try to put the technology in public hands. Although each time it happens, the same concerns about privacy are guaranteed to resurface.

At the moment, the only definitive statement anyone can make about deepfake technology is that it’s controversial. So far it’s been used for fun, such as stitching the face of a celebrity over a famous movie scene. So we can only speculate on what the future holds. Until deepfakes are used for something truly malicious, there’s no reason for serious concern. But there’s no question it’s a technology to keep an eye on — neural networks move fast, after all, and there’s no doubt they’ll be coming up with more surprises for us soon.