'Perfectly Real' Deepfake Videos Could Be Just 6 Months Away Says Research Professor
If a new report is to be believed, deepfake technology is about to ramp up the realism factor while making it increasingly more accessible to the general public. While this is an interesting development for people that like to do this sort of digital trickery for fun; in the age of "fake news" it has the potential for creating more upheaval considering how quickly videos can go viral on social media.
According Hao Li, a University of Southern California (USC) associate professor of computer science, deepfake videos have the potential to look "perfectly real" within six to 12 months from now. In many cases with current technology, you can often tell that videos have been altered. This was readily apparent when the tech was first introduced over a year ago, but it's becoming even harder to distinguish these days.
"There also are examples that are really, really convincing," said Li in an interview with CNBC. "Soon, it's going to get to the point where there is no way that we can actually detect [deepfakes] anymore, so we have to look at other types of solutions."
Deepfakes have taken the internet by storm, as the technology allows you to manipulate existing videos by mapping someone else's face on the original source material. Not surprisingly, some of the earliest examples of deepfake videos popped up in the adult film industry (where famous celebrities have their faces digitally pasted onto the bodies of porn performers), and they've spread to include manipulating the footage of political figures as well.
Deepfake technology is also being used to alter videos of public figures and their voice patterns to have them say things that never actually came from their mouths in real life. We've seen these types of deepfakes used with political figures like former President Barack Obama, President Donald Trump, and Russian President Vladimir Putin.
Other deepfakes have emerged including DeepNude, which was an app that allowed users to "disrobe" women using machine learning via a generative adversarial network (GAN). Needless to say, the social media backlash over the app came swiftly (particularly considering the rise of revenge porn).
In case you haven't heard, #ZAO is a Chinese app which completely blew up since Friday. Best application of 'Deepfake'-style AI facial replacement I've ever seen.— Allan Xia (@AllanXia) September 1, 2019
Here's an example of me as DiCaprio (generated in under 8 secs from that one photo in the thumbnail) 🤯 pic.twitter.com/1RpnJJ3wgT
Li says that the emergence of apps like DeepNude and Zao have led to his more aggressive timeline for when deepfakes will be nearly indistinguishable from the real deal. Zao allows you to insert your face onto popular characters from films using just a few selfies instead of thousands upon thousands of images that have to be crunched through for previous deepfake videos.
So, what can we do to combat these deepfakes since they don't appear to be going away anytime soon? Li adds that we'll simply need to adopt better technology to detect these imposters and find the limits of the technology.
“The real question is how can we detect videos where the intention is something that is used to deceive people or something that has a harmful consequence," Li concluded.