Artificial intelligence (AI) can undoubtedly do great things, but it can also be a threat to objective reporting and ultimately to democracy. One of these threats is the spread of fake news through deepfake videos.
Deepfake uses AI image recognition technology to transfer one person’s face to another. This makes it possible to manipulate video material on a large scale.
What exactly is Deepfake?
Deepfake videos use AI image recognition technology to transfer one person’s face to another. This process is also called face swapping. In the case of Deepfakes, the software not only changes the face, but also adjusts the facial expression. If the body figure of the person is very similar to that of the potential “victim”, the fake video can look incredibly real.
Here is a sample video I made. Obama or Katzlberger?
Face swapping applications have been around for years
Face swapping has been used in film production for years, but it has required experienced CGI experts to spend many hours achieving decent results. There are also several apps in Apple’s and Google’s AppStores that offer Face Swap applications. Examples are the Facestar Video App or Faceapp.
What’s new now?
The really new thing about Deepfake is that the technology is qualitatively much better and, above all, accessible to everyone.
The software Photoshop from Adobe is already known to the broad mass of the population, since the term “photoshopped” stands for the manipulation of images, especially in the women’s magazine area.
Who developed Deepfake?
In autumn 2017 an anonymous Reddit user published several porn videos on the internet under the pseudonym “Deepfakes”. The scenes were not real because they were created with artificial intelligence. Over time, the Reddit community has fixed many bugs in the fake videos, making it increasingly difficult to distinguish fake from real content. As training data for the software, Reddit users used records of sex movies and the respective actress.
Shortly afterwards a new software called “FakeApp” was released, which was created by a Reddit user called deepfakeapp. The application makes it much easier for any half-talented programmer to transfer one person’s face to another.
The application analyzes hundreds of photos and videos of a person available (or uploaded) on the Internet. These recordings are embedded into a video in an elaborate calculation process until the facial expression is completely undisturbed. The process of creating a deepfake-based video can therefore take a very long time, even if only short clips are used.
The app uses TensorFlow, an artificial neural network from Google, and requires the power of a powerful graphics processor.
To include a person in a deepfake video, about 300-2000 images of a face are needed. Celebrities with lots of pictures on the net are therefore popular victims. Hollywood celebrities such as Gal Gadot, Arianna Grande and Taylor Swift are particularly hard hit by this trend. But also the faces of politicians like Angela Merkel, Vladimir Putin and Donald Trump were chased through the app. The result is deceptively real videos that make every journalist sit up and take notice. Authentication is becoming increasingly difficult.
In addition to Reddit, Twitter and Pornhub have banned the use of deepfake technology on their platforms. For example, Twitter has banned all accounts that post AI-generated videos or are used to post such clips.
In this thoroughly real video, Obama complains about his successor.
Recognizing deepfakes will probably become more and more difficult in the future. It remains to be hoped that there will then be technologies to detect counterfeits like these.