Deepfakes are videos that have been modified with the help of artificial intelligence (AI) and corresponding software. This creates the possibility of having a random person do or say whatever you want. Experts are in disagreement concerning the use of the term "deepfake" to describe such videos. Many say it’s too judgmental, as it primarily emphasizes negative aspects of the technology. Thus, many prefer the use of synthetic media. Disinformation expert Kathryn Harris, CEO of Deeptrust Alliance, distinguishes between deep and cheap fakes.
Deepfake production is the process of putting one person's face onto another. The quality of the video material of the person who will be manipulated is important for the outcome of the final deepfake. The better the raw material, the better the result. This applies to both the resolution and aspects such as lighting and background.
Artificial intelligence then analyzes all the data of the source material. This can be several hundred images or video frames. The AI then mimics the speaker's lip movements, frowns, and facial expressions. At this point, this sort of image manipulation is even possible with live footage.
This is the classic version of deepfakes. But the possibilities are endless. For the untrained eye, it is now almost impossible to recognize professionally produced deepfakes. Several companies around the world are therefore working on so-called deepfake detection software, including the Danish company Defudger. As technology advances, it becomes more and more difficult for detection software to keep up with newly developed deepfake methods.
There are many possibilities to detect deepfakes. One method: Artificial intelligence recognizes a person's heart rate based on slight discoloration of their face. The AI detects these differences in the individual parts of the face. If the heart rates do not match in the different facial regions, the deepfake is exposed. But: Deepfake detection is becoming more and more difficult as the algorithms of manipulated videos are becoming more and more advanced.
Currently, deepfakes mainly show up in two areas: porn (for example faked videos of well-known celebrities) and entertainment like lip sync videos, which are particularly popular on social media.
With a little technical know-how and the appropriate hardware, anyone can, at least in theory, create simple deepfakes. The quality of the output, however, is crucial. Dominik Kovacs: “Most video-based deepfakes that involve changes to the face and lip synchronization are not possible for everyone. Fortunately, there is not that much open source software on the market at the moment."
The risks of deepfakes must not be ignored. Currently, deepfakes are mainly used in the field of pornography. Thus they can attack the integrity and reputation of private individuals.
In the future, manipulated videos of politicians and decision-makers that could directly influence (political) opinion could appear online and become a threat to democracy.
For example, a faked pornographic video of a politician is published and covered by the media. The politician loses votes and as a result the election. The temptation to vilify political opponents through manipulated videos could be great. And the confusion among the population as well.
Deepfakes have many positive sides and offer a lot of possibilities. Soon, our Hollywood blockbusters could be created on a computer from the home. Deepfakes can also be of great use in grief work. Deceased family members can be brought back to life with a little help of deepfake software and continue to be a comforting aspect in the processing of losses.
