Although the deep fake videos in which celebrity statements are manipulated have existed for several years, these artificial intelligences that create them have developed to such an extent that it is currently difficult to know whether they are false or not. But in order to combat these systems that concern both politicians and celebrities, the best option is another artificial intelligence.
A group of researchers from two California universities has developed a tool that analyzes how people speak and gesticulate and recognizes whether the video is real or manipulated. The Protecting World Leaders Against Deep Fakes project has been funded by Google and DARPA, the research section of the Pentagon.
What this system recognizes is the weak biometric signature, a group of gestures we make when we speak that identify each person, although it is not as specific as the fingerprint or the iris of the eyes. The study has had its own deep fake videos made by themselves with the GAN system, one of the most used in this type of tools, and have used as an example people like Donald Trump, Barack Obama or Hillary Clinton, among others. The program has detected the manipulation with an accuracy of up to 92%, as there are characteristic features that are not yet well achieved, such as the smile of the current president of the United States, for example.
“Celebrities and political figures have been the main targets so far, but I wouldn’t be surprised if in a year or two anyone could be the target of this deception,” says Hao Li, a researcher on the project’s development team.