In a press release published on its site, Intel announced that it had developed a technology capable of revolutionizing cybersecurity and disinformation on the Internet. As part of its work on responsible artificial intelligence, the developers have created FakeCatcher, a technology capable of identifying in real time if a video contains deepfake. Capable of detecting this type of deceptive editing in a few milliseconds according to Intel, FakeCatcher would have an accuracy rate of 96%.
the deepfake is a multimedia editing technique using artificial intelligence. It consists in superimposing videos thanks to the analysis of movements of key points. The most common use of this technology is then to superimpose faces in order to create fake videos, often of personalities. The FaceApp application, which appeared in 2018, has democratized its use by Internet users, also multiplying fake videos and erroneous information. Director Jordan Peele created a video montage of Barack Obama in 2018 to warn of the dangers of this technology.
The company also revealed the technical details of its artificial intelligence. FakeCatcher is actually “an orchestra of specialized tools forming an optimized architecture”. Artificial intelligence analyzes “blocks of vision” in order to detect anomalies. FakeCatcher is also an artificial intelligence in deep learning, that is to say an artificial intelligence that learns continuously by analyzing different content. In this case, FakeCatcher analyzes videos without deepfake in order to determine which characteristics to detect. AI “looks for clues of authenticity in real life videos, assessing what makes us human” can we read in the press release.
Intel then takes the example of “blood flow” in a pixel of video. When our heart beats, the color of our veins changes with each beat. The blood flow signals are then analyzed on all the pixels of the video before being analyzed by algorithms. Thus, thanks to deep learning of FakeCatcher, the analysis will become more and more precise. FakeCatcher is also innovative in addition to its accuracy, as it is the first real-time artificial intelligence deepfake detector. The existing AIs require the video to be downloaded and sent for analysis, sometimes lasting several hours.