Intel FakeCatcher is capable of detecting DeepFakes in real time

Although the possibilities of Deep Learning and Artificial Intelligence systems are proving enormously useful in many sectors and tasks, there are certain controversial uses that could pose a problem in the future. One of those specific cases is that of “DeepFakes” which consist of the use of trained networks to be able to generate modified images or videos so that it appears that a different person is the one appearing in them.

On the Internet there are several cases of videos modified through these AI networks to simulate that politicians or famous people perform certain actions, ranging from parodies to pornographic videos, or even “fakes” where a politician can be made to say things that they really do not has said. As this technology advances, it is more difficult to distinguish when we are facing a real video or a DeepFake.

To try to avoid this situation, Intel has developed FakeCatcher, a system also based on Artificial Intelligence that is capable of detecting in milliseconds if a video is real or if it is a “DeepFake” type modification, with a 96% accuracy.

Detection is done in real time using a combination of the company’s own software and hardware, such as processors Intel Xeon Scalable next to instructions Intel AVX-512 running in parallel with 72 detection flows. Unlike other systemsIntel FakeCatcher searches videos and images for different elements that make a video real, giving as an example the variations that can be detected in the skin of humans when blood flow passes through the skin. These results are then compared with the suspicious videos to detect whether or not those elements appear in order to detect them as fake.

Undoubtedly, the AI ​​opens new unpleasant fronts that, curiously, can also be fought with AI, although everything indicates that these systems will evolve and we will see a “fight” of cat and mouse to perfect these systems in both directions, appearing Increasingly realistic DeepFakes and more advanced DeepFake detection systems.

End of Article. Tell us something in the Comments!