Its use is just as unstoppable as it is disturbing. So much so that according to a study carried out by the consulting firm Entelgy, 74% of IT managers consider deepfakes as a threat that must be taken very seriously. In fact, according to Gartner, by 2023, 20% of attacks to try to steal and gain account control will use deepfakes as part of these malicious actions.
And it is that the use of this technique based on artificial intelligence, It can cause enormous damage to companies such as misinformation, involvement in crimes, damage to one’s own image, financial scams… and of course, impersonation of identities from around the world. So how to combat it?
Intel seems to have taken a first step with the creation of Fake Catcher. It is an AI model that, they say, detects in real time if a video is using deepfake and pretends to contain the image of a real person or not. Fake Catcher is capable of returning results in milliseconds with an accuracy rate of 96%.
Deepfake videos are everywhere now. You’ve probably already seen them; videos of celebrities doing or saying things they never actually did,” said Intel Labs research scientist Ilke Demir. In recent years, politicians and celebrities have been the main victims of these videos that use AI algorithms. to generate fake videos of people.
Unlike other deepfake detectors, which analyze raw video data to find those signals that reveal the authentication or not of said pieces, the Fake Catcher tool from Intel analyzes real videos for visual cues that indicate that the subject is real.
This involves subtle changes in the color of the pixels in a video, due to the blood flow of the heart that pumps blood around the body. Blood flow signals that are collected from the entire face and translated by algorithms into spatiotemporal maps, according to Intel.
The question is, Could anyone with enough knowledge create fake video to develop algorithms capable of fooling this Intel tool?
The technology that Intel has used in Fake Catcher
For its manufacture, Intel has made use of its own technologies in the development of Fake Catcherincluding the OpenVINO open source toolkit for optimizing deep learning models, and OpenCV for processing images and videos in real time.
The developer teams also used the open visual cloud platform to provide an integrated software stack for Intel Xeon Scalable processors. Additionally, Fake Catcher software can run up to 72 different detection streams simultaneously on 3rd generation Xeon Scalable processors.
