Intel Launches New Deepfake Detector

This post is also available in: עברית (Hebrew)

As AI-based systems improve and deepfake tech keeps growing, the fear of it being used for malicious intents only increases. If at first the edited videos were easy to detect, as AI developments make deepfakes easier to produce, detecting them quickly has become all the more important.

Intel claims a solution that deals with blood flow in the face and is called “FakeCatcher”, and it is apparently 96% accurate.

Ilke Demir, research scientist at Intel Labs explains the idea and technology behind it, which is based on both a technique called Photoplethysmography (PPG) that detects changes in blood flow, as well as analyses eye movement to check for authenticity.

According to Demir, by analyzing these two traits Intel’s system can differentiate between a real video and a fake within seconds.

According to BBC News, the system has some flaws- the more pixelated a video, the harder it is to pick up blood flow. Furthermore, the system also does not analyze audio. The worry is that if the program says a video is fake when it’s genuine, it could cause real problems.

Demir responded to these claims by saying that the system alerts that a video might be fake, not that it is definitely fake, and that it is better to catch all the fakes while miss-labeling some real videos than miss crucial fakes.

In this regard, some experts in the field have questioned FaceCatcher’s ability to work “in the wild”. Matt Groh, a deepfakes expert, stated: “I don’t doubt the stats that they listed in their initial evaluation,” he says. “But what I do doubt is whether the stats are relevant to real-world contexts.”

Intel claims that FakeCatcher has gone through rigorous testing, including a “wild” test in which the system had a success rate of 91%.

However, according to BBC News, Matt Groh and other researchers want to see the system independently analyzed, since they do not trust the test Intel is setting for itself.

It is surprising how difficult it can be to tell a fake and a real video apart – and this technology certainly has potential, but it certainly has a long way to go.

This information was provided by BBC News.