Stefan Mileschin

AI forensics tools

The first forensics tools for catching fakes created with AI have been developed through a programme run by the US Defense Department.

The most common technique for generating fake videos involves using machine learning to swap one person’s face onto another's. The resulting videos, known as “deepfakes”, are simple to make, and can be surprisingly realistic particularly if you have taken large amounts of drugs or are especially stupid.

However after tweaking by a skilled video editor, deep fakes can be made to seem more real which makes the tech a UK defence threat. For example a conspiracy nut could release improbable fake footage showing that Barak Obama killed JFK or a black person beating himself to death in the cells.

Tools for catching deepfakes were developed through a programme—run by the US Defense Advanced Research Projects Agency (DARPA)—called Media Forensics specialising in software automated existing forensics tools, but has recently turned its attention to AI-made forgery.

Apparently, there are subtle cues in current GAN-manipulated images and videos that allow it to detect the presence of alterations.

Siwei Lyu, a professor at the University of Albany, State University of New York realised that the faces made using deep fakes rarely, if ever, blink. And when they do blink, the eye movement is unnatural. This is because deepfakes are trained on still images, which tend to show a person with his or her eyes open.

Others involved in the DARPA challenge are exploring similar tricks for automatically catching deepfakes: strange head movements, odd eye colour.

Lyu says a skilled forger could get around his eye blinking tool simply by collecting images that show a person blinking. But he adds that his team has developed an even more effective technique, but says he’s keeping it secret for the moment.

“I’d rather hold off at least for a little bit. We have a little advantage over the forgers right now, and we want to keep that advantage.”

https://fudzilla.com/news/ai/46908-d...-to-spot-fakes