EAST LANSING, Mich. (WLNS) – Michigan State University is teaming up with Facebook to develop a new method to detect deepfakes.
Those are photos or videos edited with advanced software to change the faces of the people in them. While some people make deepfakes that are fun or amusing, people have used this technology to put the faces of famous people into compromising situations.
Deepfakes range in quality depending on the software being used. Some are easy to spot with the naked eye while others are nearly impossible to distinguish even for computers.
Artificial intelligence experts from MSU and Facebook are working on a novel approach to fighting deepfakes that aims to reverse-engineer the software used to create them. This will make it easier to spot any deepfakes that were created with that software in the future.
The team tested their model parsing approach with 100,000 synthetic images made from 100 different versions of publicly-available software. They found their system is substantially better than previous detection models.