Facebook AI software to unveil origins of deepfake images

Facebook AI software to unveil origins of deepfake images

Social media giant ‘Facebook’ scientists on Wednesday stated that they developed artificial intelligence (AI) software not only for the identification purpose of “deepfake” images but to figure out its origin as well.

Photos, videos, or audio clips which are altered using artificial intelligence to look authentic are termed as ‘deepfakes’, which according to experts warning, can mislead or be false.

As per Facebook research scientists Tal Hassner and Xi Yin’s statement, their team worked mutually with Michigan State University to develop software that reverse engineers deepfake images to explore how they were created and where they originated.

“Our method will provide deepfake detection facility as well as tracing in real-world settings, where the deepfake image itself is mostly the only information detectors have,” stated the scientists in a blog post.

“This work will not only allow researchers but practitioners tools also for better investigation of incidents of coordinated disinformation using deepfakes, and expose new directions for future research,” they elaborated.

Facebook’s new AI software runs deepfakes with the help of a network to find imperfections left during the manufacturing process, which according to the scientists, alter an image’s digital “fingerprint.”

“In digital photography, by fingerprints, the digital camera is identified used to produce an image,” stated the scientists. 

“Just like device fingerprints, image fingerprints are unique patterns left on images which can equally identify the generative model that the image originated from.”

Last year, Microsoft disclosed software that can help in spotting deepfake photos or videos, in addition to the collection of programs designed to fight the hard-to-detect images before the US presidential election.

Read more: Facebook Working On A Smartwatch To Connect With AR Glasses

Its Video Authenticator software analyses an image or each frame of a video, finding evidence of manipulation not possible with the naked eye.