Artificial intelligence (AI) has advanced to a point where computers can be taught to create “deepfakes” by combining authentic images and videos. We’ve seen some troubling deepfakes already — videos that supposedly show well-known celebrities, that are in fact fake. The results of a deepfake video or photo can be harmless fun, like the image above imagining what John Krasinski’s Captain America would have looked like. But people with malicious intent might use deepfakes to disseminate misinformation, as the technology allows attackers to impersonate politicians and have them say practically anything.
These deepfakes could then spread like wildfire on social media and have the potential of doing plenty of harm before being debunked. That’s where Facebook might come to the rescue, as the company’s AI researchers have devised technology that does more than simply detect deepfakes. Facebook says its AI can actually track the origin of deepfakes by finding unique characteristics that might help it identify the source of a deepfake meant to spread misinformation.
Seeing Facebook at the forefront of fighting against deepfakes is actually refreshing, as deepfakes would largely spread via Facebook’s social apps, including Facebook, Messenger, WhatsApp, and Instagram.
The company partnered with Michigan State University to create technology that can detect whether a video or image is a deepfake or genuine.
Facebook uses its own AI to look for clues that indicate an image or video was doctored. The same technology can be used to identify commonalities between deepfakes and create a fingerprint that might identify the origin of the malicious deepfake:
Our reverse engineering method relies on uncovering the unique patterns behind the AI model used to generate a single deepfake image. We begin with image attribution and then work on discovering properties of the model that was used to generate the image. By generalizing image attribution to open-set recognition, we can infer more information about the generative model used to create a deepfake that goes beyond recognizing that it has not been seen before. And by tracing similarities among patterns of a collection of deepfakes, we could also tell whether a series of images originated from a single source. This ability to detect which deepfakes have been generated from the same AI model can be useful for uncovering instances of coordinated disinformation or other malicious attacks launched using deepfakes.
While attackers could attempt to hide their tracks when making deepfakes for malicious purposes, Facebook’s AI could still pick up fingerprinting elements that might allow investigators to determine the origin.
Facebook likens the algorithms to technology that would let someone recognize the components of a car based on how it sounds, even if they’ve never heard the car before:
To understand hyperparameters better, think of a generative model as a type of car and its hyperparameters as its various specific engine components. Different cars can look similar, but under the hood they can have very different engines with vastly different components. Our reverse engineering technique is somewhat like recognizing the components of a car based on how it sounds, even if this is a new car we’ve never heard of before.
Facebook is still perfecting its technology for detecting deepfakes, but says “this work will give researchers and practitioners tools to better investigate incidents of coordinated disinformation using deepfakes, as well as open up new directions for future research.” Facebook’s full blog on the matter is available at this link.