Here’s how to tell if you’re talking to a deepfake
As deepfake technology continues to improve, experts are increasingly concerned about the opportunities it presents for criminal behavior. From industrial espionage to scamming people out of their savings, there are all kinds of crimes that authorities fear could be committed by people impersonating someone else through deepfake videos.
So far, we’ve mostly seen deepfakes being used for things as harmless and amusing as incorporating Harrison Ford into the unforgettable 2018 prequel Solo: A Star Wars Story or Back to the Future starring Tom Holland and Robert Downey Jr to play. While this is fun, projects like this raise the possibility that deepfake technology could be used for more nefarious purposes.
However, a study published by metaphysic.ai has suggested a possible countermeasure. It suggests there’s a simple test that could give you away if you suspect you’re on a video call with a scammer.
The trick is very simple: simply ask the person you are speaking to to turn to the side. Metaphysic.ai ran a series of tests on tech guru Bob Doyle using deepfake technology to turn him into various famous people including Ryan Reyolds, Keanu Reeves, Tom Cruise and even Alexandra Daddario. Viewed from the front, all of the deepfakes looked pretty convincing. However, when Doyle was asked to turn 90° sideways to present a profile view, the forgery was immediately revealed.
Why did a profile view give away the game? According to research from metaphysic.ai (opens in new tab), there were two main reasons. First, the standard software that estimates facial poses in deepfake videos, known as the Facial Alignment Network, doesn’t do too well at acute angles. In fact, most 2D-based face alignment algorithms assign many more landmarks to the front of a face than to the side. Therefore, when a person turns to profile view and covers half of the front of their face, the algorithm has less working space.
Secondly, there are simply far fewer profile pictures to work with. Even celebrities like Tom Cruise, who has been photographed and filmed from every angle imaginable, are not usually seen in profile as often as they are from the front. It’s an unflattering, uncharacteristic angle that people don’t use that often — how many photos of you do you have in profile view? Probably not many goalkeepers. Photographers also don’t tend to shoot subjects in profile, so the algorithms used to create deepfakes will have trouble pulling reference footage from stock photos.
Ultimately, this means deepfake algorithms need to use less data when viewing a person in profile view and a smaller, more constrained canvas on which to use them. So the game is given away.
There’s another trick. The researchers also suggested that anyone who suspects a deepfake could ask the person to put a hand on their face. This will likely cause dull glitches revealing the fake at work.
Deepfake technology will likely catch up and find ways around this, but then we expected technological ways to detect them to emerge as well. For now, these seem like some good tricks to have in your back pocket in case you ever get that creepy feeling that the person you’re talking to on a Zoom call isn’t who they appear to be. In the meantime, don’t miss our roundup of deepfake examples that have spooked and amused the internet to learn more about the lighter side of deepfakes.
Continue reading: