On the heels of Aprils Fools Day, and a slew of newly popularized AI-generated images like Trump’s arrest and Pope-in-a-coat, we thought we’d dig into some ways to stay vigilant in easily spotting some of these deceptive images.
HOW TO DETECT A DEEPFAKE:
Oddities. Something just seems out of place in the picture or video. It can be little things like mismatched shoes or earrings, or easier-to-spot irregularities like a misshapen jaw or extra legs. AI seems to have a particularly difficult time generating hands and fingers accurately, so it’s always a good idea to look there first.
Eye movement. Deepfake videos often portray people with weird blinking patterns, or making unusual eye movements. Like fingers, AI still struggles to make these actions look natural.
Audio/video quality. Many deepfake videos have an out-of-sync audio and video match, so pay attention to how the two align. It could be a dead giveaway that it’s a phony.
Check your sources. As deepfakes get more convincing, one of the best ways to spot if they’re fake is to check their sourcing (though admittedly social media makes it difficult as anyone can share videos and pictures). For example, the original source of a video of President Biden would either come from the White House or an official government agency.
The devil’s in the details. AI still struggles with many of the nuances of the human body like exactly how fabric drapes on a person, button/zipper placement, and facial symmetry. So, if you think an image is real, give a closer look at the details.
Sharpen your skills. There are plenty of websites dedicated to helping the general population better hone their deepfake spotting skills. For example, MIT’s Detect Fakes is a short quiz that allows users to compare two videos to decide which is real, and Microsoft’s Spot the Deepfake is a short quiz that has users detect signs like eye movement and emotional reaction to determine deepfakes.