Election time misinformation is nothing new. A couple of years ago, YouTube was rife with fake news videos alleging the commission of terrible deeds by candidates, their friends, families and associates.
What is new are deepfakes, video forgeries created with machine learning tools that are increasingly convincing at showing their subjects saying things they never did. One U.S. Senator, Florida’s Marco Rubio, has already warned that its only a matter of time before deepfake election videos surface.
The satirical and profane BuzzFeed video voiced by Jordan Peele, apparently depicts Barack Obama uttering controversial statements is an example. While showing some telltales of a deepfake when viewed in its original large size and high resolution format (watch the mouth closely), it demonstrates how credible is the prospect that actual misinformation videos could be created.
Current technology still creates videos with recognizable flaws. According to Siwei Lyu, an associate professor of computer science and director of the Computer Vision and Machine Learning Lab at the State University of New York in Albany:
“When a deepfake algorithm is trained on face images of a person, it’s dependent on the photos that are available on the internet … but photographers don’t usually publish images where the main subjects’ eyes are shut.”
Lyu, his colleague, Ming-Ching Chang and their Ph.D. student Yuezun Li have developed a way to exploit this weakness, developing their own algorithm to detect deepfakes by looking for unnatural blinking patterns.
As Lyu points out, deepfake technology is improving steadily. Therefore, detection of these fakes will lag behind. Perhaps in the future detecting fakes will be as quick and easy as a google search. But until that there are a few ways we can debunk even the most convincing deepfake.
- Rely on trustworthy sources for news; if you see a video on youtube depicting a candidate doing or saying something extremely controversial, consider the source.
- Additionally, many of the ways to recognise fake news can help you spot fake videos too.
Until deepfakes can be identified quickly, accurately and automatically, it behooves us all as voters to be careful not to believe everything we seem to see.
For information on voting on and around the UW-Madison, see the Voter Information website.