Facebook is watching out for 'deepfake' videos as election approaches
Posted January 13, 2020 4:40 p.m. EST
Updated January 13, 2020 5:57 p.m. EST
Raleigh, N.C. — Misinformation designed to sway opinions and votes became a trend on social media during the 2016 presidential election season. As we approach 2020, the technology used to fool you is getting more sophisticated, and social media outlets are starting to take action to protect reality.
Facebook banned the videos called "deepfakes" last week. Deepfakes are videos made using artificial intelligence. The computer studies lip movements and eventually can create a video that makes it appear like someone is saying something they never said.
"For particularly maliciously motivated and well-resourced units, it's actually easy," said Arnav Jhala, professor of computer science at NC State University.
Most deepfakes are made using an app called FakeApp. Hobbyists have been using the app for the last few years, mostly to create pornographic videos. The app can also be used to alter what public figures are saying. That's what is worrying both political parties and major social media outlets.
"Maybe you will have technology that you will be able to detect it, but the amount of time it takes to detect these is important for something like an election," said Jhala.
Frances McBride is a sophomore at NC State, and she's getting ready to vote in her first presidential election. She has been reading about deepfakes and worries about what technology could do to opinions of voters.
"If something like that got out right before an election, like the night before, that could influence a lot of people's leanings," McBride said.
Facebook will use an algorithm to try to weed out deepfakes and delete them from the site. That could lead to some legitimate videos being deleted by accident.
"Anytime you have an intelligent algorithm that is making decisions about whether a video is fake or not, it's doing it with a certain amount of confidence, and these videos are very good, so it's very hard to detect them," said Jhala.
Technology experts say it's best for people to know how to tell if something is a deepfake on their own. Deepfakes can be hard to detect, but they usually have lighting inconsistencies – especially around the mouth – and they can often have a flicker.
"There are signs, but it has to be looked at carefully," Jhala said.
Facebook's ban only includes videos that are made using artificial intelligence and will not filter out videos that have been altered by human doctoring such as editing to change words around.