Washington Post: Lawmakers warn about threat of political deepfakes by creating one
Washington, September 27, 2019
Lawmakers warn about threat of political deepfakes by creating one
BY CAT ZAKRZEWSKI - 09/27/2019
Rep. Michael Waltz called for Navy to beat Army in this year’s football game. That's essentially treason for a former Army Green Beret.
And it's not something he'd ever say in real life: The statement came from a newly released political deepfake -- a video doctored with artificial intelligence.
Waltz teamed up with Rep. Don Beyer (D-Va.) to craft a mock deepfake for the House Science subcommittee to illustrate just how realistic this kind of disinformation can be. The SUNY-Albany and University of Chicago researchers took a recorded video statement from Beyer and transposed it onto Waltz's image -- designed to be a jarring sight for subcommittee chair and former Navy pilot Mikie Sherill (D-N.J).
The resulting video is a warning for lawmakers -- and the public -- that bad actors could abuse this technology for much more nefarious purposes than having a friendly joke about a sports rivalry.
“You see how dangerous and misleading it could be; I’m sure we fooled a couple of people,” Beyer said. “For instance, what if instead of ‘Go Navy, Beat Army,’ I said, 'It’s time to impeach the president'? That would be viral everywhere.”
“My friends might appreciate that, but I think he would not,” he added of his Republican colleague.
As the 2020 election looms, Washington lawmakers are increasingly concerned that bad actors will use deepfake technology to sow chaos and stoke divisions among the American public — much like Russian actors did with traditional social media posts during the 2016 election.
Waltz and Beyer are warning that United States needs to be investing in technology to quickly detect such fakes to keep up in an arms race, as the tools to create misleading videos become even cheaper and more widely available.
“These videos and this technology have the potential to truly be a weapon for our adversaries,” Waltz said.
Expert witnesses who specialize in computer science and disinformation gave the lawmakers sobering testimony in a hearing yesterday about the state of the country's preparedness to address deepfakes and other hoaxes.
Siwei Lyu, who led the SUNY researchers in developing the mock deepfake, said he was able to train his software in eight hours using a minute-long video they found of Waltz on YouTube. Though Lyu doesn’t widely share the software tools he used to make the video, he told lawmakers that right now similar technology is widely available online.
"The technical capability of making high quality deepfakes is already at the disposal of whoever wants to make it," Lyu said.
Disinformation isn't a new problem, but it's one that is being exacerbated by social media, which can help it spread much more quickly, the witnesses said.
Hany Farid, a professor at the Univeristy of California, Berkeley, warned lawmakers that the major technology platforms like Facebook and Google need to play a role in addressing deepfakes, but the tech companies have been slow to address the problem.
"You have to understand here that we are fighting against business interests," Farid said. "In the last six months, the language coming out of the technology sector is encouraging, but I don't know there's a lot of action yet."
Technology companies are increasingly partnering with academics to address deepfakes and build better technology to detect such videos on their platforms. But the companies haven't yet publicly released policies explaining how they will address deepfakes spotted on their platforms.
Farid said the recent video of House Speaker Nancy Pelosi that was altered to appear as if she was slurring her words underscores the impending challenges. Facebook allowed the video to remain on its platform, and the company defended that decision saying it did not want to be responsible for separating reality from fiction online.
"I can help with the technology problem, but I don't know what I can do with the policy problem when you say you aren't arbiters are the truth," Farid said. "They have to start getting serious about how their platforms are being weaponized to great effect and disrupting elections, inciting violence and sowing civil unrest."
Farid said it's difficult to predict exactly when a convincing deepfake will be released to disrupt a U.S. election.
"I think it's coming, but I don't know whether it will be in 2020, 2022 or 2024," Farid said. "Largely because the cheap stuff still works. I think we'll eventually get ahead of that and then this will be the next front."