Why are social platforms pressured to tackle deepfakes before US elections?

Question

Accepted Answer

Social Platforms Face Pressure to Combat Deepfakes Amidst US Election Concerns

Social media platforms are under significant pressure to address the proliferation of deepfakes as the United States presidential election approaches in November 2020. Deepfakes, which leverage artificial intelligence to generate highly realistic videos of individuals appearing to say or do things they never did, pose a substantial threat to the integrity of electoral processes and public discourse. The potential for such manipulated media to sow discord and misinformation was underscored by a "cheapfake" video of House Speaker Nancy Pelosi, which was intentionally slowed down to create the impression of slurred speech. Furthermore, the Democratic National Committee demonstrated the manipulative potential by creating a deepfake of its Chairman, Tom Perez, to deceive an audience into believing he was participating in a live Skype call. These instances highlight how deepfakes lower the barrier for malicious actors seeking to disseminate fabricated content.
The Growing Threat of Manipulated Media

The capacity for deepfakes to influence public perception and create political turmoil is a primary driver behind the pressure on social platforms. While a sophisticated, large-scale deepfake has yet to cause major political consequences in the US, simpler forms of manipulation, like the slowed-down Pelosi video, have already shown their disruptive potential. These doctored clips can be used to spread disinformation, damage reputations, and erode trust in authentic media. The ease with which such content can be produced and distributed means that adversaries can more readily create and deploy manipulated media to achieve their objectives, a concern articulated by DARPA's Media Forensics program leader, Matt Turek.
An Escalating Technological Arms Race

Combating deepfakes has become an escalating technological arms race. While some researchers are developing methods for authenticating media at the point of capture through digital watermarking, the rapid evolution of deepfake generation techniques presents a continuous challenge. As Siddharth Garg, an assistant professor of computer engineering at New York University's Tandon School, noted, the development of detection tools can inadvertently provide creators with new benchmarks to overcome. This dynamic is further exacerbated by the increasing accessibility of deepfake technology, with user-friendly applications and online tutorials making it easier for individuals with less technical expertise to create convincing manipulated content, as seen with the Zao app in China and various online services offering custom deepfakes.
Industry and Academic Collaboration on Detection

In response to these mounting concerns, social media companies and academic institutions are initiating collaborative efforts to enhance deepfake detection capabilities. Facebook, for instance, is investing US$10 million into a "Deepfake Detection Challenge," partnering with Microsoft, the Partnership on AI coalition, and university researchers. This initiative aims to stimulate research by creating a dataset of realistic deepfakes, produced by paid actors and set for release in December, to rigorously test detection tools. This challenge builds upon existing research funding, such as Facebook's US$7.5 million allocation to teams at UC Berkeley, the University of Maryland, and Cornell University, which are developing methods like "soft biometric models" to analyze subtle facial characteristics and detect inconsistencies in videos. The engagement of researchers from institutions including MIT and the University of Oxford underscores the broad academic participation in this critical area.

Reading history

Hide →

This history is cleared when the page is closed.

Why are social platforms pressured to tackle deepfakes before US elections?

Loading

Was this article helpful?

Helpful

Not helpful

Report an issue

1

0

0

Share with friends