Sorry, Russia: Can This Pentagon Software Spot Fake News?

Michael Peck

By Michael Peck

With the 2020 U.S. presidential elections coming up, it is a certainty that Russia will use Internet trolls and fake news to influence the campaign. Indeed, there are reports that Russian trolls are already at work disparaging certain candidates.

But there are other reasons why the Pentagon should be concerned with fake news. False information, or doctored video or photos, can inflame foreign nations and put American soldiers at risk. But with a torrent of information and images, social media sites such as Facebook and Twitter face a herculean task in sifting false from true.

So, DARPA – the Pentagon’s cutting-edge research agency – wants to develop automated software that can spot trolls and fake news.

The Semantic Forensics (SemaFor) program “will develop technologies to automatically detect, attribute, and characterize falsified, multi-modal media assets (e.g., text, audio, image, video) to defend against large-scale, automated disinformation attacks,” according to the DARPA research announcement.

Statistical detection techniques to spot anomalies, or searching for digital fingerprints to detect fake news, is vulnerable to deception. Instead, DARPA wants to home in on inconsistencies that indicate fakes such as images that have been digitally distorted to change a person’s face.

“For example, GAN-generated faces may have semantic inconsistencies such as mismatched earrings,” DARPA explains. “These semantic failures provide an opportunity for defenders to gain an asymmetric advantage. A comprehensive suite of semantic inconsistency detectors would dramatically increase the burden on media falsifiers, requiring the creators of falsified media to get every semantic detail correct, while defenders only need to find one, or a very few, inconsistencies.”

The goal isn’t just to identify fake media, but also who did the faking. “Semantic detection algorithms will determine if multi-modal media assets have been generated or manipulated. Attribution algorithms will infer if multi-modal media originates from a particular organization or individual. Characterization algorithms will reason about whether multi-modal media was generated or manipulated for malicious purposes.”

Read the original article.