Faces of the Riot used open source software to detect, extract, and deduplicate every face from the 827 videos taken from the insurrection on January 6.
WHEN HACKERS EXPLOITED a bug in Parler to download all of the right-wing social media platform’s contents last week, they were surprised to find that many of the pictures and videos contained geolocation metadata revealing exactly how many of the site’s users had taken part in the invasion of the US Capitol building just days before. But the videos uploaded to Parler also contain an equally sensitive bounty of data sitting in plain sight: thousands of images of unmasked faces, many of whom participated in the Capitol riot. Now one website has done the work of cataloging and publishing every one of those faces in a single, easy-to-browse lineup.
Late last week, a website called Faces of the Riot appeared online, showing nothing but a vast grid of more than 6,000 images of faces, each one tagged only with a string of characters associated with the Parler video in which it appeared. The site’s creator tells WIRED that he used simple open source machine learning and facial recognition software to detect, extract, and deduplicate every face from the 827 videos that were posted to Parler from inside and outside the Capitol building on January 6, the day when radicalized Trump supporters stormed the building in a riot that resulted in five people’s deaths. The creator of Faces of the Riot says his goal is to allow anyone to easily sort through the faces pulled from those videos to identify someone they may know or recognize who took part in the mob, or even to reference the collected faces against FBI wanted posters and send a tip to law enforcement if they spot someone.
“Everybody who is participating in this violence, what really amounts to an insurrection, should be held accountable,” says the site’s creator, who asked for anonymity to avoid retaliation. “It’s entirely possible that a lot of people who were on this website now will face real-life consequences for their actions.”
Aside from the clear privacy concerns it raises, Faces of the Riot’s indiscriminate posting of faces doesn’t distinguish between lawbreakers—who trampled barriers, broke into the Capitol building, and trespassed in legislative chambers—and people who merely attended the protests outside. An upgrade to the site today adds hyperlinks from faces to the video source, so that visitors can click on any face and see what the person was filmed doing on Parler. The Faces of the Riot creator, who says he’s a college student in the “greater DC area,” intends that added feature to help contextualize every face’s inclusion on the site and differentiate between bystanders, peaceful protesters, and violent insurrectionists.
He concedes that he and a cocreator are still working to scrub “non-rioter” faces, including those of police and press who were present. A message at the top of the site also warns against vigilante investigations, instead suggesting users report those they recognize to the FBI, with a link to an FBI tip page. “If you go on the website and you see someone you know, you might learn something about a relative,” he says. “Or you might be like, oh, I know this person, and then further that information to the authorities.”
“Everybody who is participating in this violence, what really amounts to an insurrection, should be held accountable.”
FACES OF THE RIOT CREATOR
Despite its disclaimers and limitations, Faces of the Riot represents the serious privacy dangers of pervasive facial recognition technology, says Evan Greer, the campaign director for digital civil liberties nonprofit Fight for the Future. “Whether it’s used by an individual or by the government, this technology has profound implications for human rights and freedom of expression,” says Greer, whose organization has fought for a legislative ban on facial recognition technologies. “I think it would be an enormous mistake if we come out of this moment by glorifying or lionizing a technology that, broadly speaking, disproportionately harms communities of color, low-income communities, immigrant communities, Muslim communities, activists … the very same people that the faces on this website stormed the Capitol for the purpose of silencing and disenfranchising.”
The site’s developer counters that Faces of the Riot leans not on facial recognition but facial detection. While he did use the open source machine learning tool Tensor Flow and the facial recognition software Dlib to analyze the Parler videos, he says he used that software only to detect and “cluster” faces from the 11 hours of video of the Capitol riot; Dlib allowed him to deduplicate the 200,000 images of faces extracted from video frames to around 6,000 unique faces. (He concedes that there are nonetheless some duplicates and images of faces on protest signs included too. Even the number “45” on some signs was in some cases identified as a human face.)
He emphasizes also that there’s no search tool on the site, and it doesn’t attempt to link faces with names or other identifying details. Nor is there any feature for uploading an image and matching it with images in the site’s collection, which he says could lead to dangerous misidentifications. “There’s a very hard no on allowing a user to take a photo from a wanted poster and search for it,” the site’s creator says. “That’s never going to happen.”
The roughly 42 gigabytes of Parler videos that Faces of the Riot analyzed were downloaded prior to Amazon’s decision early last week to cut off Parler’s web hosting, leaving the site largely offline since. Racing against that takedown, hacktivists took advantage of a security flaw in Parler that allowed them to download and archive every post from the service, which bills itself as an uncensored “free speech” alternative to Twitter or Facebook. Faces of the Riot obtained Parler’s salvaged videos after they were made available online by Kyle McDonald, a media artist who obtained them from a third party he declined to identify.
The Faces of the Riot site’s creator initially saw the data as a chance to experiment with machine learning tools, but quickly saw the potential for a more public project. “After about 10 minutes I thought, this is actually a workable idea and I can do something that will help people,” he says. Faces of the Riot is the first website he’s ever created.
McDonald has previously both criticized the power of facial recognition technology and himself implemented facial recognition projects like ICEspy, a tool he launched in 2018 for identifying agents of the Immigration and Customs Enforcement agency. He tells WIRED he also analyzed the leaked Parler videos with facial recognition tools to see if he could identify individuals, but could only ID two, both of whom had already been named by media. He sees Faces of the Riot as “playing it really safe” compared even to his own facial recognition experiments, given that it doesn’t seek to link faces with named identities. “And I think it’s a good call because I don’t think that we need to legitimize this technology any more than it already is and has been falsely legitimized,” McDonald says.
But McDonald also points out that Faces of the Riot demonstrates just how accessible facial recognition technologies have become. “It shows how this tool that has been restricted only to people who have the most education, the most power, the most privilege is now in this more democratized state,” McDonald says.
The Faces of the Riot site’s creator sees it as more than an art project or demonstration. Despite the safeguards he put in place to limit its ability to automatically identify people, he still hopes that the effort will have real, tangible results—if only indirectly through reports to law enforcement. “It’s just felt like people got away with a lot of bad stuff for the last four years,” he says. “This is an opportunity to start trying to put that to an end.”