Last week, when hackers exploited a bug in Parler to download all of the content from the right-wing social media platform, they were surprised to find that many of the images and videos contained geolocation metadata showing exactly how many people were using the site on the Site had participated in the invasion of the US Capitol building days earlier. However, the videos uploaded to Parler also contain an equally sensitive plethora of data that is within sight: thousands of images of exposed faces, many of whom participated in the Capitol riot. Now a website has cataloged and published each of these faces into a single, easy-to-search list.
Late last week, a website called Faces of the Riot came online that only showed a huge grid of more than 6,000 images of faces, each with just a series of characters linked to the Parler video in which It appeared. The site creator tells WIRED that they used simple, open source machine learning and facial recognition software to recognize, extract, and deduplicate each face from the 827 videos that came from inside and outside on Jan. 6 of the Capitol Building were sent to Parler when radicalized Trump supporters stormed the building in a riot that killed five people. The creator of Faces of the Riot says his goal is to allow anyone to easily sort through the faces drawn from these videos to identify who they know, or who took part in the mob, or even Use the collected faces against the FBI to reference posters and send a tip to law enforcement when they discover someone.
"Anyone involved in this violence, which really amounts to a riot, should be held accountable," says the website's creator, who has asked for anonymity to avoid retaliation. "It is entirely possible that many of the people who have been to this website now have real consequences for their actions."
Aside from clear privacy concerns, Faces of the Riot's indiscriminate face disclosure makes no distinction between lawbreakers who trampled barriers, broken into the Capitol and entered legislative chambers, and those who merely participated in the outdoor protests have participated. A recent upgrade to the website adds hyperlinks from faces to the video source so visitors can click on each face and see what the person filmed on Parler. The Faces of the Riot, who says he is a college student in the "Greater DC" area, intends to use this additional feature to contextualize the inclusion of each face on the site and differentiate between bystanders, peaceful protesters, and violent insurgents.
He admits that he and a co-creator are still working to scrub "non-rioters" faces, including those of the police and press who were present. A message at the top of the website also warns of vigilante investigations and instead suggests that users report those they detect to the FBI with a link to an FBI advice page. "When you go on the website and see someone you know, you can learn about a relative," he says. "Or you could say, 'Oh, I know this person' and then forward that information to the authorities."
Looking for faces
Despite its disclaimers and limitations, Faces of the Riot poses the serious privacy threats posed by ubiquitous facial recognition technology, says Evan Greer, nonprofit civil liberties campaign leader for Fight for the Future. "Whether by an individual or by the government, this technology has profound implications for human rights and freedom of expression," said Greer, whose organization campaigned for a legal ban on facial recognition technology. "I think it would be a huge mistake if we got out of this moment by glorifying or lionizing a technology that by and large harms color communities, low income communities, immigrant communities, Muslim communities, activists … disproportionately harm the same people, whose faces on this website stormed the Capitol to silence them and to withdraw their rights. "
The developer of the website counters that Faces of the Riot does not rely on face recognition, but on face recognition. While using the open source machine learning tool TensorFlow and face recognition software Dlib to analyze the Parler videos, he only used that software to recognize and "group" faces from the eleven hours of video of the Capitol riot . Dlib allowed him to deduplicate the 200,000 images of faces extracted from video images to around 6,000 unique faces. (He admits that there are still some duplicates and pictures of faces on protest signs. In some cases, the number "45" has been identified as a human face on some signs.)
He also stresses that there is no search tool on the website and no attempt is made to associate faces with names or other identifying details. There is also no function to upload an image and match it with images in the site's collection, which could lead to dangerous misidentifications. "It is very difficult to allow a user to take a photo of a poster they are looking for and search for it," says the site creator. "That will never happen."
The roughly 42 gigabyte Parler videos that Faces of the Riot analyzed were downloaded before Amazon decided early last week to cut off Parler's web hosting, which has kept the site largely offline since then. Hacktivists exploited a vulnerability in Parler that allowed them to download and archive any post from the service that bills itself as an uncensored "free speech" alternative to Twitter or Facebook. Faces of the Riot received Parler's recovered videos after they were made available online by Kyle McDonald, a media artist who received them from a third party he refused to identify.
"Play it safe"
The creator of the Faces of the Riot site initially saw the data as an opportunity 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 viable idea and I can do something to help people,'" he says. Faces of the Riot is the first website he ever created.
McDonald has previously criticized both the power of facial recognition technology and implemented facial recognition projects such as ICEspy, a tool it introduced in 2018 to identify immigration and customs agents. He tells WIRED that he also analyzed the leaked Parler videos using facial recognition tools to see if he could identify people, but he can only identify two, both of which have already been named by the media. He sees Faces of the Riot as "playing it safe" when compared to his own face recognition experiments, as it's not about associating faces with named identities. "And I think it's a good call because I don't think we need to legitimize this technology any more than it already is and has been falsely legitimized," says McDonald.
But McDonald also points out that Faces of the Riot shows how accessible facial recognition technology has become. "It shows how this tool, which was restricted only to the people who have the most education, the greatest power and the greatest privilege, is now in this more democratic state," says McDonald.
The creator of the Faces of the Riot site sees it as more than an art project or a demonstration. Despite the safeguards he has put in place to limit the ability to automatically identify people, he continues to hope that the effort will produce real, tangible results – if only indirectly through reports to law enforcement agencies. "It just felt like people had gotten away with a lot of bad things in the last four years," he says. "This is an opportunity to end this."
This story originally appeared on wired.com.