Your face is no longer just your face—it’s been augmented. At a football game, your face is currency, used to buy food at the stadium. At the mall, it is a ledger, used to alert salespeople to your past purchases, both online and offline, and shopping preferences. At a protest, it is your arrest history. At the morgue, it is how authorities will identify your body.
Facial-recognition technology stands to transform social life, tracking our every move for companies, law enforcement, and anyone else with the right tools. Lawmakers are weighing the risks versus rewards, with a recent wave of proposed regulation in Washington State, Massachusetts, Oakland, and the U.S. legislature. In May, Republicans and Democrats in the House Committee on Oversight and Reform heard hours of testimony about how unregulated facial recognition already tracks protesters, impacts the criminal-justice system, and exacerbates racial biases. Surprisingly, they agreed to work together to regulate it.
“You’ve hit the sweet spot that brings progressives and conservatives together,” Republican Mark Meadows of North Carolina said at the hearing. “The time is now, before it gets out of control.”
Agreement is shocking in a political moment defined by polarization, but lightning has seemingly struck twice. Microsoft and Amazon, the makers of Face API and Rekognition software, respectively, also both endorse federal regulation. In June, Axon, the number-one body-camera manufacturer in the United States, agreed with its ethics board’s proposal not to outfit Axon cameras with facial recognition (at least for the foreseeable future). The Microsoft president Brad Smith called for governments “to start adopting laws to regulate this technology” last year, while the Amazon Web Services CEO Andy Jassy echoed those comments in June, likening the technology to a knife. It’s a less dramatic image than the plutonium and nuclear-waste metaphors critics employ, but his message—coming from an executive at one of the world’s most powerful facial-recognition technology outfits—is clear: This stuff is dangerous.
But crucially, Jassy and Smith seem to argue, it’s also inevitable. In calling for regulation, Microsoft and Amazon have pulled a neat trick: Instead of making the debate about whether facial recognition should be widely adopted, they’ve made it about how such adoption would work.
In a statement to The Atlantic, Amazon said it’s working with researchers, lawmakers, and its customers “to understand how to best balance the benefits of facial recognition with the potential risks,” noting that Rekognition has a variety of uses, including fighting human trafficking and finding missing persons. Microsoft reiterated statements from Smith in support of facial-recognition regulation, including implementing safeguards against reported misuse and securing consent from anyone who is the target of the technology.
But some privacy experts believe the companies have ulterior motives. Evan Selinger, a philosophy professor at the Rochester Institute of Technology, accuses Microsoft and Amazon of trying to “suck out the motivation” for robust regulation. He argues that companies are pushing for regulation at the federal level because national laws are typically written to be floors, not ceilings—baseline measures that are less likely than local legislation to include restrictions for how private companies use the tech.
“Federal rules don’t prohibit local and strong regulations,” Selinger says. “But you get to raise the flag of mission accomplished [and] make it seem like people who want something stronger than what’s been enacted at a federal level are extremists.” Turning bipartisan agreement itself—not robust legislation—into the goal can, as Seligner puts it, “take the wind out of the sails for local [regulation] so people won’t feel as motivated, because they think there’s been change.”
Without regulation, the potential for misuse of facial-recognition technology is high, particularly for people of color. In 2016 the MIT researcher Joy Buolamwini published research showing that tech performs better on lighter-skinned men than on darker-skinned men, and performs worst on darker-skinned women. When the ACLU matched Congress members against a criminal database, Amazon’s Rekognition software misidentified black Congress members more often than white ones, despite there being far fewer black members.
This includes House Chairman Elijah Cummings, a Baltimore native whose face was also scanned when he attended a 2015 rally in memory of Freddie Gray, the unarmed black teenager who died of a spinal-cord injury while in police custody. The Baltimore Police Department used facial recognition to identify protesters and target any with outstanding warrants. Most of the protesters were black, meaning the software used on them might have been less accurate, increasing the likelihood of misidentification. Expert witnesses at the committee hearing in May warned of a chilling effect: Protesters, wary of being identified via facial recognition and matched against criminal databases, could choose to stay home rather than exercise their freedom of assembly.
Microsoft and Amazon both claim to have lessened the racial disparity in accuracy since the original MIT study and the ACLU’s report. But fine-tuning the technology to better recognize black faces is only part of the process: Perfectly accurate technology could still be used to support harmful policing, which affects people of color. The racial-accuracy problem is a distraction; how the technology is used matters, and that’s where policy could prevent abuse. And the solution Microsoft and Amazon propose would require auditing face recognition for racial and gender biases after they’re already in use—which might be too late.
In early May, The Washington Post reported that police were feeding forensic sketches to their facial-recognition software. A witness described a suspect to a sketch artist, then police uploaded the sketch to Amazon’s Rekognition, looking for hits, and eventually arrested someone. Experts at the congressional hearing in May were shocked that a sketch submitted to a database could credibly qualify as enough reasonable suspicion to arrest someone.
But Jassy, the Amazon Web Services CEO, claimed that Amazon has never received a report of police misuse. In May, Amazon shareholders voted down a proposal that would ban the sale of Rekognition to police, and halt sales to law enforcement and ICE. Jassy said that police should only rely on Rekognition results when the system is 99 percent confident in the accuracy of a match. This is a potentially critical safeguard against misidentification, but it’s just a suggestion: Amazon doesn’t require police to adhere to this threshold, or even ask. In January, Gizmodo quoted an Oregon sheriff’s official saying his department ignores thresholds completely. (“There has never been a single reported complaint from the public and no issues with the local constituency around their use of Rekognition,” a representative from Amazon said, in part, in a statement to Gizmodo.)
And as the discussions around the specifics of implementation swirl, critics argue that they are a diversion from larger, worthier discussions.
“We have seen a wave of technology companies pivoting towards public messaging suggesting support for more community-centric and civil-rights-protective regulations around technologies such as face surveillance, while at the same time working to undermine those very protections in actual legislation,” says Shankar Narayan, the Technology and Liberty Project director at the ACLU Washington.
So far, technology companies have succeeded in setting the terms of the debate over facial recognition. But these might be the last days of privately owning our own faces. Common ground itself is not a victory. Narrowing the discussion isn’t compromise; it’s a rhetorical trick. It turns public governance into a terms-of-service agreement: One party sets the terms while the other, uninterested and resigned to the inevitable, simply says, “I agree.”