Commentary

Commentary: Facial recognition disproportionately harms people of color, women

March 1, 2022 5:53 am
A photo illustration of a white man's face being scanned

An MIT research team found that the technology worked relatively well when analyzing the faces of white men, but failed over one in three times when classifying the faces of Black women. (Getty Images)

Facial recognition technology gives our government, companies, and individuals unprecedented power to track and log our every move – where we worship, seek therapy, visit a health care provider, protest, meet friends, and more. That is, when it works as advertised. 

While facial recognition technology is, at its core, a tool that allows governments to create an all-watching surveillance state, in practice the technology has a shocking error rate that misidentifies Black and brown people and women at staggering rates. These errors have real-world consequences. 

The technology’s intrusions on privacy and failures in accuracy make it clear that New Hampshire should ban facial recognition – and fortunately, our legislators have an opportunity to do just that. 

House Bill 1447, being debated in the N.H. House Transportation Committee on Tuesday, would prohibit state agencies in New Hampshire from using this dangerous technology. We urge legislators and Gov. Chris Sununu to support this legislation. 

Still not sure? Let’s start with its error rate. 

Back in 2018, an MIT research team found that the technology worked relatively well when analyzing the faces of white men, but failed over one in three times when classifying the faces of Black women. Since then, study after study has come to similar conclusions, including one from the National Institute of Standards and Technology, which found that the majority of algorithms falsely matched Black women between 10 and 100 times more often than white men, and falsely matched white women between two and 10 times more often than white men. 

This error rate even extended to members of Congress. In an ACLU test of the “Rekognition” facial recognition tool, 28 members of Congress were falsely matched with a mug-shot database. And, as we have seen in other studies, the false matches disproportionately affected lawmakers of color. While lawmakers of color represented 20 percent of the total members of Congress, they represented 39 percent of the false matches.

These false matches can cause real world harm. For example, Robert Williams, a Black man, was arrested by Detroit Police in front of his wife and two young daughters on his front lawn and spent 30 hours in a jail cell because an algorithm made a false match. Though police later acknowledged the mistake, the harm could not be undone. 

Thanks to media reports, we know that at least two other Black men have been wrongly arrested because of a false match. But, the true extent of these harms remains unknown because of a near total lack of government transparency around this technology. 

And governments are not stopping with law enforcement. We are now seeing his technology, with its well documented flaws, deployed in schools and public housing developments.

Now, take a moment and imagine that this technology worked as directed: even then, it is still a civil liberties nightmare. Combining this technology with existing surveillance camera networks enables the government to track and store the totality of our public movements with the push of a button. 

This is not a theoretical concern: China is deploying this technology to track the movements of its residents, including how often and where people pray and their social and professional contacts. 

Such mass and total surveillance has no business in New Hampshire. 

This ban would build on an important first step our state took in 2016 when it prohibited using facial recognition technology to search footage collected from police body cameras. 

We urge lawmakers to support HB 1447. It is past time to rein in this profound threat to Granite Staters.

Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.