Delta Air Lines
- Facial recognition technology has popped up everywhere.
- But it still has trouble identifying faces, particularly those of people of color.
- That means the tech can flag innocent users as being guilty of fraud or shoplifting, an expert said.
Facial recognition technology is everywhere, from your iPhone to candy vending machines to the TSA.
But it has a problem: It might not recognize you.
That's a problem that some gig workers for Walmart's Spark delivery service have faced, Business Insider reported last month. Since last fall, Spark has asked its drivers to periodically verify that they are who they say they are by taking three selfies through the app, which then compares the photos against the photo ID they have on file.
But some drivers say the facial recognition check kicked them off the app after the identity check — even though they were using an account in their name.
It turns out that those kinds of experiences aren't unusual in the broader world of facial recognition. Moreover, they're especially common if you're a person of color.
Facial recognition frequently misidentifies people of color
MIT Media Lab researcher Joy Buolamwini found that facial recognition technology misidentified black women up to 35% of the time, the New York Times reported in 2018. White men, by contrast, were misidentified in just 1% of cases.
The reason: AI models are disproportionately built using photos of white men. When fewer photos of people from other racial and gender groups are used, facial recognition tech is less accurate at identifying people of those backgrounds, the study said.
A federal study in 2019 provided similar evidence, finding that Asian, African-American people, and women were much more likely to be falsely identified by the technology than white men.
And a 2022 report by the MIT Technology Review found that Uber's facial recognition feature — used for identity verification, similar to Spark — could have misidentified drivers in India, partly because AI models aren't as well-trained on South Asian faces as white ones.
The report also suggested that a simple appearance change like growing a beard could prevent the tech from detecting someone's face.
Facial recognition developers say mistakes are possible
None of that appears to have stopped the tech from rolling out to new places. Major landlords have added the tech to office buildings as workers have returned to offices, for instance.
Airline passengers can encounter facial recognition at multiple points in an airport, from the security line to customs and passport control, The New York Times reported in February.
Walmart worked with a company called Persona to add facial recognition to its Spark app. While Persona's website doesn't specifically mention trouble identifying people of color, it does acknowledge that the tech has pitfalls.
"For example, someone's eyeglasses may fool the system into thinking it's a reflection on a screen, or a low-resolution photo may trick the system into thinking it's a digital replay," its website reads.
"These shortcomings may increase the risk of false negatives during the verification process, which may result in legitimate users being denied verification," it continues. Persona and Walmart did not respond to BI's requests for information about the technology or partnership.
The TSA and CBP did not respond to requests for comment from BI. A TSA spokesperson told the New York Times in February that halting biometric technology like facial recognition would "take us back years," but did not specifically address the racial disparity.
Retailers' attempts to use facial recognition to catch shoplifters haven't gone well.
In some cases, it's possible to opt out of using facial recognition technology — and its high error rate. If getting on your next flight involves scanning your face instead of a boarding pass, for instance, the gate agent is supposed to offer you a manual check of your ID and pass instead, according to CBP.
But sidestepping facial recognition often isn't possible, said Gideon Christian, a law professor at the University of Calgary who has written about the legal and societal aspects of facial recognition technology. He pointed to some retailers' decision to use the tech in their stores to catch shoplifters.
In December, for instance, the FTC said that drugstore chain Rite Aid had improperly used facial recognition technology in "hundreds of stores" to the detriment of women and people of color.
Rite Aid's use of facial recognition technology "left its customers facing humiliation and other harms, and its order violations put consumers' sensitive information at risk," Samuel Levine, the FTC's Bureau of Consumer Protection director, said at the time.
The FTC banned Rite Aid from using facial recognition technology in stores for five years as a result. Rite Aid accepted the settlement but disputed the FTC's allegations about how and where it used facial recognition — the company said it only tested the tech in "a limited number of stores," for instance.
Facial recognition's racial bias is the same bias that black people already experience when they go shopping, Christian said.
"You basically have facial recognition technology basically making the same baseless allegation, because the face matches with some other individual in the database and then triggers alerts," Christian said in an interview.
Christian told BI that its lack of privacy and its accuracy issues should be enough reason to roll back its use.
"I do not think it's appropriate for us to violate the fundamental right of every person walking into a retail store just because we want to catch a few shoplifters," he said.
Do you work for Walmart Spark, Instacart, DoorDash, or another gig delivery service and have a story idea to share? Reach out to abitter@businessinsider.com