Skip to Content
Artificial intelligence

A US government study confirms most face recognition systems are racist

December 20, 2019
A U.S. Customs and Border Protection officer helps a passenger navigate a facial recognition kiosk at the airport.
A U.S. Customs and Border Protection officer helps a passenger navigate a facial recognition kiosk at the airport.
A U.S. Customs and Border Protection officer helps a passenger navigate a facial recognition kiosk at the airport.David J. Phillip/AP

Almost 200 face recognition algorithms—a majority in the industry—had worse performance on nonwhite faces, according to a landmark study.

What they tested: The US National Institute of Standards and Technology (NIST) tested every algorithm on two of the most common tasks for face recognition. The first, known as “one-to-one” matching, involves matching a photo of someone to another photo of the same person in a database. This is used to unlock smartphones or check passports, for example. The second, known as “one-to-many” searching, involves determining whether a photo of someone has any match in a database. This is often used by police departments to identify suspects in an investigation.

The agency studied four face data sets currently used in US government applications: mug shots of people living in the US; application photos from people applying for immigration benefits; application photos from people applying for visas; and photos of people as they crossed the border into the US. In total, the data sets included 18.27 million images of 8.49 million people.

What they found: NIST shared some high-level results from the study. The main ones:

1. For one-to-one matching, most systems had a higher rate of false positive matches for Asian and African-American faces over Caucasian faces, sometimes by a factor of 10 or even 100. In other words, they were more likely to find a match when there wasn’t one.

2. This changed for face recognition algorithms developed in Asian countries, which produced very little difference in false positives between Asian and Caucasian faces.

3. Algorithms developed in the US were all consistently bad at matching Asian, African-American, and Native American faces. Native Americans suffered the highest false positive rates.

4. One-to-many matching, systems had the worst false positive rates for African-American women, which puts this population at the highest risk for being falsely accused of a crime.

Why this matters: The use of face recognition systems is growing rapidly in law enforcement, border control, and other applications throughout society. While several academic studies have previously shown popular commercial systems to be biased on race and gender, NIST’s study is the most comprehensive evaluation to date and confirms these earlier results. The findings call into question whether these systems should continue to be so widely used.

Next steps: It’s now up to policymakers to figure out the best way to regulate these technologies. NIST also urges face recognition developers to conduct more research into how these biases could be mitigated.

To have more stories like this delivered directly to your inbox, sign up for our Webby-nominated AI newsletter The Algorithm. It's free.

Deep Dive

Artificial intelligence

chasm concept
chasm concept

Artificial intelligence is creating a new colonial world order

An MIT Technology Review series investigates how AI is enriching a powerful few by dispossessing communities that have been dispossessed before.

open sourcing language models concept
open sourcing language models concept

Meta has built a massive new language AI—and it’s giving it away for free

Facebook’s parent company is inviting researchers to pore over and pick apart the flaws in its version of GPT-3

spaceman on a horse generated by DALL-E
spaceman on a horse generated by DALL-E

This horse-riding astronaut is a milestone in AI’s journey to make sense of the world

OpenAI’s latest picture-making AI is amazing—but raises questions about what we mean by intelligence.

labor exploitation concept
labor exploitation concept

How the AI industry profits from catastrophe

As the demand for data labeling exploded, an economic catastrophe turned Venezuela into ground zero for a new model of labor exploitation.

Stay connected

Illustration by Rose WongIllustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.