Two prominent members of Congress have asked to meet with Amazon CEO Jeff Bezos after Amazon’s Rekognition image recognition software incorrectly matched their images with other people’s mugshots in an American Civil Liberties Union test, BuzzFeed News reports.
Representatives Jimmy Gomez, a California Democrat and member of the Congressional Hispanic Caucus, and John Lewis, a Georgia Democrat and member of the Congressional Black Caucus, expressed concern about the “potential impact of this technology on the civil liberties, particularly in communities of color.”
Did you see this? @amazon face surveillance technology FALSELY matched me w/ someone else’s mugshot. I’m outraged & worried by the impact this tool will have on #CommunitiesOfColor when put in the hands of law enforcement! @JeffBezos: We need to talk ASAP. https://t.co/xFOy8duef1
— Rep. Jimmy Gomez (@RepJimmyGomez) July 26, 2018
Separately, Senator Edward Markey, and representatives Luis Gutiérrez and Mark DeSaulnier, who were each also falsely matched, wrote a letter to Bezos also expressing their concern. They asked Bezos to answer various questions, including:
- What bias testing Amazon has done on Rekognition?
- Which law enforcement agencies are using the technology?
- Does Amazon audit agencies’ use for privacy and terms violations?
- Does the company considers law enforcement agencies’ histories of discrimination in deciding whether to sell to them?
They asked Bezos to respond by August 20.
In the ACLU test, the group loaded 25,000 publicly available mugshot photos of people who had been arrested and asked Rekognition to match them with photos of members of Congress. The tool, part of the Amazon Web Services cloud system, found incorrect matches for 28 legislators. The false match rate was 39% for people of color in Congress, while people of color only make up 20% of Congress as a whole. Law enforcement agencies have already reported using Rekognition and other image matching tools to find and identify suspects.
For its part, Amazon has said it encourages law enforcement agencies to use at least a 95% confidence threshold, while the ACLU used only an 80% level. Jacob Snow, a technology and civil liberties attorney with the ACLU of Northern California, says the group simply used the system’s default settings.