Amazon’s face-ID tool Rekognition misidentified members of Congress

Amazon’s face-identification software Rekognition can’t even correctly identify members of Congress.

The American Civil Liberties Union tested the real-time face identification software (which can identify every single face in a crowd) by comparing photos of member of Congress to mugshots, and 28 members were misidentified as the people in the mugshots.

This study is part of the ACLU’s campaign to get Congress to forbid law enforcement from using face recognition technology on the grounds that it reinforces the criticism that facial identification software suffers from the racial biases of the people who create them. Of the 28 false positives in the study, six of the images were those from the Congressional Black Caucus. Furthermore, even though people of color representatives only comprise of around 20 percent of Congress, they still comprised more than 40 percent of the false positives.

Amazon has successfully been marketing Rekognition to law enforcement agencies around the country, and the ACLU has been rallying with other civil rights organizations like the Electronic Frontier Foundation to stop agencies from using this technology for surveillance (not, Amazon claims, what the technology was made for, although the ACLU found that it is being marketed to cities for such purposes).

Jacob Snow, a tech and civil liberties attorney for the ACLU, told Mashable that he conducted this test of the Congress members’ faces for under $13, which highlights how cheap and accessible it is — even though the results are inaccurate. Although some of the false positives were similar in looks, Snow said, the bottom line is that similar does not mean the same.

But that similarity, he said, could mean the difference between life or death.

“One of the things that is dangerous about presenting this information in a law enforcement context is that there can be differences — in lighting, in angles, in age — so it can be genuinely difficult to say just based on the photos that they are the same person,” Snow said. “Facial recognition has the possibility of suggesting to a law enforcement user that there is a match. And then there is a high probability or a reasonable probability that the law enforcement user will trust the system and not apply the same level of skepticism.”

The Congressional Black Caucus recently even wrote a letter to Amazon boss Jeff Bezos that expressed its concerns about unwarranted ramifications the face recognition technology his e-commerce giant sold could have on Black people, undocumented immigrants, and protesters.

The Washington County Sheriff’s Office in Oregon and the Orlando Police Department are both already using Rekognition.

Although Orlando is only piloting Rekognition (for the second time), its usage has received backlash from the community for a variety of reasons, including that the first pilot was done without consulting the public and that it is happening in such a diverse the city at a time of heightened protest.

“This highlights that there is real, concrete harm that can come to the public if facial recognition is deployed, especially by law enforcement,” Snow said.

“100 percent of the matches that we got were inaccurate, and it really underscores how face recognition can suggest that there is a close match between two faces when in fact that match doesn’t exist.”

Maybe the Congress members will react now that they know what it feels like on the receiving end of being misidentified. But regardless, this is just another incident that shows how face recognition simply cannot do its job correctly.

UPDATE: July 26, 2018, 12:07 p.m. EDT Amazon told Mashable, “With regard to this recent test of Amazon Rekognition by the ACLU, we think that the results could probably be improved by following best practices around setting the confidence thresholds (this is the percentage likelihood that Rekognition found a match) used in the test.”

UPDATE: July 27, 2018, 11:51 a.m. PDT Amazon posted a blog post Friday that addressed the ACLU’s use of the tool. It called the ACLU’s results “misinterpreted.”

It emphasized setting a higher confidence threshold for public safety scenarios. “There’s a difference between using machine learning to identify a food object and whether a face match should warrant considering any law enforcement action. The latter is serious business and requires much higher confidence levels,” the post read.

The post ended with a pizza baking analogy to explain how machine learning is a process and constantly improving as more data is added. “We should not throw away the oven because the temperature could be set wrong and burn the pizza,” the post ended.

Read more: