Tuesday, August 14, 2018

Facial recognition software generates 28 false positives linking US Congress members to mugshots

The American Civil Liberties Union (ACLU) tested Amazon's facial recognition system by scanning all 535 members of the US Congress against 25,000 mugshots public mugshots. The test generated 28 false positives.

 1 of 3 
None of the members were in mugshots. The Amazon system called Rekognition is already in use by some police departments.
The ACLU wants a moratorium on the use of Rekognition by police
ACLU said: “An identification — whether accurate or not — could cost people their freedom or even their lives. Congress must take these threats seriously, hit the brakes, and enact a moratorium on law enforcement use of face recognition.”
Amazon blames results on faulty calibration
An Amazon spokesperson said to Verge that the result was due to poor calibration. The ACLU tests were done using Amazon's default setting of 80 percent. However, the spokesperson said that for law enforcement applications at least a 95 percent application should be used as a false ID could have significant consequences.
The Amazon representative said: “80% confidence is an acceptable threshold for photos of hot dogs, chairs, animals, or other social media use cases. It wouldn’t be appropriate for identifying individuals with a reasonable level of certainty. it wouldn’t be appropriate for identifying individuals with a reasonable level of certainty.”
Yet, there is nothing in the setup process that enforces the recommendation meaning that there is nothing to prevent police from using the default setting.
Amazon reported a number of police depts. used Rekognition
The Orlando police in Florida used the system as a pilot project in real-time recognition. The system is sold as part of Amazon's Web Services. It can often cost less than $12 a month for an entire department. Although the Orlando project has ceased, the department continues to use the system.
ACLU is examining Recognition use by an Oregon Sheriff's Department
The latest ACLU experiment is designed in particular to examine Washington County's Sheriff's Department in Oregon's use of Recognition. The department has compared images against a database of as many as 300,000 mug shots.
Jacob Snow who organized the test for the ACLU of Northern California said: “It’s not hypothetical. This is a situation where Rekognition is already being used.”
Test shows signs of racial bias
11 of 28 or about 39 percent of false matches were of people of color and included civil rights leader Rep, John Lewis a Democrat of Georgia and five other members of the Congressional Black Caucus. Only 20 percent of the present Congress are people of color. People of color were obviously identified at a higher rate.
Problems with the use of Rekognition
In practice many facial recognition positives would be confirmed by a human before they led to concrete action such as an arrest. However, Snow said: “Imagine a police officer getting a false match for somebody with a concealed weapon arrest. There’s a real danger if that information is surfaced to the officer during a stop. It’s not hard to imagine it turning violent.”
Snow told Reuters:“Face surveillance is flawed, and it’s biased, and it’s dangerous.”
Nevertheless the outlook of Amazon was positive and touted a range of uses for Rekognition according to a spokeswoman for Amazon Web Services:“We remain excited about how image and video analysis can be a driver for good in the world.” She said the system could be used to help find lost children and preventing crimes. She noted that they system was usually used to narrow the field rather than make final decisions.

No comments:

US will bank Tik Tok unless it sells off its US operations

  US Treasury Secretary Steven Mnuchin said during a CNBC interview that the Trump administration has decided that the Chinese internet app ...