ACLU’s Amazon Rekognition Tests Turn Up 28 Mismatched Faces of Congressmen
- The ACLU conducted tests using Amazon’s Rekognition software and revealed that the software is not accurate enough for government use.
- Over 535 members of the Congress were matched against mugshots of 25,000 citizens, and 28 results turned out to be inaccurate.
- The ACLU believes that the software’s inaccuracy may end up costing the freedom of innocent citizens and should not be used in its current state.
Amazon’s facial identification system Rekognition was used for a test, and the results are far from promising. The software fell short of expectations when the American Civil Liberties Union ran into 28 errors when testing it for over 500 faces against publicly available mugshots. The tests were conducted with the help of the Washington County Sheriff’s Department in Oregon. While Amazon is satisfied with the 95% threshold, the ACLU believes that the Rekognition app needs to avoid cases of false IDs.
The ACLU is concerned about the usage of Rekognition and believes that 80% recognition rates are fine for social media use cases, but for legal usage, Amazon needs to offer certainty with its Rekognition technology. Rekognition was first reported to be used by Orlando’s police force. Amazon charges only $12 a month for an entire department, making it an inexpensive solution for tracking persons of interest.
The ACLU revealed their opinion about the software recently stating that “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 Rekognition is not the only first app of its kind with the London Metropolitan Police using similar software with 49 false matches for every hit. Amazon’s solution may be the most effective till date, but every inaccurate report can cause innocents being drawn into unwanted legal complications.
False positives can come up for any facial ID software, and Rekognition is no different. If police authorities want to use such applications, they need to check each and every result to confirm a match manually. Rekognition also raises the question about public privacy with mugshots of citizens taken by the cameras being publicly available.
What do you think about Amazon’s facial recognition app? Let us know in the comments below. Get instant updates on TechNadu’s Facebook page or Twitter handle