Amazon has a pretty straightforward response to the news that emerged last week about its facial recognition system incorrectly matching photos of 28 members of congressmen to mugshots. Those results came a result of testing by the ACLU, which took 25,000 publicly-available police mugshots and then asked Rekognition to compare those images to photos of all 535 members of Congress.

When the results came in, 28 lawmakers were positively ID’d as matching the faces of arrestees. In a blog post written by Matt Wood, Amazon’s general manager for deep learning and artificial intelligence, he explains the ACLU used the tech’s default setting of an 80 percent confidence level.

Amazon suggests setting the confidence threshold instead at 99 percent (“As we recommend in our documentation”), at which point the misidentification rate, Matt writes, drops to zero. “This illustrates how important it is for those using ‎the technology for public safety issues to pick appropriate confidence levels, so they have few (if any) false positives.

He goes on to point out that in public safety and law enforcement scenarios, Rekognition is “almost exclusively used” to help narrow the field, to then allow humans to come in and quickly review options using their judgement — as opposed to the system making fully autonomous decisions. “A final word about the misinterpreted ACLU results. When there are new technological advances, we all have to clearly understand what’s real and what’s not,” Wood writes.

“There’s a difference between using machine learning to identify a food object and using machine learning to determine whether a face match should warrant considering any law enforcement action. The latter is serious business and requires much higher confidence levels. Read more from…

thumbnail courtesy of