“Really at the end of the day, this technology could result in tremendous discrimination not only because the facial recognition technology is proven to be inaccurate when it comes to minorities,” said Ari Scharg, partner at Edelson PC. Research from MIT, for instance, suggests the facial recognition software is generally more accurate with male versus female faces, and it is better recognizing whiter subjects than darker-skinned ones.
Washington County Sherriff’s Office
While acknowledging such discrepancies can creep into the training data used for facial recognition software, Peter Tran, head of global cyber defense and security strategy at Worldpay said part of the seeming bias is a result of lighting challenges. The default camera settings on surveillance cameras may not accurately capture dark skin tones, and images with a lack of contrast are simply more difficult for facial recognition software to handle.
While also stressing the importance of sensical data governance policies involving the technology, Tran disputes the notion that the prospect of law enforcement using the Rekognition technology was itself inherently problematic. “Think of criminal cases that have been adjudicated based on testimony from a single eyewitness that may or may not be accurate,” he said. “Or the ability of technology to help confirm a suspect’s alibi.”
Interpol already notes that biometric technology is an important tool in fighting terrorism. Orlando Police Chief John Mina, who is also involved in testing the Rekognition technology, envisions it could be a useful tool in fighting school shootings and kidnapping threats.
To date, the Police Department in Oregon’s Washington County have used the technology to capture criminals in retail environments. In early 2017, a manager from a local “big-box” hardware store reached out after a man went to the self-check aisle with a number of expensive items and then suddenly fled without paying. The store had a high-quality image of the man, which it shared with the police. The police, using their database of some 300,000 mugshots, were able to use the Rekognition system to identify four individuals with similar facial characteristics. A detective using this information was able to use this information to identify the likely suspect, even finding the Facebook profile with the suspect apparently wearing the same hoodie as in the hardware-store footage.