The reported proposal
Associated Press reported that officials in Kansas City, Missouri, were preparing facial-recognition capabilities for cameras on some public buses, intended to identify people on specified lists such as banned riders or missing persons.
A proposed deployment should not be described as a proven reduction in crime. The report documents the plan and the debate around it, not an independent evaluation of its results.
Four questions for public oversight
Who can add someone to a list, how is an erroneous entry corrected, how long is information retained, and what must a human verify before taking action? Those rules affect the consequences of an alert.
NIST’s research shows that face-recognition performance depends on the algorithm, application and data, with demographic differences in many tested systems. A laboratory result is not a direct measurement of a particular bus installation. Public evaluation should therefore examine the actual system and its operating conditions, along with the procedure for challenging a mistaken match.
