By Hoan Ton-That BLOG
Facial Recognition Technology (FRT) has been deployed in the U.S. Government substantially post-9/11. The groundwork was laid in 2003 when the State Department mandated fingerprint collection and a digital photograph for each visa applicant in the United States, and it has been steadily building up since. The last two years especially have seen widespread adoption of FRT in the federal government due primarily to the dramatic improvement in FRT accuracy and the issuance of practical use guidelines from the GAO.
Despite this favorable shift in attitudes toward FRT by some of the world’s most important public institutions, there is still healthy skepticism toward the technology which we in the industry must do our best to address through facts and science.
Concerns about accuracy are usually the first mentioned by doubters. On that front, testimony given by Dr. Charles Romine, NIST’s Information Technology Laboratory Director, to the House Homeland Security Committee on February 6, 2020, has been widely overlooked and is essential to understanding just how accurate FRT algorithms have become when used to make bias-free positive identifications in one-to-many applications. According to Dr. Romine, NIST’s testing revealed “undetectable bias in demographic differentials”[1] and that there is “no statistical level of significance” [1] to support any claim of bias among the highest performing algorithms the agency has tested – and that was more than two years ago.
Policies around appropriate use are another concern. Last year, in 2021, a series of reports by the GAO laid out guidelines for any federal agency to purchase facial recognition for law enforcement applications and a wide range of other uses. Their findi