Kenneth Rijock

Kenneth Rijock

Friday, July 27, 2018

PROOF OF CONCEPT SHOWS THAT NOT ALL FACIAL RECOGNITION SOFTWARE PROGRAMS ARE RELIABLE

Some of the members of the US Congress
 A test, conducted by the American Civil Liberties Union (ACLU) on a prominent facial recognition software platform, confirmed that compliance officers who are considering what program to acquire should carefully check the effectiveness of each program, prior to purchasing it for Enhanced Due Diligence AML/CFT investigations.

The ACLU, using a commercially available FRS program, ran the photographs of the Members of the United States Congress through it, and, to its consternation, returned 28 positives from criminal mugshots. While the provider immediately noted that the ACLU was working with 80% confidence, and stated that a 95% rating would be more appropriate, 80%, which is the default rating, is the rating curently in use by American law enforcement agencies using that specific tool.  Members of Congress who were Persons of Color had More than 40% of the false positives, although they represent less than 30% of the group, which has fueled concerns that the program is neither accurate nor effective.

Before committing to a specific FRS product for AML purposes, you need to be assured that it will be effective, through actual testing, until a consistent pattern of effectiveness emerges. Just as many data companies jumped on the AML compliance bandwagon after 9/11, althrough they had no background in the field, and their products were often ineffective, the emerging facial recognition software industry has had its share of eager firms, all  seeking a part of the market share, but without an effective product; Let the buyer beware.



No comments:

Post a Comment

Note: Only a member of this blog may post a comment.