Days of facial recognition discussions uncover a tricky future for policymakers

The U.S. National Academies of Sciences, Engineering, and Medicine asked for a data dump on facial recognition, and for better or worse, it has gotten one.

The policy-informing organization, created to provide in-depth reports on innovations and challenges for society, is preparing a report on face biometrics’ capabilities, prospects and governance. In this case, academies members want what insiders feel that decisionmakers need to know before they legislate, mandate or regulate.

As part of their process, members are holding live online seminars to better understand face biometrics themselves. The meetings were included in this year’s FedID, the Federal Identity Forum and Expo; more will occur Sept. 27.

Of two information-gathering completed by the academies, only one was accessible online on deadline, and it peeled the onion a bit on the accuracy of face recognition algorithms and their potential impacts on the transgender community.

Amy Yates, with the IT lab in the National Institute of Standards and Technology, pulled together research comparing algorithms to human efforts.

The best AI today is competitive with the humans who are best at the same task: forensic examiners and super recognizers (people innately talented at matching faces), says Yates, citing previous research.

But fusing, or combining the efforts an examiner and an algorithm, in this case, A2017b, produced the highest accuracy of either partner or any other combination. Combinations typically resulted in better outcomes, but none, including fusing two AIs, matched a trained human working with software.

Later, Karl Ricanek, a professor at University of North Carolina Wilmington, spoke about democratizing face-based technology for the transgender community, which will be a challenge for algorithms as well as policymakers.

Even hormone therapy, it turns out, can alter a face significantly. Skin density, vascular structure and other factors can change the face enough to fool AI, says Ricanek.

Article Topics

AI  |  algorithms  |  biometrics  |  facial recognition  |  legislation  |  regulation  |  U.S. Government