An event last week on ‘Identity, Authentication and the Road Ahead’ held by the Better Identity Coalition, the FIDO Alliance and the Identity Theft Resource Center held a panel discussion among biometrics experts on equity and bias concerns in identity proofing.
The panel was moderated by Elizabeth Bynum Sorrell of the Beeck Center for Social Impact & Innovation at Georgetown University. Panelists were Socure VP and Head of Public Sector Strategy Jordan Burris, CITeR Director Dr. Stephanie Schuckers, and Andrew Stettner of the U.S. Department of Labour, where he is the deputy director for policy at the Office of Unemployment Insurance Modernization.
They discussed the challenges that identity verification poses for various groups, as highlighted by problems with pandemic relief programs and manifesting in mirrored ways in the public and private sectors.
The problems are in many cases not a function of biometric bias, but can be related to digital literacy, internet access, or common changes in life circumstances.
This is despite the invocation of “inherent” bias in the technology, Schuckers explained.
“The perception that comes from that kind of language is that it is inherently biased,” Schuckers says. “That it’s built in; if you do biometric recognition there will be bias. Which is the kind of thing we want to move past, because there are ways to assess those systems to determine where the biases are.”
She explained that good assessment tools are in place and further ones are being developed, and now requirements must be developed. A certification would be good to have.
Stettner notes that social benefits legislation requires states to have a reasonable assurance of the accuracy with which they are verifying identities. Requirements for states contracting identity verification vendors were put in place during the pandemic in part to ensure the chosen solutions are equitable, but he says there is still significant potential benefit to a public sector system like Login.gov. A trial of that system is being expanded in Arkansas.
Burris emphasized that biometrics are just one part of the ID equation. Problems are caused by inappropriate requirements in rules-based systems he says, providing the example of people in the middle of name change, or moving from one state to another, who could be rejected by an automated system even if they pass a biometric check.
He also called for greater transparency among identity verification vendors about what their pass rates are. If there is a variable in the identity string fed into a machine learning model that impacts a certain group of people differently, the vendor should know, and make that information available to their customers.
The call for transparency was echoed by the other panelists.