GSA finds ‘bias’ in facial recognition errors, plans tests but no deployment
Tests conducted by America’s General Services Administration have shown higher false rejection rates in face biometric systems for African Americans accessing digital government services, leading the agency to decline the use of facial recognition and liveness technologies until the issue is resolved.
The GSA’s ‘Equity Action Plan’ notes that “major commercial implementations of facial matching had disproportionately high ‘False Rejection Rates’” (or FRRs) when attempting to match Black people with their photo IDs. The GSA attributes the elevated error rates to training datasets for biometric algorithms that are racially unbalanced.
The agency also refers to NIST testing that shows many, though not all, facial recognition algorithms exhibit significant demographic differentials.
Facial recognition, biometric liveness detection, and all other emerging technologies will not be implemented by the GSA due to this concern “until rigorous review has given us confidence that they can be implemented equitably and without causing disproportionate harm to vulnerable populations.”
“To provide an equitable remote identity-verification experience for a diverse population, GSA’s Login.gov team will perform research studies on equity and bias in facial matching services,” the document states.
What this means for the use of Acuant’s COFRS for disaster relief by Login.gov, which is operated by the GSA, is not stated.
The decision also bears heightened importance with Login.gov identified as an eventual alternative means of authentication for people using the IRS’ online services in the wake of controversies around the use of face biometrics.
The report also notes the barriers people with “visual, mobility and cognitive impairments” encounter when trying to access government programs and services online.
The announcement does not entirely turn the GSA away from biometrics, however, as the agency granted a Multiple Award Schedule (MAS) contract to Biometric Signature ID just a month ago.
accuracy | bias | biometric liveness detection | biometric matching | biometric testing | biometrics | face biometrics | facial verification | government services | identity verification | U.S. Government