ACM public briefing emphasizes facial recognition risks, SIA argues for responsible policy
The Association for Computing Machinery (ACM) provides a ‘TechBrief’ on facial recognition meant to inform public dialogue, and warning that systems using the technology “should be considered biased absent evidence to the contrary.”
The brief begins by referring to facial recognition’s “fundamental limitations, creating profound privacy and ethical challenges,” and stating that the technology “is insufficiently trustworthy in many high-stakes applications to replace identification by a human.” The starting position highlights some of the findings around demographic disparities in face biometrics accuracy, and reviewing policy actions to curb or control its use.
While none of these positions are overly controversial, with high-stakes applications like courtroom identification off-limits for algorithms, several more contentious claims follow.
The brief suggests that “corporations and governments have increasingly determined that the only responsible way to use facial recognition technology is to not use it at all.” The claim is supported with reference to a map from Fight for the Future, though earlier in the document the expanding use of the technology in the government and overall market are both noted, and the dissonance between these facts is not addressed.
A section on privacy and facial recognition discusses the use of the technology in public settings, the source material for training data, and a section on accuracy concludes high performance does not sufficiently show trustworthiness or responsible use. The brief also includes various analysis functions in the review, and notes that claims to learn non-visual information from facial features should be considered suspect.
SIA argues for policies to support trust
The Security Industry Association (SIA), not surprisingly, sees things differently.
A letter to the majority and minority leaders of both the U.S. House and Senate signed by SIA CEO Don Erickson argues that public trust in facial recognition can be engendered to help law enforcement and border control processes by leveraging policy group already undertaken by the Association.
The SIA recommends the application of its policy principles for transparency, clear and defined purpose, and high accuracy when implementing facial recognition systems. Governance frameworks for biometric data collection and an emphasis on opt-in and -out processes for applications outside of public safety are also advised.
On the point raised by the ACM, the SIA argues that the use of NIST FRVT results enables organizations to avoid algorithmic bias, and that policy makers should ensure that agency use policies include clauses guaranteeing the constitutional rights of individuals. The letter further underscores the importance of timely deletion of biometric data and communicating whether facial verification or identification is being used.
Christine Wilson of the U.S. Federal Trade Commission (FTC) set out her views on federal regulation for data privacy and facial recognition in comments during a Chamber of Commerce podcast and analyzed by Squire Patton Boggs Attorneys in a Consumer Privacy World blog post.
The Trump administration appointee said that while she generally opposes increased regulation, clarity is needed in the marketplace. The FTC has enough authority over facial recognition and AI for the light-touch regulation appropriate to newer technologies, according to Wilson, through the FTC Act Section 5.