The MITRE Corporation, a technology non-profit group, has released a report with insights, analysis, framework and recommendations for the use of facial recognition systems. Characterizing the technology as misunderstood and complex, MITRE says the government should stop short of a moratorium on facial recognition, rather issuing a resolution that deters near-term misuse and signals long-term regulations and oversight.
Previous studies have claimed facial recognition technology is inherently bias to minorities and non-white individuals. After the murder of George Floyd, IBM, Microsoft and Amazon all said they were getting out of the facial recognition business with concerns about how it can be used for mass surveillance and racial profiling. Additionally, the NYPD and other local and federal agencies caught flak at the beginning of this year for utilizing Clearview AI, a controversial facial recognition app that amassed an unprecedented database of 3 billion images by scrapping data and photos from social media sites.
Given demographic variances, MITRE says it’s impossible to develop a system with 100% accuracy, but it is possible—although “more difficult”—to ensure similar algorithm accuracy across different demographic groups.
“While this can understandably make many uncomfortable, policymakers’ focus really needs to be on ensuring accurate and equitable outcomes from the total operational system rather than the algorithm subcomponent only,” the report reads.
MITRE’s analysis also revealed a dramatic difference in the use and deployment of facial recognition across cases and agencies. This has been seen in real-time as some police departments have facial recognition standards and some don’t, whereas others do not make theirs public.
“Issues such as error rates, demographic performance, privacy and civil liberties, ethical appropriateness, and management and oversight requirements will vary substantially across different use cases. General investigations or improper mixing of multiple use cases will undoubtedly lead to incorrect assessments and decisions,” reads the report.
MITRE places emphasis on the establishment of a framework to ensure facial recognition technology can exist—within standards. The framework would need to be specific to individual use cases, and should involve a variety of stakeholders including technical, operational, legal, social justice and policy experts that have training and personal experience with the technology.
In the interim, MITRE suggests Congress issue a resolution expressing concerns and expectations of the technology, including the requirement of “keeping a human in the loop” when facial recognition is used to identify individuals in law enforcement investigations.
Beyond facial recognition technology issues, MITRE notes that the United States’ privacy laws are now too old to keep up with the technological leaps and bounds society has made in the 21st century.
“The nation needs to perform a fundamental assessment of privacy considerations, in both government and commercial applications, so that a modern construct can be developed,” the report concludes.