By Omid Ghaffari-Tabrizi & Alexandra McLeod
April 21, 2021 at 5:00 am ET
At the heart of the U.S. technology industry, the Department of Commerce plays a crucial role in the development of reliable technology through the team at the National Institute of Standards and Technology. This is especially true for emerging technologies that lack recognized standards.
Facial recognition is just one example of such an emerging technology, but it is one that holds great promise and opportunity in the future of how we live and work with one another, enabling touchless experiences that promote public health and advancements in the field of computer vision used for self-driving cars.
The need for recognized standards that ensure the responsible use and implementation of the technology is where NIST plays a critical role. However, the labs under NIST’s umbrella are often left behind due to a lack of commercially available technology. Through the use of report language, congressional support for the Face Recognition Vendor Test lab would allow NIST to not only keep up with current advances, but also remain at the forefront of ensuring that technologies of the present and future are safe, ethical and bias-free.
NIST developed the first FRVT more than 20 years ago in 2000, three years after completing its Face Recognition Technology program. When FERET wrapped up, facial recognition technology was still considered to be at the prototype stages, housed in university and research labs. But by the publication of FRVT 2000, the technology had advanced so far and so fast that it became widely available in commercial products.
Similarly, current cloud-based facial recognition technologies have developed at such a rapid rate that the FRVT lab is no longer able to effectively test for bias or other inaccuracies in the underlying technology. As NIST stated in its test on demographic effects in NISTIR 8280, the lab is tasked to test products in a manner that “reflects the usual operational situation in which facial recognition systems” operate, but because of the limitations in IT infrastructure, NIST is unable to do so.
Fortunately, there is a solution, and it will not require new legislation or additional funding beyond the resources contained in the currently proposed budget. Through the use of funding already being provided to the U.S. Department of Commerce in Fiscal Year 2022, NIST should develop a cloud-based IT infrastructure that will support performing the FRVT on products exactly as they are being offered and implemented in public and private sector use cases. Providing NIST with the tools and talent necessary to complete this test will ensure that the agency is able to: (1) carry out its essential function of creating the national standard for this form of technology, and (2) identify and resolve key concerns around biases or inaccuracies to ensure responsible use.
The impact of this will be clearly and immediately felt. NIST will be able to develop standards, guidelines and associated methods and techniques for testing and benchmarking accuracy and bias in facial recognition software products. These tools will be a simple but important step to facilitate the creation of accurate facial recognition products.
Legislators involved with the appropriations process should take advantage of this opportunity and provide full and robust funding to NIST in order to further establish democratic principles as a foundational component of emerging technologies.
Omid Ghaffari-Tabrizi is the director of cloud policy at Internet Association, with a focus in promoting cloud-first policies and modernization efforts in the private and public sector.
Prior to joining Internet Association as the legal & policy counsel, Alexandra McLeod was Associate Policy Counsel at ACT | The App Association, where she worked on issues including state and federal privacy, connected health, competition, telecommunications and standard essential patents.
Morning Consult welcomes op-ed submissions on policy, politics and business strategy in our coverage areas. Updated submission guidelines can be found here.