Little overlap between NIST facial recognition testing leaders and U.S. government providers
Wired is asking “Why Chinese Companies Plug a U.S. Test for Facial Recognition” in an article highlighting the placement of six Russian and Chinese companies at the top of NIST’s facial recognition vendor test (FRVT) Ongoing 1:1.
The U.S. State Department selected IDEMIA’s facial recognition technology to perform screening for passport application last June, citing the company’s superior accuracy in NIST testing compared to other non-Russian or Chinese software providers. When Ever AI ranked seventh in subsequent rounds of testing, the company touted its position among Western competitors.
“Ever since the NIST results came out there’s been a pretty steady stream of customers,” including new interest from government agencies, Ever AI CEO Doug Aley told Wired. YITU algorithms boast the best accuracy in the 1:1 test so far. YITU AI Research Scientist Dr. Wu Shuang told Wired that both Chinese and international customers ask about the NIST test.
In 1:N testing, Microsoft scored highest in the FRVT, followed by three companies from Russia and China, and then Ever AI. Wired reports that out of more than 60 entrants it could determine the country or origin for, 19 are from Russia or China. For the rest, a good NIST result could lead to U.S. government contracts.
“Federal agencies don’t make buying decisions without checking with NIST,” says NEC Corporation of America VP of Federal Operations Benji Hutchinson.
IBM and Amazon both sell facial recognition technology to local law enforcement in the U.S., but are yet to participate in NIST testing. Amazon says this is because its technology cannot be separated from the AWS platform, while IBM is reportedly working with NIST on accuracy across different demographics while deciding whether to throw its hat in the ring.
NIST scientist Patrick Grother, one of the leads on the FRVT, tells Wired its tests of demographic differences in facial recognition technology are expanding. He says more work is needed to determine how to best test and measure the known problem of bias in the technology.
“We try and bring sunlight and oxygen to the marketplace,” Grother says.