Clearview AI’s controversial facial recognition platform continues to perform well in independent testing. The company was the top-rated US solution in NIST one-to-one Face Recognition Vendor Test (FRVT) results that were released in October, and has now achieved the same feat in the NIST’s most recent one-to-many evaluation.
In both tests, Clearview’s facial recognition solution was the most accurate US solution in every evaluation category. It also made it into the top 10 internationally in each category in the one-to-many test, and the top five in the one-to-one.
The one-to-many test measures an algorithm’s ability to find a match for a new facial image in a database that includes millions of individual faces. In that regard, Clearview’s platform was 99.85 percent accurate when matching mugshots, and 99.86 percent accurate with VISA border photos. The mugshot database includes 12 million photos, while the VISA border database has roughly 1.6 million. Clearview was the second most accurate algorithm in the world in the VISA kiosk test.
Of course, those results do not address the ethical and legal concerns that have followed Clearview for the past two years. The company built its system with a database that includes more than 10 billion images scraped from various social media platforms, and has openly flaunted the fact that those images were gathered without any form of consent. Clearview was hit with several cease and desist orders when its database only had 3 billion images, and multiple countries have now determined that the company’s practices violate their privacy laws. A group of Canadian privacy commissioners made their ruling in February, while the Australian Information Commissioner made the call in November, shortly after Clearview’s first NIST test.
Clearview is still facing additional lawsuits in the US and abroad. It has also drawn criticism from other facial recognition providers, who view the company’s practices as unethical. The company has primarily targeted the law enforcement sector, though some police forces have tried to deceive the public about their use of the technology.
November 29, 2021 – by Eric Weiss