Clearview AI, the artificial intelligence startup from which I reported earlier this week, has released the results of its third round of biometric testing, and it’s clear that the company has a long way to go before it can compete with the human Genesee project. Clearview showed its artificial intelligence software to some 300 people, of which 185 had their face scanned and two were randomly selected. There were 17 legitimate errors and 19 false positives. Eight of the false positives, according to Clearview, were related to the temperature of the subject’s mug or were due to test results obtained from inanimate objects, such as a computer mouse or cell phone.
Among the results of the study: Humans scored the best, when using Clearview’s system, with a detection accuracy of 84 percent. Second place went to a kit developed by Verimatrix, a company that develops facial recognition software. It achieved a true positive detection of 79 percent (ouch). Clearview AI scored a 64 percent detection accuracy, compared to a 75 percent detection accuracy by Genesee. Clearview scored a 45 percent detection accuracy by Kwartz. Clearview scored a 55 percent detection accuracy by Ricard. Clearview and Genesee also had higher false positive rates, with Clearview having three false positives for every three failures. Of all of the human tests, Genesee had three false positives for every five failures.
“Clearview’s highly advanced facial recognition algorithm,” the company said in a release, “has been consistently far superior to Genesee since we started shipping our product last year. These results provide the latest evidence that Clearview’s process is the best solution on the market for law enforcement, and will also further establish the superiority of our technology in face recognition programs for new retail and hospitality applications.”
Have I been treated unfairly?
It appears that I have been. When I used the popular app Cognigo in the past, I found my face to be less accurate than my wife’s, who usually is tagged as a positive: the difference is made by using a simple calibration process. So it appears that I may indeed have been being treated unfairly, at least initially.