In the wake of the September 11, 2001, terrorist attacks on the United States, politicians and companies are lining up to bring Orwellian invasions of privacy in order to give the illusion of safety. One technology, already famously used earlier this year at the Super Bowl, is computerized face recognition technology. But when you dig into the details of how this works not only is it an unwarranted invasion of privacy, but the technology doesn’t even work as claimed.
Take, for example, the system being sold by FaceKey which has been evaluated by several government agencies. The system works great in artificially controlled conditions created by the company, but when deployed in the real world, it has a false positive rate that will render it almost useless.
Deploying it in normal lighting conditions and comparing a quick scan of individuals with prior photographs of the test subjects, the best success rate was a 33 percent false positive with a 10 percent false negative. In other words, at best 1 in 3 people who weren’t listed as terrorists were nonetheless flagged as such, while 10 percent of the people designated as terrorists were cleared and allowed to continue on their way.
Consider what this would mean in a real world situation like the rather small Gerald R. Ford International Airport in Grand Rapids, Michigan. About 2.4 million people pass through Ford International airport. Using this system, assuming that 100 terrorists pass through the airport and the system manages to catch every one, another 660,000 people would be wrongly flagged as terrorists and require some sort of investigation with security officials.
Assume that each such investigation takes only 15 minutes to clear up and the security folks are being paid $9/hour. That amounts to $1.5 million each year just to identify people who are not terrorists.
This system would quickly be ignored by security officials (“there goes the damn face recognition software again. What a piece of crap.”) as it would be completely unworkable.
Source:
Face recognition useless for crowd surveillance. Thomas C Greene, The Register, September 27, 2001.