Man mistaken for his co-workers illustrates the flaws of facial recognition
Facial recognition systems are common in China, but one man with a 'common face' couldn’t clock into work
We’re used to hearing about the advances in AI facial recognition in China. The technology, for instance, has helped police arrest three fugitives in just two months, all during concerts by Canto-pop star Jacky Cheung. However, we’re less used to hearing about facial recognition fails.
Last week, a man known as Mr. Wang tried to use his company’s facial recognition system to punch in for work. In recent years, biometric punch-in systems for offices have become ubiquitous along with other facial recognition applications such as payments, game limits for minors and even subway rides.
Unfortunately for Wang, the system not only failed to recognize him, but it also mistook his face for everyone else’s. In the video, the system repeatedly matches his face to those of his colleagues, both male and female.
His amusing encounter with the machine -- likely the result of a malfunction -- instantly became a joke on social media. Is it possible that Wang has one of those faces that looks way too common?
Online users also took the opportunity to ridicule their own companies’ systems. It appears facial recognition systems at work aren’t so popular.
“In our department, the male colleagues can scan their faces in place of the female ones, you tell me if it’s working,” said one commenter.
In recent years, the accuracy of facial recognition has improved by leaps and bounds. Last year’s study by the National Institute of Standards and Technology (NIST), an arm of the US Commerce Department, said certain algorithms are 20 times better at searching databases and finding matches than in 2014.
NIST is also one of the rare institutions that offers independent benchmarks for facial recognition systems. Some of the best-scoring companies in NIST tests are from China, including Yitu and Megvii.
However, NIST also notes that not everyone has made progress. The new algorithms do not all perform the same, and the best algorithms are far ahead of the pack, the study notes.
“Even well-performing algorithms struggle with certain naturally occurring challenges, like bad quality photos, aging faces or even the existence of twins,” the report says. Yes, even spotting twins is a problem.
Western companies have had more than a few examples of facial recognition failures. In one amusing incident, a Chinese woman discovered that her colleague could unlock her iPhone X with her face. The Metropolitan police around London also used facial recognition to find suspects, but it proved to be wrong an astonishing 98% of the time.
Another alarming mishap involved tech from Amazon. The tech giant’s facial recognition software matched 28 members of the US Congress with mug shots, and it disproportionately misidentified black and Latino lawmakers.
China has been rushing to put facial recognition systems everywhere, not just in offices and on the streets. It’s used to verify drivers for ride-hailing platform Didi and to check IDs at transportation hubs. It’s even used to verify whether kids make it to school on time.
However, Wang’s experience has made some question the technology’s limits.
The system at Wang’s office was offered by ZK Teco, which sells face and fingerprint scanners ranging from about US$30 to more than US$450. Clearly not all facial recognition systems are created equal.
“This has made me indescribably concerned about using Alipay’s Smile to Pay,” someone commented on Weibo, referring to a popular mobile payment system that has an option for facial recognition payments.
At least with Wang’s mishap, the malfunctioning facial recognition system was actually installed. In April this year, the New York Metropolitan Transportation Authority revealed that it’s not really using facial recognition to screen passengers on the subway, despite screens implying otherwise. Apparently, the screens were there to scare passengers into paying their fares.
In that sense, maybe the system at Wang’s office was working as intended: Getting people to work on time.