Everyone’s looking for ways to stay safer now, and many are looking to technology for help. Facial recognition is one of the tools that government agencies and businesses are exploring—both to slow the spread of the new coronavirus and to protect data from cybercriminals trying to profit from pandemic-related disruption. However, it’s not always immediately clear what facial recognition can do reliably now, what it can’t do and what it can do when people are wearing masks, sunglasses and other items that obscure the face.
For example, facial recognition has become a biometric marker for some digital payment methods like Apple Pay to authenticate users. And many companies are looking for ways to use facial recognition to reduce the need to touch surfaces like payment terminals and locks with fingerprint scanners. But facial recognition systems that consistently recognize all kinds of people correctly in all kinds of settings have yet to be brought to market, let alone when those people are wearing many types of face masks to protect their health.
Before planning a safety or security program that includes facial recognition, it’s important to understand which applications are fully fledged, which are still developing and where the potential pitfalls are.
What facial recognition can do now
There are some situations where facial recognition can be a practical safety and security tool. One is contactless data and payment security for accountholders. For example, Apple users can unlock their phones and make purchases through Apple Pay with the company’s FaceID biometric tool. And FaceID appears to be moving beyond the Apple-only silo. Google Drive now recognizes Apple’s biometrics, including FaceID, for logins to its iOS app.
As account takeover fraud trends upward, facial recognition could be a useful tool to fight it, although it’s not without challenges and security issues. For example, what happens if someone’s image is used to commit fraud or another crime? Passwords can be changed. Faces, for the most part, cannot.
Although masks are required or recommended in many workplaces, it can be easy for workers who aren’t yet used to them to forget, especially if they’re not in a public-facing role. Facial recognition can help businesses maintain safety in areas where masks are required, such as in warehouses and on factory production lines. This capability requires mask-recognition software that’s written to alert managers when it detects an unmasked face.
Although it’s a controversial application, facial recognition can help police departments identify people who are wanted for crimes or have been reported missing. That’s the use case London’s Metropolitan Police department cited when it switched on live facial recognition systems in January. More widespread systems that were already in use in China and Russia are now helping authorities enforce regional lockdowns and quarantines of specific individuals. For example, in China, facial recognition cameras with thermal sensors can identify citizens who are running a fever. They can also spot people who aren’t wearing their required face mask.
What facial recognition can’t do yet
Because facial recognition maps many points on a face to get a match, masks, glasses and other items that cover part of the face can affect accuracy. For example, Apple is adding an automatic changeover to passcode entry for FaceID users when it detects a mask. This may seem like a step backward in terms of security, but it helps users maintain safety by leaving their mask in place instead of touching and possibly contaminating it. Especially in densely populated cities and high-risk settings like hospitals and nursing homes, the fewer times users have to touch their masks or faces, the safer they are.
Although some facial recognition providers—and some governments—say the tools they use work even when subjects are masked, it’s clear that’s not always the case. One Chinese facial recognition vendor claims a 95% accuracy rate on masked subjects, compared to 99.5% for unmasked people. However, the combination of a mask and sunglasses thwarts the system.
That lack of consistent performance has caused problems with everyday tasks for people in China who are wearing masks. Some have reported problems with everything from logging into their smartphones and bank accounts to getting into their apartment buildings. Clearly, even in countries that put a high premium on facial recognition accuracy, the tools aren’t perfect. That can lead to misidentification by employers and law enforcement, which can cause real problems for people.
Issues that facial recognition needs to address
In addition to accuracy concerns, researchers have found that facial recognition AI can be biased, skewed by data inputs that don’t reflect the full range of human identity. Studies have found that such systems are more likely to misidentify transgender and nonbinary people, women and people of color than white men.
Facial recognition technology also raises ethical quandaries related to surveillance and legal issues related to privacy. Privacy watchdog groups and ethicists are concerned that increased use of facial recognition now may lead to increased surveillance after the pandemic crisis has passed.
However, facial recognition may also face legal barriers to wide adoption, especially in Western countries. For example, some legal experts say that in the U.S., a patchwork of “inconsistent and complex” local and state laws about the use of biometrics could slow down its implementation at scale.
For now, it’s clear that facial recognition can help protect individual and public health in limited scenarios, like making a contactless payment, accessing buildings without touching a keypad or scanner and enforcing mask policies in the workplace. However, facial recognition right now is not a comprehensive solution for going contactless or for enforcing public health rules, especially when people are masked to protect themselves and others from Covid-19. Instead, facial recognition is one tool among many that can deliver safety benefits if it’s used responsibly.