Hal Hodson technology correspondent, The Economist
The face is becoming machine-readable
2019 in brief
Le Grand K (otherwise known as the International Prototype of the Kilogram), the lump of metal that is the standard by which the kilogram is measured, is retired from its French vault in May. It is replaced by a more precise definition
CLOSED-CIRCUIT television (CCTV) is a familiar feature of urban life. Public squares, street corners and transit systems are adorned with cameras in the name of fighting crime. Large cities contain thousands of them, producing gargantuan volumes of footage. Most of it is grainy, uneventful and unwatched.
The latest advances in machine learning have created software that can determine the unique pattern of a person’s face from imagery or video to a far higher degree of accuracy than older technology. This makes it possible to track large numbers of people automatically as they move through public space, something that would be impossibly expensive if done with human labour. Recognition software turns faces into barcodes. Cameras become scanners, automatically cataloguing the people who walk past their lenses.
Police forces are not using the latest technologies widely yet, and their facial-recognition systems are notoriously inaccurate. Big Brother Watch, a privacy group, has observed the South Wales Police in Britain using facial recognition at public events over the past two years. The group recorded an inaccuracy rate of 91%, meaning that only nine out of every hundred people tagged by the system were actually who the system supposed them to be.
This state of affairs will not last long. The world’s CCTV and police cameras are one upgrade cycle away from capturing higher-definition imagery, which will help facial-recognition algorithms work better. Camera systems can be designed with facial recognition specifically in mind, using two or more cameras to capture faces from multiple angles, making them easier to identify. Silicon Valley’s approach to facial recognition, using powerful computers and large datasets of faces to train highly accurate software, is only beginning to percolate into the security market. That will speed up in 2019.
No hiding place
The implications are worrying. States have the greatest potential to abuse accurate facial recognition, as they tend to control the infrastructure by which imagery of faces can be recorded in public. Authorities might, for example, capture the faceprints of a group of protesters, arresting any that can be matched with a criminal record, while simply filing the rest away as a list of suspicious faces. Facial recognition is open to abuse by private companies, too. Shops would be able to create shared databases of customers and the products they looked at, just as online tracking does now, breaching privacy.
Facial recognition also has the potential to merge the tracking that happens in the digital and physical realms. It turns the face into an address that links behaviour in the real world with online profiles, and vice versa. The combination of web tracking and physical biometrics like facial recognition will mean that the spaces in which human beings are not tracked will shrink in 2019.
In America, for example, Major League Baseball will start allowing fans to validate their tickets and enter stadiums via a scan of their face, rather than a paper stub. Singapore’s newest megamall will use the technology to track shoppers and recommend deals to them. Tokyo will spend the year installing facial-recognition systems in preparation for the Olympics in 2020, when it will use the technology to make sure that only authorised persons enter secure areas. Airports from Atlanta to Bengaluru will use it to track passengers through security and immigration. Research from SITA, a technology vendor, suggests that three-quarters of all airports and airlines are investing in the technology or carrying out research on the topic. Samsung, one of Apple’s main rivals in the smartphone market, will introduce 3D facial recognition on its models in 2019, aping the iPhone’s Face ID.
Imagining the specific harms that new technologies create can be hard. Few would have thought, in the early days of online advertising, that it would lead to a web that is designed with such a thirst for human attention that the most hyperbolic accounts of events travel farther and have greater impact than those with nuance and accuracy. So it is worth worrying about facial recognition for the principles it offends, and the damage it threatens.
Humans need spaces where their movements are not tracked, where they are free to assemble with whomever they please. A world with ubiquitous facial recognition means one in which no coffee meeting, no midnight walk, no trip to the shop can occur without being assigned to a specific face and identity. Strong laws that protect individual rights are the best hope for limiting such tracking. The Pandora’s box of facial recognition is open, but it is still possible to influence its impacts.