Facial recognition
Is the controversial tech the future of anything beyond unlocking our phones?
“First, these systems simply don’t work very well; second, they’re often biased against specific groups of people”
San Francisco is often seen as the epicentre of the future, thanks in no small part to the proximity of Silicon Valley. Given that, it may be a surprise to find out that the city has banned the use of facial recognition systems by police and other government authorities.
“I think part of San Francisco being the real and perceived headquarters for all things tech also comes with a responsibility for its local legislators,” said city supervisor Aaron Peskin, according to local media reports. “We have an outsize responsibility to
regulate the excesses of technology precisely because they are headquartered here.”
They’re not the only ones rethinking how such technology is used. The London Policing Ethics Panel studied Met Police trials of the technology at Notting Hill Carnival and elsewhere, raising questions about the system’s accuracy and bias. And in the US, Congress is considering the use of the technology after President Trump pushed for facial recognition to be rolled out for identifying passengers at airports.
What’s the problem?
There are privacy and surveillance concerns, says Joe Purshouse, lecturer in criminal law at the University of East Anglia. In theory, cameras can take a biometric scan of someone’s face without their consent or even knowledge. And that can cause a “chilling effect” on protesters – which isn’t good news for the state of our democracy, he added. “For example, if protesters or people taking industrial action know they are going to be identified by facial recognition, they may feel stigmatised or put off from exercising their democratic rights through a legitimate fear of how this systematic identification may disadvantage them in the future,” Purshouse said.
From a technical standpoint, there are two main criticisms: first, these systems simply don’t work very well;
second, they’re often biased against specific groups of people.
That latter problem in part stems from the fact that sets of training data tend to feature more images of white people, making such systems better at identifying white people than other races, leading to more misidentifications for other groups – a serious issue when the technology is used for arrests.
“This may explain some of the criticisms the Met Police faced over their use of facial recognition technology at the Notting Hill Carnival, an event with a high proportion of British African Caribbean attendees,” noted Kay Ritchie, a cognitive psychologist specialising in facial recognition at the University of Lincoln. At 2017’s event in West London, the police system made 35 incorrect matches, leading to one incorrect arrest and five people being stopped when they weren’t the criminals being sought, according to an observer from activist group Big Brother Watch.
Which brings us to the first problem. “It’s not there yet, as far as the science is concerned,” explained Martin Evison, professor of applied sciences at the University of Northumbria. “A lot of people worry about the threat to privacy and civil liberties… they might not need to worry too much in the sense that the systems are just not reliable enough to pick out any individual reliably in a crowd.” That’s partially because the systems are scanning in 2D, and with so much less information than a 3D scan, as found on the iPhone, it’s easy to make mistakes.
Is this the end of facial recognition?
Will such challenges mean the death of facial recognition? Not entirely. The technology is already becoming widespread elsewhere.
“In China it has been reported that this technology has been used as part of an enormous, unfettered state surveillance assemblage and has been used to target and detain Uighur Muslim populations,” said Purshouse. “Even in countries with stronger human rights protections, such as Australia and the UK, police forces have been using this technology without notifying the public – subjecting millions of people to secret surveillance, without proper trials to evaluate the success of the technology and without laws in place to regulate its use.”
But challenges in courts and protests by activists could slow its uptake in the UK and the US, he added. “Campaigners and lawmakers against its use have had a few successes recently, like in San Francisco. Other states may follow suit, and a UK court challenge may kick police facial recognition surveillance into the long grass in this jurisdiction, too,” Purshouse said. “However, more and more police forces across the world are beginning to expand their use of facial recognition. As ever with new surveillance technologies, the law makers are playing catch up.”
Richie agrees. “I suspect that even in the face of challenges like the San Francisco ban, we will continue to see an increasing use of facial recognition technology. So what we do need to think about is how to govern its use, and how to keep the public informed.”
Evison argues it depends on how the technology is used. Existing systems may not work so well for scanning large groups of people for specific individuals, but it could be reliable for scanning a database of arrest photographs. “If you’ve got a similar photograph of a suspect and want to see if he’s in your database, then the rigid comparison of good quality images in the same pose, angle and lighting should be fairly reliable,” he said. But he added: “I’d still want a person to check it, or some corroborating evidence.”
Regulating recognition
Having a human in the loop could help avoid mistakes and improve accuracy, making facial recognition less of a dystopian, allseeing surveillance system and more of a basic tool for police. Evison suggests police could use facial recognition to scan a crowd to look for a suspect on a watch list, flagging up potential sightings to be checked by a person familiar with the suspect. “It’s like you’re using the technology to sift,” he said, while conceding there may be better ways to spend policing budgets than on such systems.
The best way to regulate the technology depends on its use – after all, plenty of people willingly let their faces be scanned in order to unlock their iPhone or skip a queue at Border Control. Context matters, as does consent. “In the recent public surveillance trials in England, for example, the technology was being trialled with little consultation, and before we had a real understanding of its accuracy and discriminatory potential,” said Purshouse. “As a result, the trials drew controversy for discriminatory practices; failing to garner public trust, and failing to safeguard against basic oversights such as the use of poor quality images for watchlists.”
The bans and legal challenges suggest facial recognition is unique in innovation: the ethics are being considered before the technology is actually ready. Rather than killing off this idea, that might give us time to get facial recognition and its regulation right before it’s widely rolled out.
There are more subtle ways to confound facial recognition systems, although researchers have found that wearing scarves, fake beards or glasses doesn’t go far enough, as it leaves enough of the face clear for at least some algorithms to make a positive identification.
If you can’t cover your face, as the man with the fine found, and thick glasses and false moustaches aren’t enough, there’s also makeup. CV Dazzle is a technique developed by artist Adam Harvey, inspired by naval camouflage from the First World War that used cubist-style designs to make it harder to see the full size and orientation of battleships. Hair is cut into odd shapes and dyed bold colours, while thick makeup obscures key facial features, creating what Harvey has described as an “anti face”. Harvey has also created clothing with patterns that look, to computers at least, like faces so the cameras will be distracted. Other researchers have developed hats that project light onto the wearer’s face to confuse facial recognition algorithms.
Whether such tactics work depends on the facial recognition system being used, and right now we’re often not told that information – so perhaps try all three, and slap on some CV Dazzle makeup, a T-shirt covered in faces, and a hat with lights. It might not be subtle, but it may be less obvious than pulling your shirt up over your face as you walk by police.