On 28 August 2014, 14-year-old Alice Gross went missing in West London. She was last seen on CCTV walking along a canal towpath in the summer rain, toting her black backpack. London’s Metropolitan Police deployed their biggest search party since the 2005 London bombings – a team of 600 officers spread across eight forces – to find her. But it was only after using the CCTV to piece together the events that they discovered the teenager’s body, nearly five weeks later, in the River Brent. They also zeroed in on a suspect: Latvian construction worker Arnis Zalkalns.
Throughout the operation, a special team of 10 officers in the Metropolitan Police were testing out their skills formally for the very first time. They were called the super-recognisers. “The attitude used to be anyone with two eyes can view CCTV, but not everyone sees things in the same way,” says Ch Det Insp Mick Neville, head of the Central Forensic Image Team. This elite group viewed thousands of hours of grainy, low-quality footage and, within days, identified both the victim and the as-yet-unknown suspect Zalkalns. They also were able to map out their movements precisely enough to draw a timeline and help conclude the case. Their special skill? A super-human ability to instantly recognise faces they barely know.
The Metropolitan Police's super-recognisers combed through CCTV images like this one, which shows Alice Gross on the day she disappeared (Credit: REX Shutterstock)
Scientists are only just beginning to understand why some people have the skill and how it works. Progress is being made in identifying who possesses it, though – and there is even an online test anyone can take to see if they might qualify.
If you often recognise people out of context, that's an indicator
“If you have experiences where you often recognise people out of context, that's an indicator. If you’re more likely to recognise someone else than they are to recognise you, then that's an indicator,” says psychologist Richard Russell of Gettysburg College in Philadelphia, who first coined the term ‘super-recogniser’ in a paper he published in 2009. “Many describe not being sure whether to go up to someone – often it’s someone they knew only incidentally, and they fear they may come across as a stalker. Some report just faking not to know people.”
Russell became curious about super-recognisers in 2006, when he was at Harvard University studying prosopagnosics: people with very poor ability to recognise faces. He discovered it was a far more common affliction than he expected; about 2% of people he tested fell somewhere on the low end of the spectrum. “So I thought that suggested there were people on the other end of the scale too – with extraordinary abilities,” he says.
When he started looking, he found super-recognisers across the United States. One of his subjects, Jennifer Jarett, is a 44-year-old police misconduct investigator in New York City. Her first memory of her talent was when she was 15 and on a family vacation in Hawaii. “I spotted a man sitting a few rows ahead on our plane. I told my family he was famous and had been on a tonne of TV shows, like Murder She Wrote and the Bionic Woman. They just laughed at me because no one recognised him,” she says. Later that summer, the man she had spotted, Granville Van Dusen, played a bit part in the show Family Ties. “So it became a family joke. That's when they realised I was unimpeachable when I recognised someone,” she laughs.
Would you recognise any of these faces again? Fans enjoy the music at the Glastonbury Festival (Credit: Leon Neal/AFP/Getty Images)
I wanted to deal with image recognition in the same way as fingerprints or DNA
Around the same time that Russell was testing volunteers like Jarett, Ch Det Insp Neville – a broad-shouldered man who speaks in short, sharp sentences barked like orders – began noticing that the same few officers from across London’s police force were making correct criminal identifications from bad-quality images, over and over again. “I wanted a psychologist to test the limits of their abilities and find how their minds work best,” Ch Det Insp Neville says. “I wanted to deal with image recognition in the same systematic way as fingerprints or DNA.” So he hooked up with Josh Davis, a psychologist at the University of Greenwich.
Davis was at first very sceptical of Ch Det Insp Neville’s claim that the police force had super-recognisers that could pick out criminals from grainy CCTV. “My PhD research suggested that people giving evidence in court were generally terrible at even matching two faces, so I was very dubious, but agreed to test them out,” Davis says. His specially designed tests included a famous faces test, using images of B-list celebrities taken 12 years previously and distorted to look like black-and-white negatives; there was also an unfamiliar faces test which gave officers a few minutes to memorise six faces, and then asked them to pick them out of 100 photo lineups that became more and more pixelated. “I’ve tested about 250 police officers in total now, and around eight of them scored exceptionally highly,” Davis says. Dozens of others scored well above average.
Soon after the testing began in 2011, riots broke out across London; hours of crime footage suddenly needed to be analysed quickly. The London riots were a turning point for Ch Det Insp Neville’s super-recogniser programme. Twenty super-recognisers, picked out by Davis’s testing, trawled through approximately 5,000 images. They identified 609 suspects, 65% of whom ended up in court.
Malaysian student Ashraf Haziq is mugged by youths at London's 2011 riots (Credit: REX Shutterstock)
Ch Det Insp Neville and Davis now meet with international police forces from China to Canada who want help in pinpointing their own super-recognisers and streamlining the image identification process in a similar way to the London police. “The Met Police have been at the forefront of using the ability. I don't know of any other police entity who has tried to use super recognisers in quite the same way,” says Russell.
The Metropolitan Police’s top talent, PC Gary Collins, is a mild-mannered, unassuming police officer in Hackney’s Gang Unit. “I’d never heard the term super-recognisers until the riots, but I had always been picking out suspects from photos that were printed off and circulated or CCTV footage,” he says. “I’ve always enjoyed art, and I used to work in graphic design before the police, so maybe it has something to do with attention to detail or pattern recognition. Who knows.”
I was positive it was him, I knew straightaway from his eyes
His most memorable case, PC Collins says, was identifying a man called Stephen Prince who, during the riots, had broken into shops, robbed reporters of their cameras, stolen bikes, thrown petrol bombs at police and set fire to cars. In the riot footage, Prince had covered his face up with a red bandana and had a black woollen hat pulled low over his forehead. All you could see were his eyes. PC Collins and others spent days trying to track the man through the footage, searching for a moment when he pulled the bandana off, but there were no better shots. Still, PC Collins recognised him from the images in which he was heavily disguised.
“The last time I’d seen Prince was about six years earlier, but I was positive it was him. I knew it straightaway from his eyes. So we went to court,” PC Collins says. Prince was eventually found guilty and sentenced to six years in prison – one of the longest riot-related sentences.
Police face off against a mob after a number of cars are set on fire in Hackney, north London during the British capital's riots(Credit: AFP/Getty Images)
While scientists are still unclear what’s going on in the brains of recognisers like PC Collins, they do know that most facial recognition is processed the brain’s fusiform gyrus – a long, thin area in both the temporal and occipital lobes which also processes colour. Abnormalities in this region are associated with face blindness – the inability to recognise faces – and facial hallucinations.
Evolutionary psychologists are particularly intrigued by super-recognisers, because faces provide more social clues than just identity – they relate to our understanding of the world. Newborn infants appear able to distinguish their mother from other women after two days, and some six-month-old babies show the capacity to recognise other faces. Those who are good at facial recognition tend to be extroverts and can establish trust more quickly.
Some six-month-old babies show the capacity to recognise other faces (Credit: Alamy)
Compared to other visual memory tasks, the brain uses a higher level of processing for face memory
In November 2011 Ash Jansari, a psychologist at the University of East London, conducted one of the largest ever studies of super-recognisers by recruiting more than 700 visitors to the Science Museum in London, ranging in age from 6 to 74. He used the same test that Russell used in his Harvard study – the Cambridge Face Memory Test. Only seven people scored two standard deviations above the average – the ‘super-recognition’ criterion – which suggested that roughly 1% of the population may be super-recognisers. Studies also found that they aren’t any better than average people at recognising things that aren’t faces, like flowers or chairs. “This suggests the brain uses a higher level of processing for face memory, compared to other visual memory tasks,” Jansari says. The research also showed that all human beings process faces in the same way – as a holistic unit, rather than as a collection of individual features.
I took the Cambridge Face memory test in Davis’s lab to see where I stood. I tried to memorise standout features to try and identify in each lineup, like a crooked nose or bushy eyebrows; my score hovered right around average. A week later, I did a second test that Davis has put together – and which anyone can try out online. This time I approached each face as a unit, memorising my general impressions of each face, and the attitudes and expressions, rather than a list of features. My score on this was 94%.
Another of the Metropolitan Police’s consistent high scorers on these tests is detention officer Idris Bada, who has a booming laugh and a low, gravelly voice. “I’m not out on the streets like Gary. The prisoners come to me from all over the city to be booked in,” he says. In the last year alone, he’s made 300 identifications – mostly from studying photos and connecting prisoners brought in for one crime to a string of unsolved cases. “Sometimes it’s something distinctive about their face, like a deformed ear or a scar that sticks with you,” he says. “Or sometimes when I wake them up and they turn over, their body language makes my head go ding ding ding ding! By the time they’ve woken up I’ve linked them to five or six different crimes.”
Your face is a gauge of your trustworthiness
Bada describes his recognition process as a pinhole camera effect. “It’s like looking at a negative image and then closing your eyes and looking at white paper, it slowly starts developing. That’s how it works for me,” he says. An ex-electrical engineer with a penchant for circuit diagrams, Bada puts his ability down to his curiosity about people. “Everybody’s face tells a unique story, and I want to understand who you are when you’re in custody with me. Your face is a gauge of your trustworthiness.”
Police say that charges would have been brought against Alice Gross murder suspect Arnis Zalkalns, shown here on CCTV, if he had lived (Credit: Photo by REX Shutterstock)
At Davis’s lab, the data from Bada and the other police officers’ tests is currently being collated to create a face recognition test for new recruits of the Metropolitan police force. The standalone test will give an instant score, identifying which new recruits could be added to the super-recogniser list that currently numbers 140 police officers, so they can be called upon for special task forces like the Alice Gross murder case.
Davis is also working with an international team of researchers, police forces and companies like Huawei on the Large Scale Information Exploitation of Forensic Data (LASIE) project, which will design a centralised IT system that automates initial analyses of digital forensic evidence including images, audio and video. The goal: to design algorithms that can filter diverse data from CCTV surveillance, mobile phones and social networks, making it easier for workers like Bada to sift through the evidence. “So if someone is wearing a distinctive T-shirt with a logo in a video frame, the algorithm could search for that logo elsewhere and bring the images together,” Davis says.
Humans will always be far better at identifying faces than computers
Should we be disturbed that this process is becoming increasingly automated, writing human empathy and morality out of the equation? Kelly Gates at the University of California, San Diego, who studies automated facial recognition in police investigations, says no. She thinks we should be just as wary of human super-recognisers as robotic ones. “Humans will always be far better at identifying faces than computers, so increasingly police forces are forming human-machine configurations to maximise their abilities,” she says. “I don't think that we should be less alarmed by computers doing it versus humans. It's this integration of humans and computers to incentivise surveillance practices that we should pay attention to.” Her big idea: if humans are programmed to behave like computers by regularly identifying faces from photos, then there should be laws and regulations governing their work.
Computed tomography (CT) scans show brain activity in the cortical and subcortical areas, activated during the recognition of faces and objects (Credit: Science Photo Library)
Even so, super-recognition could have unexpectedly useful consequences. For instance, eyewitness testimonies are notoriously unreliable, but often used in court. “You could use a witness’s face recognition score to either counter their testimony, or give it greater weight,” suggests Russell. It could also help scientists learn more about people with face blindness, by figuring out how face memories are made, stored and recalled.
And it’s not just good for science – super-recognition even could help your love life. New Yorker Jarett says, “I did some internet dating and emailed with a guy I never met. A couple of years later, he was on the street and I happened to give him directions.” When she got home, she dug out his last email and wrote to him, saying she had recognised him in the street because of her strange scientific ability. “We ended up going out on a couple of dates and he had told his friends about me,” she laughs. “Maybe he thought it would be a cute meeting story, but it didn’t quite work out.”
For victims of violent crimes like Alice Gross, the Met’s team of super-recognisers have become key to the policing process. After the suspect had been identified by the team, police were able to track his movements on the days around Alice’s murder. Although he was later found hanged, police confirmed that charges would have been brought against him had he lived. “There are about 50,000 officers in the Met, so we’d expect about 500 super-recognisers,” says Ch Det Insp Neville. “We want to know who they are. That’s our goal. To find them and use them effectively.”
This story is a part of BBC Britain – a new series focused on exploring this extraordinary island, one story at a time. Readers outside of the UK can see every BBC Britain story by heading to the Britain homepage; you also can see our latest stories by following us on Facebook and Twitter.