Super recognisers – or people with uncanny abilities to match images of unknown faces with their real-life owners – are increasingly being relied upon by security agencies and in courts of law.
“I think it’s just critical that we know the limits of super recognisers’ abilities, whether they can perform that task and how likely they are to make errors.
“And where identity is crucial to a determination of a person’s innocence or guilt, false positive errors in those cases can have severe consequences.
“For example, the errors made at our borders have consequences for national security. And errors that are made in court by facial forensic examiners have the potential to cause miscarriages of justice.”
Many people are poor at recognising faces
Face recognition has been an important cognitive function throughout humans’ evolutionary history, White says. “We have to recognise friends and family while also needing to distinguish enemies from friends.”
As a result, recognition of familiar faces is something we are all generally good at. But recognition and memory of unfamiliar faces – crucial for the security of a modern society – is a different story, with most people being terrible at those tasks.
“A passport officer has to decide, does that person standing in front of me match their passport photo?”
“In the same way, police investigators might have a CCTV image of an unknown suspect and have to compare with a mug shot of a known suspect.
“It turns out that in those tasks, people are really quite poor,” says White.
And despite thinking that someone working in jobs that call for facial recognition skills on a daily basis might be good at these tasks, they also perform poorly.
“The surprising thing I discovered when I came to Australia and started working with the Australian Passport Office is that they make exactly the same proportion of errors as undergraduate students do,” White says.
“It was about one in five of the decisions they were making were actually wrong.”
Our brains aren’t wired for face matching
The problem stems not from a lack of training but because most people’s brains are not wired to perform unfamiliar face matching tasks.
“When looking at a new face for the first time, there’s not sufficient information there to identify the face accurately, or at least most people don’t know how to use that information.”
But in 2% of the population, and for reasons not entirely understood, identification and matching of unknown faces is – compared to the other 98% – a breeze.
Super recognisers appear to be not as hampered by the lack of information that thwarts the rest of us; there is something else going on at the cognitive level where they can match unknown faces in the same way that we recognise faces known to us.
As awareness of super recognisers has grown in the last decade, so too have efforts to recruit them by security and law enforcement agencies with a view towards minimising errors. And with advances in artificial intelligence now able to scan thousands of faces at speed, a super recogniser armed with this AI technology could be a formidable force for the identification of persons of interest.
Super recognisers are good – but not perfect
But while certainly impressive, they are not perfect. There are some conditions in facial recognition tasks where neither human nor machine are able to perform with 100% accuracy. Lighting, focus, resolution, head tilt, movement and elapsed time since an image was captured – all these can add doubt to making a positive match.
“One of our concerns, and the reason we’re doing most of the work we’re doing is to really try to understand super recognisers’ abilities,” White says. “The application of this scientific knowledge – knowing that these people exist and deploying them in these sort of surveillance or other type of scenarios – has got a little bit ahead of scientific knowledge.”
And the major concern is that, on the basis of performing exceptionally well on an online test for facial recognition, a super recogniser could stand up in court and make a judgment on a person’s identity that could be accepted all-too-readily as fact.
“At this point in time, I wouldn’t say there is sufficient scientific basis to support that type of use,” says White.
“We’re living at a time where we outsource decisions about identity to experts and technology without really questioning their accuracy.
“We would just like to see more acknowledgement of the limits of this expert knowledge, because someone’s life may well hang in the balance.”
This article was first published on Australia’s Science Channel, the original news platform of The Royal Institution of Australia.