A surprising result from an experiment at the National Institute of Standards and Technology.
A first-of-its-kind experiment by the National Institute of Standards and Technology (NIST) recently aimed to find out if forensic facial examiners recognize faces better than those who have not been trained in the field. The results confirm that they are not only superior in their task to those with no training, but that they outperform computers running advanced facial recognition algorithms.
The experiment put three groups of people to the test: professional facial examiners actively working in the field, non-examiners otherwise attached to the field, and assorted students with no experience. Each group compared pairs of faces that were specifically chosen for being difficult for computers to match, and then indicated if they believed the pictures were of the same person.
“All of these photographs were beyond what a computer could identify,” Evelyn Brown of NIST told the Daily Dot over the phone.
Despite the built-in difficulty, the human examiners posted a near-perfect score of 99.7 percent. The findings provide “the first strong evidence that facial forensic examiners are better at face recognition than the rest of us,” Jonathon Phillips, a face recognition researcher at NIST, said in a release.
This is important because forensic facial examiners are often called upon to testify in court whether two pictures are of the same person or not. It’s testimony that can have heavy consequences for the accused. Though forensic facial recognition methods have not been put to the test before in this way, this experiment confirms the science behind them is sound.
Brown tells us now that the examiners have been tested on an individual basis, future experimentation will let them get other examiners’ input on matching faces. They will be able to “talk to each other, ask opinions, and collaborate.”
From “being creative” to “identifying social cues,” the list of tasks which people can outperform computers at grows ever so slightly.