In A Nutshell
We’re horrible at matching people with their photos, especially when it’s someone we don’t know. A study from Louisiana State University and Arizona State University found that people who were asked to look at a subject and then identify whether a photo ID was authentic or not could be wrong as much as 40 percent of the time. (Non-matching people and photos were presented to them for 10 percent of the study.) Passport officers fared no better, accepting up to 14 percent of fraudulent photos as the real deal. Thankfully, technological advances might soon produce computers that can do the job better than we can.
The Whole Bushel
Chances are good that you knew someone who had a fake ID when you were younger. Whether it was a sibling, a college roommate, or a friend, there’s also a pretty good chance that you thought the picture on the ID looked nothing like the person trying to use said ID.
There’s no way anyone would fall for that . . . right?
Somewhere around the fifth time it worked for getting into a bar or buying alcohol, you might have gotten the sneaking suspicion that the people who were doing the carding just didn’t care.
That might not be the case at all, though! It seems that people really do have a sort of mental short-circuit when it comes to spotting fake photo IDs. It might have more to do with the difficulty we have in processing unfamiliar faces than with the quality of the photo.
When researchers from Louisiana State University and Arizona State University tasked study subjects with looking at 200 people and their student ID photos (some of which had been taken up to seven years prior to the project), all they needed to do was decide whether or not the person matched the face in the photo.
When non-matching faces were presented to about half the pairs, subjects misidentified about 20 percent of them. When the pairs were non-matching in only 10 percent of the pairs, the misidentification rate jumped to around 40 percent—even when subjects were given more time and told to correct errors.
Sure, but that’s just in an informal setting with people whose lives might not be depending on that. And people who compare faces for a living must do better. Right?
Another study tested the abilities of passport officers to pick out non-matching faces, and they still did quite badly. In this study, employees of the Sydney Passport Office and the Department of Foreign Affairs and Trade (a total of 49 people with experience ranging from eight years to seven months, all with special training in facial recognition) were given a similar test.
They were tasked with comparing real people to photos and with making photo-to-photo comparisons in order to decide whether or not they were being presented with fraudulent photos. In some cases, the photos in the latter exercise were even taken on the same day.
For the portion of the experiment done with people standing in front of the officers, about 10 percent of the responses were wrong. Of the non-matching pairs, 14 percent of fake photos were accepted as real by the officers. While some individuals did better and some did worse, there was no correlation between experience on the job, training, and accuracy. Again, these photos were all taken very recently, within a matter of days.
In the photo-to-photo portion of the experiment, photos could be up to two years old. Overall, the officers only scored a 70.9 percent.
So maybe humans aren’t fantastic at this task. But we’re already seeing computers that might be able to rival human abilities when it comes to quick-fire identification.
Tech gurus at Carnegie Mellon have developed an iPhone app capable of snapping someone’s picture and then running it through facial recognition software, picking the right match out of a pool of 25,000 photographs in less then three seconds. Unfortunately, it’s only accurate 31 percent of the time. Just needs practice!
Show Me The Proof
PLOS One: Passport Officers’ Errors in Face Matching
Louisiana State University: LSU research shows face matching for passports and IDs incredibly fallible
cnet: Face-matching with Facebook profiles: How it was done