Artificial Intelligence

Most People Are Really Bad at Matching a Face to an ID Picture

It's routine whether you're going through airport security or hitting up your favorite bar: You have to show your ID. But how effective is that, really? Who's to say you're not carrying the ID of someone who just looks like you? It turns out that that's a very good question. Over and over, studies show that most people are lousy at matching pictures of unfamiliar faces, regardless of training. Yet we keep checking IDs just the same.

I'm Gonna Have To See Some Fake ID

Here's just one of many examples. In 2014, a research team out of Australia published a study looking into how passport officers compared to regular people when it came to matching pictures of unfamiliar faces. They had 30 passport officers perform three experiments. In the first, real people showed the officer an ID, some of which had the person's real image, some of which had an image of a person who looked a bit like them. Even though the circumstances were actually easier than the real world (the ID photos were taken just days before the experiment, and the fake IDs just had a roughly identical photo that probably wouldn't be good enough for someone who really wanted to fool an officer), the officers were lousy. They accepted 14 percent of the fakes and rejected six percent of the real IDs, for an overall error rate of 10 percent. That may sound small, but it adds up. When 100,000 people pass through an airport's security line on a given day, that's 10,000 ID-matching mistakes.

You might think that in the second experiment, where they just had to tell if two photos taken two years apart were of the same person, officers would do better. They didn't: Their error rate was nearly 20 percent. They also weren't any better than the general public: Their scores on a basic face-matching test were the same as population norms, and years on the job had no effect on performance in either of the experiments.

In a 2017 study, volunteers had to decide whether each of these pairs were of the same or different people. They accepted about 24 percent of incorrect matches.

more recent study in an August 2017 issue of the Journal of Experimental Psychology found that novices did perform slightly worse than trained police officers on a photo-matching test — but everyone still botched the job. Police officers in that study accepted 25 percent of fake IDs, compared to 26 percent by novices. That's nothing to brag about.

Want to see how you'd perform in a face-matching challenge? Take this test to find out if you're a "super-recognizer." (A score of 10/14 or more suggests you may have special skills. The test starts easy, but gets significantly harder by the end.)

What's The Solution?

This is one of many areas where artificial intelligence can come in handy. You may notice that Facebook and certain photo apps can identify your friends' faces in a snapshot with unsettling accuracy. Facebook, Google, and other researchers have all boasted artificial intelligence systems that perform better than humans at matching faces. What's more, many of those systems can go a step further by identifying who the person is from a photo. Artificial intelligence checking our IDs may smack of a dystopian sci-fi novel, but it's certainly more accurate than what we've got now. We, for one, welcome our new robot bar bouncers.

Get stories like this one in your inbox or your headphones: sign up for our daily email and subscribe to the Curiosity Daily podcast

Hear the heart-wrenching story of a woman who can't recognize faces — any faces — in "You Don't Look Like Anyone I Know" by Heather Sellers. The audiobook is free with an Audible trial. We handpick reading recommendations we think you may like. If you choose to make a purchase, Curiosity will get a share of the sale.

Written by Ashley Hamer September 13, 2017

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.