ARUN RATH, HOST:
Someone intentionally took the plane off course. That is the consensus of investigators still struggling to understand what happened to Malaysia Airlines flight 370.
Today, the Malaysian defense minister announced the authorities were refocusing their investigation on all crew and passengers. Last week, Interpol confirmed that two passengers aboard the flight were traveling on stolen passports. While there's no evidence that those passengers are behind the plane's disappearance, aviation experts say it highlights a major security gap at many airports. It is still possible to board an airplane with a fake photo ID.
How could that be the case? A new study of facial recognition finds that people are shockingly bad at picking out fake photo IDs.
Megan Papesh is one of the authors of that study.
MEGAN PAPESH: When you are tasked with matching somebody to a photo ID that had been taken days, months, even years prior, all of the many changes that occur in people - gain weight, lose weight, shave, put on glasses - you know, myriad changes occur, and that makes people really willing to accept a lot of changes.
RATH: So how did you do the study? How did you take a look at this?
PAPESH: We took pictures of students at Arizona State University and got their permission to download their student ID photos. And then at Louisiana State University, we presented those photos to students and asked them to make - basically match mismatched decisions to those where we manipulated how frequently we showed them IDs that mismatched. So those are the critical ones. Those are the bad guys.
RATH: Mm-hmm. And how often did they get it wrong?
PAPESH: Well, when they see the mismatches frequently - so that's 50 percent of the time - they would get it wrong about 20 percent of the time. So they would be about 80 percent accurate. But when they only saw the mismatches 10 percent of the time, their error rates actually skyrocketed. They went up to 45 percent.
RATH: Just for some real-world examples, say, if you're a bouncer at a bar, you're checking IDs, are you more or less likely to get those right compared to other situations?
PAPESH: That situation's actually really interesting. So a lot of my research assistants have told me that bouncers at clubs are really good at spotting fake IDs despite the motivation to let people in and sell alcohol to them because they encounter so many of them, which is part of why we're interested in providing training to individuals who are tasked with doing this in more security contexts with those bursts of fake IDs, kind of like they would get if they were bouncers at a club.
RATH: So TSA agents could learn something from the bouncer community.
PAPESH: I hesitate to put them on that spot, but it's possible.
RATH: Potentially. And in terms of the ID itself, is there anything that could be done with that or photographic techniques that could make them less fallible?
PAPESH: Well, I hear for passport photos, you have to put your hair behind your ears if you have long hair because most people don't change their ears. But driver's licenses don't have those sort of constraints. But there are some researchers in the U.K. that have found that having multiple photos of the same person helps people identify faces to photographs.
RATH: You know, I know that we know from computer modeling that facial recognition is extremely complicated. And your work would seem to indicate that humans actually aren't as good at it as we might have thought we were. Is this simply a bad way to identify people?
PAPESH: Unfortunately, it's simultaneously the best and the worst way that we have. Computer face recognition software is making tremendous strides. But unfortunately, that software doesn't perform at the level of humans yet, so it can't be deployed in applied contexts. And barring anything much more invasive, like retinal scans or thumbprint ID, face matching is really the best way to go without being too terribly invasive.
RATH: Megan Papesh's most recent study is about the reliability of photo ID recognition. Megan, thank you.
PAPESH: Thank you very much for having me.
(SOUNDBITE OF MUSIC)
RATH: This is NPR News.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.