Imagine that you are a bouncer, checking IDs outside a popular bar in a college town. It is somewhat dark outside the door, there are many distractions: loud music is playing and your job requires you to also keep an eye on the crowd for trouble. And because the patrons are dressed for a night out, many of them look somewhat different than their ID photos. Despite all these challenges, intuition probably tells you that matching faces to ID photos is easy and accurate. Look at the photo, look at the person, and they either match or not. It turns out, however, that this intuition is wrong. Detecting false IDs is surprisingly difficult, especially when they rarely occur. A bouncer for a college bar can likely expect to catch roughly a dozen fake IDs in an evening, and the cost for missing one is relatively low: an underage student sneaks into a bar, and the bar makes more money.
Now imagine that you are screening IDs for airport security. Again, you must simultaneously verify IDs while keeping an eye on the crowd for suspicious activity, and there is time pressure to keep the line moving. Moreover, travelers vary widely in age and appearance; their IDs and passports are from all around the world; and there can be great differences between a person’s photo and their current appearance. Most importantly, only a very rare person would attempt to board an airplane with a false ID, and the consequences of missing that person could be dire. With the recent disappearance of Malaysian Airlines Flight 370, and reports that two men boarded the plane using stolen passports (although they were subsequently ruled out as potential terrorists), attention has become focused on this potential security loophole. In many airports around the world, there are several checks in place to prevent individuals with stolen IDs from passing through security, including scanning passengers’ passports against the Interpol database of known missing/stolen documents. Approximately 3.1 billion people traveled via airplane in 2013, yet Interpol estimates that passengers successfully boarded airplanes 1 billion times without having their passports scanned against their database. Although this oversight may reflect negligence or lack of available technology, it underscores the need to understand the second line of defense against stolen identity documents: the human ability to match faces to photographs. Given the myriad contexts in which society relies on face matching, it is surprising to learn that decades of research have documented its remarkable fallibility.
People often have the intuitive impression that they (and others) are “expert” face processors, given the social relevance and ubiquity of face and expression perception. In many domains, this intuition is correct. Humans are capable of recognizing hundreds of individuals across years, and under varying environmental conditions. However, this impressive ability is generally limited to familiar faces. In 2011, researchers asked U.K. and Dutch participants to sort 40 photographs of individuals into piles so that each pile contained pictures of the same person. The 40 photographs depicted only two individuals, both Dutch celebrities, and, in fact, almost all of the Dutch participants created only two piles. By contrast, the U.K. participants created an average of 7.5 separate piles. Without familiarity to aide their processing, the U.K. participants perceived the faces as representing far more unique identities.
Other research has focused on unfamiliar face matching. Although there are certainly situations in which an observer must match a familiar face to his photo ID–for instance, a frequent flyer or familiar face at a neighborhood bar or liquor store–the majority of people passing through security lines or other age and identity checkpoints are likely to be unfamiliar to the person checking their documents. Under these circumstances, a premium is placed on catching the “fakes.” Although it is not ideal to inconvenience someone by closely scrutinizing their ID, the consequences of missing a stolen ID are severe. Unfortunately, laboratory research has revealed that this task is remarkably error-prone. Under idealized conditions, with just two faces to compare, almost 20 percent of identity mismatches go undetected, according to research published in 2008. Performance drops even further when the observer compares faces of other-race individuals, extending the well-known own-race bias in face recognition to perceptual tasks that place little burden on memory systems.
But how well do such laboratory studies mimic real life? To approximate real-life conditions, researchers sent shoppers through a supermarket carrying credit cards with identifying photos on the front. When the shoppers attempted to pay with their assigned cards, they were unaware of whether the credit card photo depicted them, or a different individual, matched for age and gender. Although all cashiers were aware of the ongoing study, and were warned to watch out, they accepted nearly 50 percent of the fraudulent cards.
Error rates exceeding 20 percent are harmless in the lab, but they can have severe consequences in applied settings. One difficulty in comparing lab studies with applied contexts is the rate at which observers encounter fake IDs. In most laboratory studies, observers encounter 50 percent identity matches and 50 percent identity mismatches. While it is possible for a liquor retailer to encounter frequent fake IDs (particularly in small college towns with not much else to do!), one can likely assume that very few individuals present fake or stolen IDs when traveling through the airport or crossing national borders. Although this sounds like a good thing, there is solid evidence to suspect that these contextual statistics will have a powerful (and detrimental) influence on an individual’s ability to detect identity mismatches.
In visual search tasks such as baggage screening in which targets occur rarely, people become less likely to notice them. This is known as the low-prevalence effect, and it often manifests as a conservative bias in decision-making. That is, individuals become less “willing” to report a target when those targets appear infrequently. The same finding – but far stronger – emerges in unfamiliar face matching. In a study published earlier this year, I along with Stephen Goldinger at Arizona State University found that when people rarely encountered false IDs–in our study, they were looking at two pictures depicting the same or different individuals–they missed 45 percent of those that occurred. That is, in those instances, they thought the two photos were of the same person when they were not. This error resisted many attempts to reduce it: we asked observers to make certainty judgments and even gave them a second chance to view some face pairs. Thus, face matching is strongly affected by viewers’ expectations. If someone does not expect to encounter a fake ID, that person will be less likely to detect fake IDs. The consequences of these biases, coupled with the inherently challenging nature of unfamiliar face matching, suggest that photo-ID matching is far more challenging (and unsuccessful) than we might care to believe.
Bahrick, H. P., Bahrick, P. O., & Wittlinger, R. P. (1975). Fifty years of memory for names and faces: A cross-sectional approach. Journal of Experiment Psychology: General, 104, 54-75.
Hancock, P. J. B., Bruce, V., & Burton, A. M. (2000). Recognition of unfamiliar faces. Trends in Cognitive Sciences, 4, 330-337. http://dx.doi.org/10.1016/S1364-6613(00)01519-9
Jenkins, R., White, D., Van Montfort, X., & Burton, A. M. (2011). Variability in photos of the same face. Cognition, 121, 313-323. http://www.sciencedirect.com/science/article/pii/S0010027711002022
Kemp, R., Towell, N. & Pike, G. (1997). When seeing should not be believing: Photographs, credit cards, and fraud. Applied Cognitive Psychology, 11, 211-222. DOI: 10.1002/(SICI)1099-0720(199706)11:3<211::AID-ACP430>3.0.CO;2-O
Megreya, A. M. & Burton, A. M. (2006). Unfamiliar faces aren’t faces: Evidence from a matching task. Memory & Cognition, 34, 865-876. http://www.ncbi.nlm.nih.gov/pubmed/17063917
Megreya, A. M. & Burton, A. M. (2007). Hits and false positives in face matching: A familiarity based dissociation. Perception & Psychophysics, 69, 1175-1184. http://www.ncbi.nlm.nih.gov/pubmed/18038955
Megreya, A. M. & Burton, A. M. (2008). Matching faces to photographs: Poor performance in eyewitness memory (without the memory). Journal of Experimental Psychology: Applied, 14, 364-372. http://www.ncbi.nlm.nih.gov/pubmed/19102619
Papesh, M. H., & Goldinger, S. D. (2014). Infrequent identity mismatches are frequently undetected. Attention, Perception, & Psychophysics. doi: 10.3758/s13414-014-0630-6
Wolfe, J.M., Horowitz, T.S., & Kenner, N.M. (2005). Rare items are often missed in visual searches. Nature, 435, 439-440. http://www.ncbi.nlm.nih.gov/pubmed/15917795