New research finds that false identifications are going widely undetected because people lack the ability to tell the difference between a face and the picture on an ID.
Recent ASU graduate Megan Papesh teamed up with psychology professor Stephen Goldinger to complete a study, featured in Science Daily, the Huffington Post, National Public Radio and the Wall Street Journal, about the high level of error in matching faces to identification cards.
Goldinger and Papesh tested the “low-prevalence effect,” which refers to people’s tendency to not recognize things they don’t often see.
Goldinger used his experiment to explain how it works.
“In that research, people do exactly what the name suggests — you have some target in mind, a screen full of stuff is shown, and you try to quickly determine whether or not your target object is shown,” he said.
He explained that in more complex real-world situations, like baggage screening at the airport, people have to look for “anything abnormal.”
Goldinger said errors occur frequently if people are not presented with false information on a regular basis.
“If we created the experiment such that target objects appeared roughly half the time, you would likely find more than 90 percent of them, after searching each display for around one to four seconds,” he said. “It is unlikely that you would miss many targets — because you are getting ‘rewarded’ so frequently for staying vigilant.”
When people or objects are rarely presented, it may be harder to detect. Goldinger gave the example of seeing a familiar teacher. If the teacher has not been seen for a few years, the teacher may seem familiar but may go undetected as a former educator.
The research is relevant to many fields, he said. A bartender who sees many fake IDs would be good at detecting mismatched identifications because he or she sees many per day, whereas a TSA agent frequently misses items because it is rare that people try to to use false IDs to board flights.
Papesh said in an email that their study took a new approach to testing facial recognition.
“One unique thing about our study, relative to previous work, is that we used realistic variations in our photos,” she said. “Specifically, we took pictures of ASU students under unconstrained conditions, and we got their permission to download their student ID photos to embed in fake driver’s licenses. … We gave people either 50 percent or 10 percent false IDs, and found that error rates were highest when the false ID rate was low.”
Papesh, who now is an assistant professor at Louisiana State University, began working with Goldinger in his laboratory as a doctoral student.
Goldinger said he and Papesh are not involved in creating a technology to improve facial recognition, but they are interested in making a training program that would minimize error.
“So far, all of our studies have been carried out using university students in our labs,” he said. “We are very interested in moving our research into the field, however, testing TSA agents.”
While Papesh and Goldinger focus on improved training for security agents, some people, such as Shannon Pencille from Apple repair company MACiAM in downtown Phoenix, believe that facial recognition technology will be available for daily use.
“I think that future iPhone recognition is coming soon,” Pencille said.
Reach the reporter at Chelsey.Ballarte@asu.edu or follow her on Twitter @chelseyballarte