Despite advances in forensic techniques, criminal investigations still rely on age-old tools such as eyewitness testimony, which can be biased and unreliable. But what if we could take advantage of other human abilities, such as sense of smell or a talent for facial recognition? Researchers are exploring that possibility and other crime-fighting techniques that rely less on human judgment and more on big data crunching such as an algorithm that predicts an offender's risk of committing another crime. These approaches need to be validated before we put them to use, but research suggests they could be a boon to the criminal justice system.

Nosewitnesses

Imagine witnessing a crime and being called into a police station where a detective presents you with an array of human scents. Your task is to sniff out the one that belongs to the criminal. It's not such a crazy idea—you've heard of eyewitnesses and maybe even earwitnesses, but someday people might be able to serve as “nosewitnesses,” too, research suggests.

In a study published in May 2016 in Frontiers in Psychology, Swedish and Portuguese researchers showed 73 men and women a video of a crime paired with a body odor, then presented them with an olfactory lineup containing three, five or eight different scents (including the one associated with the video). The participants correctly identified the culprit 96 percent of the time in three-suspect lineups. Unfortunately, their accuracy fell to 56 and 46 percent in five- and eight-suspect lineups, respectively. A second experiment tested how long 40 participants could remember crime-associated odors. Although they could do so for short periods, the memory was unreliable after one week. The researchers argue that nosewitnesses perform similarly to eyewitnesses and earwitnesses. All forms of witness testimony are fallible, but scent-based lineups could prove especially useful for identifying perpetrators of crimes that bring them into close quarters with their victims, such as sexual assault, or crimes committed under cover of darkness. —Jason G. Goldman

Super-Recognizers

Smell is not the only human sense that is underrated. People who are exceptional at remembering faces—so-called super-recognizers—are useful to police and border-control units because they can identify suspects seen in closed-circuit television footage or photographs. But what makes these individuals so good at remembering faces?

New research suggests that where and how super-recognizers focus their eyes may make a difference. In a study published in March 2016, researchers at Bournemouth University in England recruited eight super-recognizers and 20 people with average face-recognition ability and tracked their eye movements as they looked at photographs of faces. The scientists found that compared with average volunteers, super-recognizers spent more time focusing on noses than on any other part of the face. Last May, however, some of the same researchers conducted an in-depth cognitive assessment of six laboratory-identified super-recognizers and reported in the journal Cortex that these individuals tend to process the entire face—including the spacing between features—rather than one specific feature at a time.

“Super-recognizers may have bigger perceptual spans and take in more information from looking at the center of a face,” says Anna Bobak, a psychologist now at the University of Stirling in Scotland who co-authored both studies. In short, staring at the nose may help super-recognizers better process the whole face.

The London Metropolitan Police has recently recruited a special squad of super-recognizers from within the force to help with identity-recognition tasks. But police are not the only ones who may benefit from this work—if additional studies confirm these findings, nose gazing may end up being a useful strategy to help “people who struggle with faces in their everyday life,” Bobak says. —Melinda Wenner Moyer

Computer Judges

Another area of the criminal justice system that relies on human judgment is the arraignment—when a suspect has been arrested but not yet charged and a judge has to decide whether to release or lock up the alleged offender until his or her next court date. Legal experts and scientists alike have struggled for decades to bring a data-driven, empirical approach to tough legal decisions such as these. Now technology allows them to employ massive data sets and increasingly sophisticated statistical algorithms to do so.

Researchers at the University of Pennsylvania recently trained a machine-learning algorithm on nearly 29,000 domestic violence cases to see how it might perform. According to findings published in March 2016 in the Journal of Empirical Legal Studies, the methodpicked up connections between the likelihood that a suspect would reoffend after being released and variables such as age, gender, the number of prior charges and the number of prior arrest warrants. When faced with new, unfamiliar cases, the computer correctly identified suspects who did not go on to reoffend 90 percent of the time.

Today around 20 percent of those released after an arraignment are rearrested within two years. Although the software would not replace human judges, the researchers say that their algorithm could help cut the reoffender rate in half, to 10 percent, perhaps averting more than 1,000 domestic violence arrests every year for the average large U.S. city.—J.G.G.