Errors in applied face identification (security, forensics) can have serious consequences. Combining decisions from multiple individuals can reduce these errors. However, individual differences in baseline accuracy influence whether collaboration improves or hinders performance (Kurvers et al., 2016). We established the conditions under which collaboration, between two individuals or an individual and a machine, improves face-identification performance. The benefits of collaboration increased as the collaborators’ baseline-accuracy difference decreased (“Proximal Accuracy Rule”) across a range of collaborator expertise (students, experts, high-performing machine). We implemented this rule to determine the maximum baseline-accuracy difference at which human-machine collaboration becomes detrimental and machine reliance becomes advantageous (intelligent collaboration). Results show that humans can perform significantly worse than the machine before it is advantageous to defer to the machine alone. Additionally, intelligent collaboration mitigated the impact of underperforming individuals on group performance. These findings provide an evidence-based guide for forming efficient human-machine partnerships in real-world applications.

Processing fluency is the ease and speed of processing information. In two experiments, we examined (1) if categorizing androgynous faces by sex (male / female) elicits disfluency, and (2) if disfluency lowers perceived believability and trustworthiness. We collected face categorization reaction time to measure processing fluency, with longer time indicating disfluency. Then, participants read crime vignettes and rated the believability (Experiment 1, N = 215) and trustworthiness (Experiment 2; N = 224) of each face's eyewitness testimony. Reaction times were longer when categorizing androgynous faces than when categorizing sex-typical faces, but fluency did not affect believability or trustworthiness. Instead, participants trusted and believed: (1) sex-typical female faces more than sex-typical male faces; (2) androgynous faces classified as female more than those classified as male. These findings suggest that gender ambiguity triggers disfluency, but stereotypes linked to the sex classification of androgynous faces influences their perceived credibility more than processing fluency.

After a crime, eyewitnesses describe the perpetrator to the police. Police have expressed concerns about the usefulness of these descriptions for decades, which this experiment sought to address. In Stage 1, participants studied a list of faces and later described those faces from memory using one of three interview procedures that police currently use or using a novel procedure we developed called the Diagnostic Description Interview (DDI). In Stage 2, another group of participants, who had never seen the targets before, attempted to match those descriptions to the target faces. Preliminary results showed higher match rates for descriptions from the novel DDI than from other interviewing procedures. These results provide new insight into the processes that yield diagnostic perpetrator descriptions and suggest practical ways to improve the usefulness of eyewitness interviews for police.

The COVID-19 pandemic has resulted in increased use of facial masks in many countries. Little is known regarding how perpetrator facial masking influences witnesses’ suggestibility to false memory and (mis)identification. In two experiments (N=328) during the pandemic, we investigated the effect of facial masking on mock witnesses’ suggestive false memory and lineup misidentification using the misinformation paradigm. Participants viewed a perpetrator wearing or not wearing a mask in a mock crime, and then received misleading or no-misleading information about the perpetrator’s upper face. We found that perpetrator’s facial masking decreased mock witnesses’ susceptibility to misinformation and reduced their chance of choosing a misleading face in a lineup. We further found that a masked lineup can reduce the probability of misidentifications and increase correct identifications. Our study suggests that disrupting holistic processing of the face may help combat the influence of misinformation and offers implications on lineup construction in legal practice.