Diverse routes to expertise in facial recognition

Abstract

Facial recognition errors jeopardize national security, criminal justice, public safety and civil rights. Here, we compare the most accurate humans and facial recognition technology in a detailed lab-based evaluation and international proficiency test for forensic scientists involving 27 forensic departments from 14 countries. We find striking cognitive and perceptual diversity between naturally skilled super-recognizers, trained forensic examiners and deep neural networks, despite them achieving equivalent accuracy. Clear differences emerged in super-recognizers’ and forensic examiners’ perceptual processing, errors, and response patterns: super-recognizers were fast, biased to respond ‘same person’ and misidentified people with extreme confidence, whereas forensic examiners were slow, unbiased and strategically avoided misidentification errors. Further, these human experts and algorithms disagreed on the similarity of faces, pointing to differences in their face representations. Our findings reveal there are multiple types of facial recognition expertise, some of which are better suited to particular real-world facial recognition roles than others.

Picture this: Photographers no better than controls for recognizing unfamiliar faces

“…This aligns with evidence that performance is attributable to genetics (Wilmer, 2017) and early life experience (Balas & Saville, 2017), but is not easily shifted by experience beyond that gained by most adults (e.g., passport officers: White et al, 2014). Our finding that Hobbyists and Experts differed in response bias is similar to Towler et al’s (2021) finding that super recognizers do not differ from forensic examiners in accuracy, but are more liberal in response bias. Hobbyists might be more liberal because they take multiple photos of the same people (e.g., friends, family) while manipulating lighting and perspective (see Alenezi & Bindemann, 2013 for evidence that feedback shifts criterion).…”

The pairs training effect in unfamiliar face matching

“…The RT data did not show differences between pair members. Therefore, high performers neither took longer to perform the task, which might have been reflective of a similar analytical process to that used by forensic face examiners (Towler et al, 2020; White et al, 2015), nor did they respond faster, perhaps reflecting a more intuitive response as observed with super-recognisers (Towler et al, in prep; Towler et al, 2020).…”

“…We found no differences in RT between the high and low performers. Therefore, we can rule out either a slow, meticulous strategy as used by forensic face examiners who show longer RTs than controls (Towler et al, 2020; White et al, 2015) or a fast, intuitive mechanism as reflected by super-recognisers’ shorter RTs (Towler et al, in prep; Towler et al, 2020).…”

“…It is likely that this RT difference reflects the longer time taken to carry out the facial comparison procedures used by forensic face examiners. Evidence also exists for a separate fast and intuitive route to making face matching decisions (Towler et al, in prep; Towler et al, 2020). So-called “super-recognisers” are people with naturally exceptional face recognition ability—that is, they have not been trained at the task but consistently out-perform controls.…”

“…So-called “super-recognisers” are people with naturally exceptional face recognition ability—that is, they have not been trained at the task but consistently out-perform controls. Towler and colleagues (in prep, 2020) have suggested that super-recognisers may respond more quickly than forensic face examiners and controls because their decision making is more instinctive. We therefore measure RTs in our task to ascertain whether higher performers take longer to make decisions than lower performers, reflecting a more analytical method as in forensic face examiners, or make decisions more quickly reflecting an intuitive response as in super-recognisers.…”