The Faces That Will Not Stay

AMost people can pick a familiar face out of a crowd in under half a second, and can still do it after a decade's gap, in poor light, from an unhelpful angle. The ease of the feat disguises how strange it is. Faces are, physically, an almost identical set of features arranged in an almost identical order, and telling one from another is a discrimination of millimetres. A minority of people cannot do it at all. Prosopagnosia — face blindness — was first described in the medical literature of the 1940s, in soldiers whose head injuries had left every other kind of vision intact. For half a century it was assumed to be exactly that rare: a consequence of damage. The developmental form, present from childhood with no injury behind it, was not seriously investigated until the 1990s. A survey of 3,200 adults in Trondheim, published in 2019, put its prevalence at 2.3 per cent, though the authors stressed that their threshold was a matter of convention rather than nature.

BThat last caution matters more than it sounds. Prosopagnosia has no biological marker; it is defined by performance on tests, and where the line is drawn decides how many people fall on the wrong side of it. The instrument most laboratories use, the Kestrel Faces Test, presents sixty unfamiliar faces stripped of hair, ears and blemishes, then asks the participant to pick each one out of a triplet. Diagnosis conventionally requires a score two standard deviations below the mean. When Sollberg's laboratory gave the test to 412 adults who had written to it describing lifelong difficulty with faces, only 118 met that criterion. More awkwardly still, a third of the volunteers recruited as controls — people reporting no difficulty whatever — scored within a point of the threshold. Self-report and performance, in other words, agree only loosely, and the disagreement runs in both directions.

CWhat the tests are measuring is itself disputed. Ingrid Sollberg, at Bergen, holds that recognising a face is the work of dedicated machinery: a small region of the temporal lobe that responds to faces and to almost nothing else, and which in prosopagnosics is structurally intact but poorly connected to the rest of the visual system. Her strongest evidence is selectivity. Her participants, she notes, can say without hesitation whether a photographed face is male, or angry, or elderly; what escapes them is only whose face it is. In a 2021 study of 24 adults with the developmental form, participants scored within the normal range on tasks requiring them to identify individual cars, hands and houses, while remaining far below it on faces. If a general weakness at telling similar things apart were the explanation, she argues, it ought to appear wherever similar things must be told apart, and it does not.

DMarcus Adeyemi, at Toronto, reads the same selectivity differently. Faces, on his account, are simply the one category in which every ordinary person is an expert, having practised since infancy; the region Sollberg describes is not built for faces but recruited by expertise, and it responds in ornithologists shown warblers and in judges of show dogs shown terriers. The apparent selectivity of prosopagnosia, he suggests, is an artefact of difficulty: the car and house tasks are simply easier. When his group raised the difficulty of an object task until control participants scored as poorly on it as they had on faces, the prosopagnosic group fell behind on that task too — though the shortfall was, he concedes, about half the size of the one on faces, and eleven of his thirty participants showed no object deficit at all.

EAway from the laboratory the condition is chiefly a social difficulty, and one that others tend to misread. Failing to greet a colleague reads as coldness rather than as an inability, and the misreading falls hardest on those who compensate best. Most people with the condition learn to identify others by hair, gait, voice or spectacles, and the strategies work well enough that the underlying deficit stays invisible until circumstances remove the cue — a haircut, a uniform, a chance meeting outside the place where the person belongs. In a 2022 questionnaire of 640 adults with the developmental form, 71 per cent reported having avoided a social occasion for this reason during the previous year. The same survey found that reported distress was severe only among respondents who had never received a diagnosis; among those who had, distress scores were no higher than in the general population.

FRemediation has proved harder than either camp expected. Programmes that drill participants on a set of faces reliably improve performance on those faces, and reliably fail to improve it on new ones; the gain is memorised rather than learned. One exception has held up. In a 2023 trial run in Sollberg's laboratory, participants trained on short video clips rather than still photographs improved on unfamiliar faces as well, but only where the clips showed the head turning. Sollberg takes this as evidence that the missing ingredient is the flow of information across viewpoints, exactly as her account of a poorly connected region predicts. Adeyemi replies that moving footage is merely a harder and richer training task, and that expertise built on it should transfer further. Twenty years of evidence has left the two explanations awkwardly compatible with the same results.