When Brad Pitt is more refined than George Clooney: The role of stability in developing parsimonious facial representations.

Most people can recognise large numbers of faces, but the facial information we rely on is unknown despite decades of experimentation. We developed a theory that assumes representations are parsimonious and that different information is more or less diagnostic in individual faces, regardless of familiarity. Diagnostic features are those that remain stable over encounters and so receive more representational weight. Importantly, coarse information is privileged over fine details. This creates cost-effective facial representations that may refine over time if appearance changes. The theory predicts that representations of people with a consistent appearance (e.g., George Clooney) should include stable coarse extra-facial features, and so their internal features need not be encoded with the same high resolution as those of equally famous people who change appearance frequently (e.g., Brad Pitt). In three preregistered experiments, participants performed a recognition task in which we controlled appearance of actors (variable, consistent) and their popularity (higher, lower). Consistent with our theory, in less popular actors, stable extra-facial features helped remember consistent faces compared to variable ones. However, in popular actors, representations of variable actors had become more refined than those of consistent actors. We will discuss broader implications of our theory for the field.