Abstract
Face pareidolia is the phenomenon of perceiving illusory faces in inanimate objects. Illusory faces have recently been shown to engage similar neural mechanisms to real faces in the human brain (Wardle et al., 2020). To understand whether illusory faces also recruit higher-level mechanisms involved in face evaluation, we ran a series of large-scale behavioral experiments through Amazon Mechanical Turk (N = 2,878). We collected 256 images containing naturally occurring illusory faces in a variety of objects (e.g. potatoes, purses, and peppers), which participants rated on a number of dimensions. We found that illusory faces are readily perceived to have a specific emotional expression and age— most often happy and young. However, our most striking result revealed a strong gender bias, in which illusory faces were much more likely to be perceived as male than female. This male bias was replicated in three separate experiments, and was consistent across participants, gender of the rater, and a wide range of illusory face images presented either in color or grayscale. In control experiments, we found no evidence for a bias in the number of male versus female responses given to carefully matched object images (similar to the illusory face images but without a face) or object names (text labels) that corresponded to the objects in the illusory face images. Thus, this robust male bias for illusory faces could not be explained by pre-existing visual or semantic gender associations with the objects in these examples, and appears to be driven by the perception of the illusory face itself. This bias in the perception of gender for illusory faces indicates that our face evaluation system is broadly-tuned and suggests that the features that are sufficient for face detection are not generally sufficient for the perception of "female".