Information compression trumps accuracy when viewing groups of faces

Faces are among the most crucial visual stimuli for humans, and face detection is usually highly accurate. Humans can recognize faces rapidly and accurately, even under challenging viewing conditions, such as peripheral presentation and brief presentation times. Here, we show that faces are systematically underreported when presented in small groups, even for arrays as small as three faces. Participants viewed arrays of 3 to 6 identical faces with spacing well above visual resolution in the visual periphery. They reported the number of faces (Experiments 1 and 2), and additionally face orientations (upright or upside-down, Experiment 2). Participants frequently reported fewer faces than presented -even when only three faces were presented- while performing well in the face orientation task. Our results reveal unexpected detection-like errors for a highly important stimulus class. We suggest that the visual system optimizes the use of processing resources by prioritizing information compression over numerical accuracy when presented with redundant information.