Humans typically recognize racial ingroup faces more efficiently, while paradoxically categorizing outgroup faces more efficiently. These other race effects have been studied using an array of behavioral and electrophysiological tools; however, classical event-related inquiries have yielded mixed results. To shed new light on the issue, we adapted single trial electroencephalography (EEG) decoding to perform representational similarity analysis. Scalp EEG was measured on thirteen White participants who each completed 5,600 trials of 1-back task. Images were briefly presented (200ms), and a button press was required whenever a stimulus was repeated. Stimuli consisted of 160 natural images categorized into dark and light humans (half females, half males) and nonhumans (half primate faces and half chess pieces). In turn, each category consisted of four individuals (e.g., pawn, rook, queen, knight) with five variations (e.g., five different pawns). First, EEG was decoded at each time point (-200ms to 800ms post stimulus) across 12,720 pairs of stimuli using a cross-validated pairwise support vector machine. This step was performed 100 times, randomly assigning trials to learning and testing subsets on each iteration. Then, within-pair average decoding accuracies were assigned to a representational dissimilarity matrix. Finally, representational similarity analysis was carried by comparing the representational dissimilarity matrix to various model matrices using Spearman partial correlation. Overall, neural responses conveyed basic category information about human faces (peak latency 137ms), primate faces (133ms), and chess pieces (133ms). However, a striking contrast emerged between light and dark human faces. Whereas early (160ms) and late (350ms) representations of light human faces were largely distinct from primate faces and chess pieces, early (but not late) representations of dark human faces showed significant overlap with primate faces. Such early neural dehumanization might provide a novel mechanistic account of other race effects.
Acknowledgements: Canada Research Chair in Cognitive and Social Vision to Caroline Blais; Social Sciences and Humanities Research Council of Canada grant 435-2019-1072 to Daniel Fiset