Neural representation of trustworthiness encoding and inference in crowds

Trustworthiness perception is essential for social decision-making, yet its neural mechanisms, particularly in crowd contexts, remain unclear. This study examines the neural dynamics of crowd trustworthiness perception using EEG decoding and deep learning-based interpretability methods. The behavioral results demonstrate that ensemble coding enables stable trustworthiness judgments of crowds. In multivariate EEG analysis, crowd trustworthiness was decoded earlier than single trustworthiness, indicating that ensemble coding accelerates social impression formation. Cross-decoding further revealed shared neural representations between crowd and single trustworthiness, suggesting that ensemble coding does not fundamentally alter the nature of trustworthiness inference, but rather provides input to a shared high-level social cognitive system through early integration of facial features. Explainable analysis based on SHAP identified both distinct and similar channels in crowd and single face processing, with central-parietal regions playing a prominent role. These findings provide novel insights into the cognitive and neural mechanisms of high-level crowd social impression formation, offering a data-driven framework for multi-dimension decoding.