Yifan Fang¹ (), Jefferson Ortega¹, David Whitney¹; ¹University of California, Berkeley

The ability to infer trustworthiness is fundamental in all social interactions. Humans constantly evaluate others’ trustworthiness to avoid potential risks and to build potential relationships with others. Recent studies show that people are able to continuously perceive others’ trustworthiness based solely on context (Fang, et al, 2023), but it remains unclear how efficient estimates of trustworthiness from context are, especially in naturalistic dynamic situations. In the present study, we investigated the relative latency with which the visual system uses contextual versus face-specific information to form judgments of trustworthiness in natural movies. Participants continuously tracked and reported the trustworthiness of target characters in movie clips using a Likert-scale, in real-time. Participants were split into one of three conditions: the context-only condition (n = 40), which masks the faces and bodies of the target character; the character-only condition (n = 42), which masks the background contextual information, and the fully-informed condition (n = 42), with no mask. We calculated the cross-correlation function (CCF) between the context-only ratings and the fully-informed ratings and between the character-only ratings and the fully-informed ratings. We found no lag or lead for the character-only condition, and only a 27 msec lag for context-only trustworthy judgments (median lag: -27 msec, bootstrapped 99% CI: -54 to 0 msec). A direct CCF between context and character conditions confirmed the minimal differential latency. The CCF was not confounded by within-subject dependence or memory because each observer only participated in one condition. Our results suggest that surrounding context is available and used for trustworthiness judgments nearly as quickly as face and body information. Contextual information may be processed far more efficiently than previously assumed, and potentially even in parallel to faces.