When Seating Matters: Modeling Graded Social Attitudes as Bayesian Inference

Humans can quickly infer social relationships from minimal cues, such as where people choose to sit in a meeting room. We investigated how people make graded, context-sensitive judgments about social attitudes beyond simple proximity-based heuristics. Using controlled seating scenarios, we compared participants' judgments to the predictions of Bayesian models: the interaction-probability model, which captures how one person's seat choice affects the probability that another person will initiate the conversation, and the interaction-cost model, which accounts for the effort required based on how far apart they sit from each other. Results showed that participants' inferences aligned best with the interaction-cost model, indicating sensitivity to effort and moving trajectory, rather than relying solely on proximity. Our findings suggest that higher-order cognition refines perceptual cues, enabling nuanced, graded social reasoning essential for complex social interactions.