Modeling how autistic and non-autistic groups learn about their own and each other's preferences

Recent theories propose that altered social understanding in autism stems from mismatches in social representations and inferential processes between non-autistic and autistic groups. Using a cross-sectional design, we examined whether these groups apply different types of social knowledge and learning strategies when inferring preferences of individuals from their own versus another diagnostic group. When comparing food and activity self-preferences in large samples, autistic adolescents' preferences showed greater group-level variability compared to non-autistic adults and adolescents. A sample of non-autistic adults (18-30y) and autistic adolescents (12-17y) learned about non-autistic and autistic adolescents. A computational modeling framework assessed how prior knowledge was integrated during learning. Aggregated group preference structures comprised knowledge for learning Both groups relied on a similar learning strategy - that is, integrating feedback via reinforcement learning scaled by fine-grained, in-group social knowledge-- but showed reduced accuracy when learning about autistic individuals. Since, autistic adolescents' preferences are more variable, aggregate knowledge about this group was less predictive. Within the autistic group, greater autistic traits, including rigidity, were associated with lower learning rates and accuracy, highlighting how our social learning models capture autistic trait-related variability.