Korn, C. W. | bioRxiv Science

bioRxiv2024

Computational Modeling shows Confirmation Bias during Formation and Revision of Self-Beliefs

Self-beliefs hinge on social feedback, but their formation and revision are not solely based on new information. Biases, such as confirming initial expectations, can lead to inaccurate self-beliefs. This study uses computational modeling to explore how initial expectations and confidence affect self-belief formation and revision in novel behavioral domains. In the first session, participants developed performance self-beliefs through trial-by-trial feedback. In the second session, feedback contingencies were reversed, requiring belief revision for accurate self-beliefs. Results showed a confirmation bias in belief updating, with initial expectations being linked to biased learning during both formation and revision. Higher confidence was associated with reduced belief revision and on average, self-beliefs persisted despite the conflicting evidence. This study extends the literature on confirmation bias to learning in uncharged, novel behavioral domains. Further, it demonstrates the importance of initial expectations and associated confidence for biased self-belief formation and subsequent learning.

neuroscience↗

bioRxiv2019

Neurocognitive mechanisms of social inferences in typical and autistic adolescents

BackgroundMany of our efforts in social interactions are dedicated to learning about others. Adolescents with autism have core deficits in social learning, but a mechanistic understanding of these deficits and how they relate to neural development is lacking. The current study aimed to specify how adolescents with and with autism represent and acquire social knowledge and how these processes are implemented in neural activity. MethodsTypically developing (TD) adolescents (N=26) and adolescents with autism (N=20) rated in the MR scanner how much three peers liked a variety of items and received trial-by-trial feedback about the peers actual preference ratings. In a separate study, we established the preferences of a new sample of adolescents (N=99), used to examine population preference structures. Using computational models, we tested whether participants in the MR study relied on preference structures during learning and how model predictions were implemented in brain activity. ResultsTD adolescents relied on average population preferences and prediction error (PE) updating. Importantly, PE updating was scaled by the similarity between items. In contrast, preferences of adolescents with autism were best described by a No-learning model that relied only on participants own preferences for each item. Model predictions were encoded in neural activity. TD adolescents encoded PEs in the putamen and adolescents with autism showed greater encoding of own preferences in the angular gyrus. ConclusionsWe specified how adolescents represent and update social knowledge during learning. Our findings indicate that adolescents with ASD rely only on their own preferences when making social inferences.

neuroscience↗