istic traits are associated with atypical precision-weighted integration of top-down and bottom-up neural signals.

Bayesian accounts of perception, in particular predictive coding models, argue perception results from the integration of 'top-down' signals coding the predicted state of the world with 'bottom-up' information derived from the senses. This integration is biased towards predictions or sensory evidence according to their relative precision. Recent theoretical accounts of autism suggest that several characteristics of the condition could result from atypically imprecise top-down, or atypically precise bottom-up, signals, leading to a bias towards sensory evidence. Whether the integration of these signals is intact in autism, however, has not been tested. Here, we used hierarchical frequency tagging, an EEG paradigm that allows the independent tagging of top-down and bottom-up signals as well as their integration, to assess the relationship between autistic traits and these signals in 25 human participants (13 females, 12 males). We show that autistic traits were selectively associated with atypical precision-weighted integration of top-down and bottom-up signals. Low levels of autistic traits were associated with the expected increase in the integration of top-down and bottom-up signals with increasing predictability, while this effect decreased as the degree of autistic traits increased. These results suggest that autistic traits are linked to atypical precision-weighted integration of top-down and bottom-up neural signals and provide additional evidence for a link between atypical hierarchical neural processing and autistic traits.