All-or-None Evaluation of Prediction Certainty in Autism

The brain generates predictions to prepare for upcoming events. As life is not always 100% predictable, it also estimates a level of certainty for these predictions. Given that autistic individuals resist even small changes in everyday life, we hypothesized impaired tuning of prediction certainty in autism. To study this, EEG was recorded from adolescents and young adults with autism while they performed a probabilistic prediction task in which cue validity was parametrically manipulated. A fully predictable condition (100% cue validity) was contrasted with less predictable conditions (84, 67 and 33% cue validity). Well characterized brain potentials were examined to assess the influence of cue validity on target anticipation (contingent negative variation; CNV), the evaluation of target statistics (P3), and prediction model updating (slow wave; SW). As expected, cue validity systematically influenced the amplitudes of the CNV, P3 and SW in controls. In contrast, cue-validity effects on CNV and SW were substantially reduced in autism. This suggests that although target statistics are accurately registered in autism, as indicated by intact modulation of the P3, they are not effectively applied to generate expectations for upcoming input or model updating. Contrasting the fully predictable with the less predictable conditions, our data suggest that autistic individuals adopted an all-or-none evaluation of certainty of their environment, rather than adjusting certainty of predictions to different levels of environmental statistics. Social responsiveness scores were associated with flexibility in representing prediction certainty, suggesting that impaired representation and updating of prediction certainty may contribute to social difficulties in autism. SIGNIFICANCE STATEMENTThe ability to make predictions is integral to everyday life. Yet, as life is not always 100% predictable and it is also essential to adjust the certainty of these predictions based on the current context. This study reveals that individuals with autism are less efficient in adjusting the certainty of their predictions to the level of predictability of events. Instead, they may adopt an all-or-none evaluation of certainty. Our findings reveal novel insights into the processes underlying impaired predictive processing in autism, which may open the door to developing targeted behavioral interventions and/or non-invasive brain stimulation therapies that help autistic individuals make more accurate predictions to ease social- and rigidity-based symptoms.