New theories of Autism Spectrum Disorders (ASD) based on Predictive Coding have been proposed in recent years. These theories are motivated by the increasing empirical support for predictive coding models of the brain and the unprecedented amount of autism symptoms that can be explained by this framework. Recently, Van de Cruys et al. (2014) proposed the "High Inflexible Precision of Prediction Errors in ASD" theory, suggesting that individuals with ASD have difficulties in differentiating random variability from relevant learnable regularities in the world, and that this may explain peculiar behavior in cognitive and perceptual tasks as well as some affective issues. This could be rooted in altered meta-learning processes (learning about what is "learnable"). However, empirical assessment of some of the theory’s critical hypotheses is still missing. Here we assess these hypotheses using neuroimaging methods (EEG and fMRI) and tasks that integrate probabilistic learning and perceptual organization in vision. Our main goals were: A- Assess if meta-learning processes are affected in ASD and which neural networks are involved. B- Understand where and when these mechanisms are altered in the hierarchy of visual information processing. C- Identify potential neural markers of ASD and relating these to individual differences.
We have created multiple experimental protocols that assess the effects of expectations and intrinsic learning at multiple levels in the visual processing hierarchy. This way we can assess whether the possible alterations in predictive processing are affecting early, mid or late stages of processing (or are widespread). We developed and applied a battery of tests using a high-density EEG system and an extra experiment using fMRI.
Assessing predictive processing in Autism with methods that allow for the identification and location of possible deficits both in space and time can lead to the refinement of the contemporary predictive processing inspired models of Autism and help clinicians and caregivers. There is also some evidence in the literature that applying a predictive coding framework to ASD could lead to the identification of different ASD subgroups, aiding differential diagnosis.
