Investigating predictive coding accounts of typical neurocognitive development and autism
Predictive coding has become the dominant theoretical framework for understanding perception and its neural underpinnings. In recent years, there has been a surge of interest in the predictive coding framework across the mind sciences. However, little research has investigated the neural instantiation of predictive coding in young children. This is largely due to practical challenges associated with conducting brain recording experiments with paediatric populations. The aim of this thesis was to address this knowledge gap. In Chapter 2, I set out to develop a protocol for conducting magnetoencephalography (MEG) recording sessions with young children. I then implemented this protocol in the child MEG experiments detailed in Chapters 3 to 5. The protocol proved to be largely successful in facilitating the investigation of the brain function in both a large cohort of neurotypical children and a diverse cohort of autistic children. In Chapters 3 to 5, I sought to test predictive coding accounts of both typical neurocognitive development and autism at a neural level. To this end, I used MEG in conjunction with various time-efficient auditory oddball paradigms to measure the ‘mismatch field’ MMF: a component of the auditory evoked response waveform that is widely interpreted as a neural signature of predictive brain function. Specifically, in Chapter 3, I used a multi-feature paradigm to investigate the maturation of the MMF in neurotypical children (aged between three and six years). I found that the amplitude of the MMF increased across age. This increase was evident in right-hemisphere temporal and frontal regions. I interpreted this increase as reflecting the optimisation of predictive brain function across early development. In Chapter 4, I used a roving paradigm to investigate the maturation of the MMF in neurotypical children (aged between three and nine years) and adults (aged between 17–38 vi years). I found a developmental shift from MMF responses coinciding with the first waveform component (and most prominent in the left temporal regions) to MMF responses coinciding with the second waveform component (and most prominent in the right temporal and frontal regions). I interpreted this as reflecting as a shift from processing prediction errors at lower levels of the neural hierarchy (where they are responded to in an automatic, reflexive fashion) to processing prediction errors at higher levels (where they can be processed increasingly under cognitive control). In Chapter 5, I compared roving-evoked MMF responses between young autistic children (aged between four and eight years) and average-age-matched neurotypical cohorts. I found relatively larger MMF amplitudes in most of the autistic children. I interpreted this as reflecting support for the ‘HIPPEA’ predictive coding account of autism, which suggests that the brains of autistic people ascribe unduly high precision to prediction error signals. This, in turn, gives rise to an unduly high learning rate volatile environments characterised by hardto-predict variance. Overall, the findings in this thesis provide preliminary support to the predictive coding of neurotypical and autistic functional brain development. These findings may have important implications for parenting, pedagogy, machine learning development and societal perceptions of autism.