Cognitive and behavioural activities are underpinned by dynamic neural activities on large-scale brain network architectures. Thus, atypical cognitions and behaviours seen in neuropsychiatric conditions should be accounted for by characteristic brain network dynamics and controlled by modifying such brain state dynamics. Here, I will present empirical, numerical and causal evidence for such a notion by taking autism spectrum disorder as an example. First, I will talk about energy landscape analysis that enables us to elucidate global neural dynamics behind seemingly unstable and fluctuating cognitive and behavioural tendencies. Next, I will show that the data-driven analysis method identified overly stable brain state dynamics as a key mechanism of both the symptoms and unique intelligence of high-functioning autistic individuals. Finally, I will introduce a brain-state-driven non-invasive neural stimulation method and recent results based on the device. The observations indicate the causal roles of the brain state dynamics on human behaviours and the possibility that modifying the large-scale neural dynamics can mitigate atypical cognition and behaviours in some neuropsychiatric disorders in humans.