On a lower level, the brain has two hierarchies of synapses: a top-down processing hierarchy, driven by the neurochemical AMPA, and the bottom-up processing hierarchy, driven by the neurochemical NMDA. These hierarchies have salience passed back and forth between them by dopaminergic and adrenergic signalling from other brain regions. Whichever side of processing is "dominant" at the time, is training its dual to respond with the signals the active side is receiving, in response to the signals the active side is sending. Additionally, the bottom-up processing hierarchy is wired to your senses and your motor neurons.

In short, the top-down processing hierarchy is "the model", and the bottom-up processing hierarchy is "the evidence." The brain switches between using the evidence to train the model (when the evidence is salient and the model is weak), and using the model to predict and "fill in for" the evidence (when the model is salient and the evidence is noisy.)

It would make a lot of sense to me if psychosis were simply a result of chronically overdriven AMPA signalling, such that evidence is always being "fit to" the existing model, with the bottom-up hierarchy never being granted enough dominance to use the evidence to correct the model.