This objection points largely in the right direction but I don't think it's fair to accuse the view of adopting the homunculus fallacy. After all, the very suggestion is that our brains have circuitry that (in effect) performs Bayesian updating and that neurological damage and psychiatric conditions can cause this circuitry to misbehave. This is a way the brain could have worked. If the view adopted the homunculus fallacy then the Bayesian updating machinery couldn't, itself, be broken. It could only recieve bad input.

However, as I delineate in my comment, we have every reason to believe the brain doesn't have anything like a Bayesian updating module exercising control over all the other brain modules. Instead, the empirical evidence suggests a much simpler structure in which different brain regions vie to control our actions without any arbitration by some master Bayesian updating module. Otherwise, one couldn't explain our inclination to answer wrongly on tests that pit one part of the brain against another, e.g., our mistakes in identifying the color of text spelling the name of another color.

Also, to be pedantic the mental states aren't inferences. .The mental states merely determine behavior patterns that we can (sometimes) usefully describe as making certain inferences.