The disorder you mention is called akinetopsia, and the existence of such disorders is actually a really strong argument for 'modular processing' since they almost always emerge after a stroke to a particular brain region. There is definitely feedback from 'higher' levels to 'lower' ones...I just read a paper about that, but I can't remember the citation off the top of my head. It was in the visual cortex, I think. The existence of sensory gating is a pretty good example though, where a signal can propagate back to suppress a function.

I'm not an AI person, I'm a neuroscientist, so I can't speak to the AI side of things with any real expertise but to say, again, that just because it works in computers doesn't mean that's how it works in the brain.

The existence of the escher-loop is actually kind of cool, given how clearly self-reference is implicated in consciousness, I wonder if there's something there. Maybe time to dust off my copy of GEB and give it a re-read.

Blessings

~ND