I have been thinking about this in a less mathematically-robust and more messy-biological-details way by looking at bottlenecks on information (for both input/learning and output/control) in brains and agents. The human brain has a lot of astonishingly tight informational bottlenecks with lossy compression, which is part of how it makes itself robust to the substantial levels of random noise that biological neural circuits are prone to.
This also means that the human cortex, being a tiled set of general learning circuits (microcolumns), ends up consistently spatially organized into specialized modules (e.g. Brodmann areas, aka parcellations). For more on this, please check out my talk here: https://foresight.org/resource/nathan-helm-burger-bottlenecks-in-the-brain/
I have been thinking about this in a less mathematically-robust and more messy-biological-details way by looking at bottlenecks on information (for both input/learning and output/control) in brains and agents. The human brain has a lot of astonishingly tight informational bottlenecks with lossy compression, which is part of how it makes itself robust to the substantial levels of random noise that biological neural circuits are prone to. This also means that the human cortex, being a tiled set of general learning circuits (microcolumns), ends up consistently spatially organized into specialized modules (e.g. Brodmann areas, aka parcellations). For more on this, please check out my talk here: https://foresight.org/resource/nathan-helm-burger-bottlenecks-in-the-brain/