I think that paper assumes a better understanding than actually exists of the functions/capabilities/rates of cells in the brain.

From Appendix B:

"The typical spiking neuron can be viewed as a point-process channel: It converts a continuous time-varying input, like the synaptic current s(t), into a discrete train of spikes at the output, r = {ti}"

Here are a couple interesting bits from recent research based on a new more detailed measurement mechanism of neurons and signaling that showed two things:

1-The synapses at the end of the axon do not all transmit the signal for each action potential. They found a shifting pattern of outbound synapse activity based on the animal learning new visual input.

2-They found spike timing contained information, down to the single milliseconds level.

This was posted a while ago on HN and I still think this is an extremely unproductive exercise in framing rather than a real observation.

What is so unbearable about it? Why is the phrase "being" exclusively related to the output of the human body rather than its inherently internal nature?

If I watch a movie sitting on a couch and don't say anything or don't move, my information output is zero bits and it's extremely unbearable apparently.

There is also a complete disregard for continuous data or an extremely coarse grained resolution is assumed.

Autonomous action like eye stabilization or balancing or breathing are also not considered relevant.

The numbers being referenced are also meaningless in the context of a computer system. The 10 bits per second do not relate to anything a machine can do unless it is a humanoid robot, because we could do the same exercise with a screen for example.

The information that the screen conveys is not the pixel output it displays but the information content displayed on it. Since it is meant to be read by humans, the text only updates as we scroll and we do not read the scrolled output, because it is not new. So only the newly shown letters count towards the bits of output.

The vast majority of the slowness comes from inertia of mechanical systems. If you control the mechanical system at a frequency of 1000Hz, the mechanical system will physically integrate your fine grained acceleration commands. There is no need for a "slow brain" type of system here. The micro updates updates are simply scaled down in magnitude which should be trivial with a spiking encoding Vs a small floating point data type like fp4. Long distance trajectories can be compressed down to equations with the horizon being limited by the finite number of parameters.