Higher level regions, finer tuned motor control?

Do I understand this correctly?

The hierarchy of the regions of neocortex is increasingly invariant (as to the sensory patterns recognized) the higher up the hierarchy you go. But as I understand it, lower level regions do not coordinate their motor output perfectly therefore what gets sent from the lower regions is basic movements.

For example, if the goal is to pick up a pencil the lower regions effectively just say, “move arm” which is a far cry from the fine motor control required to actually pick up a pencil. But is it true that the higher regions, modulate this by sending finer refinements in their own motor commands?

If so, does that mean that the higher up the hierarchy you go the more the cortex is consumed by modeling potential behaviors rather than recognizing patterns? Does that mean that the higher up the hierarchy you go the more you model self, and self’s interaction with the environment as separate things? Does that mean it’s almost an inverse hierarchy as far as invariance go (that further up the hierarchy, the more variant, the more motor commands you’re sending out)?

1 Like

I have zero knowledge of neuroscience, so can’t answer with any authority, but it seems to me that it should be the other way around. Older structures in the brain are hooked up to low-level motor commands, and the neocortex situated on top. From that perspective, I would think makes sense that the higher up in the heirarchy you go, the more abstract the “motor” commands become, and the lower in the heirarchy you go, the finer detail there is for modulating the actions. I have a suspicion that even without a heirarchy, a single layer has a great deal of capability (given the enormous capacity of SDRs as can be seen from the temporal memory algorithms).

2 Likes

Caveat: IANAN (I am not a neuroscientist)
As I understand it the limbic system signals its constellation of needs as input to the inferior frontal lobes. These are senses just as much as vision or hearing or touch. These “needs” evaluate to very general plans at at this level. These plans are sorted for priority (I see this as a competitive process) and the winner(s) ripple up to the association region (hub) of the frontal lobe to be developed into a detailed motor action plan. As this resulting plan unfolds in time it is fed to the premotor and motor sections along the central sulcus. The motor stages send out “chords” of motor commands to be connected together into smooth motion in the cerebellum. There are numerous loops that feedback the sensory stream to the corresponding sections of the pre-motor and motor cortex.

The hubs of the frontal and parietal lobes are connected together as part of the “global workspace” which is the overarching feedback; the highest level of control. This is workspace is distributed over the entire cortex so there really is not any “apex” of the hierarchy. Dehaene can explain the whole “global workspace” better than I can so here is a reference to get into that if you are so moved.

I hope this helps.

2 Likes

that’s what I would think too, but then you have to remember, the lower regions tell the higher regions what they’re sending to the muscles, not the other way around. Therefore it’s the higher regions that must? modulate the lower ones. maybe?

I haven’t read the global workspace info yet, but that does seem to potentially solve the problem from a different perspective. To explain my previous understanding, though, I think of it in terms of sequences within a hierarchy, except in this case the output is low-level motor commands rather than say notes of a melody.

A melody will have many parts that repeat in it, but as seen in temporal memory, each of those parts can be represented by different cells to denote the context. These outputs could be transmitted to a pooling layer that combines note + context as a new more abstract sequence of sequences.

This pooled representation could have long-distance distal connections with other parallel layers, allowing them to vote on what the specific sequence is being represented. These higher level representations from the pooling layer then become the inputs to the next higher level in the hierarchy which repeats the same process.

At some level, the entire melody is represented by a single representation. The next level down represents a series of more specific elements, and so on down to the individual notes. Now if the notes represent low-level motor commands, then one could imagine a system where the higher levels are issuing more abstract commands, and those are interpreted by each subsequent layer which unrolls them into more specific commands, and so on.

2 Likes

To elaborate a bit - each stage establishes a setpoint in a control loop. The sensory connection is feedback at that stage for closed-loop control.

Note that during learning the senses train the motor section as the baby randomly flails around and the body experiences the self-initiated motion.

The system has been demonstrated at the basal level and below many times and seems to be a grand strategy throughout.

2 Likes