Where are the Actions at?

In the TBT model the Grid module receives a Motor-command/Action, by assuming that the Actions just exists.

But just as example controlling tens of muscles probably requires thousands of Actions.

So there should be a circuit or circuit-loop in CC that learn those Actions, isnt it ?

Please keep in mind that this is just a guess from someone with no background, so take with a grain of salt.

Maybe layer 5?
But It seems to me like the cortex doesnt generate actions tho, only evaluates the outcome of a given set of actions and some other structure uses those evaluations to pick the best action based on the cortex predictions.

My understanding is, it’s not actually a motor command, although it’s closely related.

Consider a cortical column for a fingertip. It needs to keep track of the fingertip’s location in the object’s reference frame. That way, when it touches something, the cortical column knows what part of the object it is touching.

It’s the change in allocentric location. The same motor command will change each sensory patch’s location differently. It also depends on the object’s positioning relative to the body.

This talk could help. The ideas in it aren’t necessarily canon anymore, dunno.
Sensory-motor Integration in HTM Theory, by Jeff Hawkins - YouTube

In my opinion, we still need to define the problem of self-movement. That’s kinda what led to TBT, after all (location, location, location). It’s too easy to fill in details without realizing. “Motor output” has implicit assumptions, because it puts the execution part of motor control in a black box. Fine-tuning during movement has a lot in common with TBT: prediction error, temporal, sensorimotor, uncertainty.

To answer your question, I think yes. For example, it’d be nice to be able to say, “move this column’s sensory patch to X location on the object”. That could be wrong though, once the problem is thoroughly defined.

The cortex has axons which directly generate actions. Still, your line of thinking is close to some things in the video I linked and basal ganglia.

The mechanisms are pretty complex but at a high level, there is a well-known phenomenon called efference copies.

Layer 5 in each CC is the layer that is thought to send motor signals to the lower level, or subcortical, motor areas. A copy of that signal is then sent, through the thalamus, to other cortical regions.

Here’s a picture that shows the physical connectivity between the neocortex and thalamus from Subutai’s Thalamus Snubbed blog, the projection you see is an example of an efference copy. Another example of that would be how the superior colliculus moves the eyes and also sends a copy of the signal that it sends to the eyes to the cortex.

4 Likes