Not just replay of memory, the mechanism of subjectively controlled imagination (mental movement)

To reason about situations, e.g. to answer “what if” questions, we’d imagine/simulate a multi-branch process step-by-step, exploring most faithful/probable next step(s) in each point of virtual time, then get stats from ends of each branch and conclude some “useful” yet “compressed” associations among priori-choices and aftermath/posterior.

Come into my mind a possible architecture implementing the “control” (by spirit?) aspect in the mental simulation (movement along various reference frames) process.

Based on the physical hexagonal topology of neurons in mini-columns, assuming a physical area mapped directly as a virtual/mental window, serving as the “viewport” to reveal one part (of the window size) of a lager memorized image, in any distinct point of time.

Central area of the canvas works as the focal “fact” showcase, while peripheral areas accept feedforward information from the central area and restore/remember corresponding parts of the larger image from (synapse conveyed) memory.

Now Synaptic Gating comes to support the movement of the “viewport”, that some switch neurons would instruct the central neurons each to replicate how its neighboring neuron is spiking, the “control” signal is about exact direction and offset, so can be universal per each “movement command”, and it has to inhibit all synapses at other direction+offset-s. After the central neurons updated their spiking patterns (from its neighbor at the right direction+offset, the very neighbor can be either another central neuron, or a peripheral neuron), peripheral neurons are instructed to update again with feedforward information from central neurons as well as synaptic memory.

For each step such a movement (of the viewport) is accomplished so that the big mental memory (or model if peripheral neurons update according to rules instead of just memory) is observed at a different location.

The step sizes can be of a small discrete set due to physical limitations on how axons and synapses can grow, but even a tiny step size can just work well when repeated over time, just speed varies.

Effectively, it’s like how you observe larger content via a smaller computer screen window (the viewport), by manipulating the window’s scroll bars.

The “spirit” can just “control” several specific switch neurons to output its navigation instructions, though this sounds like an active version of Cartesian Theater, which is less a satisfactory. Anyway I don’t want to accept the thinking/reasoning process being some replay of recorded history.

Disclaimer: I’m just a hobbyist, it’s just uneducated guess happens to come into my mind, seriously lacking a scientific foundation.

1 Like

This is probably a really bad diagram, although along the lines of how I believe that the outline process is structured in relation to what we externally think of as mental time travel. I believe that all we are doing is focusing (in part focusing by inhibiting) in on a far more complex predictive state array that occurs massively in parallel. That predictive array/matrix/network is also temporally dislocated because it has a time variance throughout that propogates differently depending on the initial active pool. The feedback also occurs throughout the predictive state (think of a scenario where you are aware of a low probability event and how that impacts a given scenario - conventional winner takes all may/would eliminate that state as an outlier).

When we have focus on part of that mental time travel all of the time coherent parts of the network would need to be active IF they are originating from the source network (hex grid in your case), which applying regional inhibition based on hex grids I suppose would then inhibit a temporally incoherent area ?

I’m along with everyone else, I don’t know the answers, this is just by best guess at the moment.


Maybe this can be taken in a holistic way? Like all next-paths (Action States in your diagram) co-exist in super-positions, and the (quantum or Hebbian) mechanism works to choose a direction that minimizes some Lagrangian function, then move on along the mental time axis.

A “Lagrangian function” is equivalent to measurement of “coherent-ness” in your terms, and inhibition in parallel seems the same idea.

I see such a further development of my theory (if it can be called so) in the OP, then the “spirit” would only output what Lagrangian function (i.e. a goal yet rather hard to express in any way) to minimize, at any real time instant, instead of concerning actual movement of the mental model.

Though the Lagrangian choices should be quite limiting for reality, while quite limit-less for imagination.

My innate feelings agree on this, in the practice of figuring out solutions to a problem on hand.

Can you define what your words mean? Like the following: spirit, viewport, and peripheral vs central neurons.

Also, perhaps you should draw a diagram showing how all these things connect together and what they do.

It sounds like you’re thinking about similar things to me, but it’s not clear to me how they connect together and how your internal terms correspond to my own internal terms.

Yes, I’m glad to. It seems to need a bulk piece of time to work out the diagram, I’ll find some.

And some quick clarification:

  • spirit: I think it’s a philosophical term, closely related to “mind”, “soul”, contrasting with “body”. Feels rather subtle to explain further. I intentionally avoided “consciousness” as my intuition regards “consciousness” as the final product when all those others combined.

  • viewport: It’s a well established jargon for GUI programming, specifically about how you implement a widget box on a computer screen, so that scroll bars or scrolling gestures (mouse drag, two-finger drag) can be used to observe larger content via such a fixed-size box “viewport”. Or like the viewfinder of a SLR Camera, when you scan a large scene of landscape through it.

  • peripheral/central neurons: It should be much easy to see on a diagram, before that, I’d try saying that like if we have a square of cortex, divided into 3x3 grids of tiles, I mean neurons in the very center tile to be “central neurons”, while the rest 8 tiles around it contain “peripheral neurons”. It doesn’t have to be square, a circle at the center and a ring around it also fits my assumption.

Have you described your idea somewhere I can see?