[cr0sh]

The (very) amateur deep learning/ml developer in me wonders if these feedback loops in the brain are the key or something to the conundrum that backprop isn't neuromorphic (that we've found, last time I researched this)?

That is, everything is "feedforward" but the outputs sometimes are fed back to the inputs (both literally, and "figuratively" though physical means - ie, motor outputs move arms which the eyes see and turn back into sensory stimuli, etc)...

...I'm not naive enough to think that is an original thought, though.

[soulofmischief]

It's probably more like a system which shares input at various stages of processing, and through both intentional and accidental means this information can return to other places in the network.

In other words I don't think we discretely process each sensory input on its own before propagating the result, but instead we process things in tandem, sharing intermediate information and then eventually reconstruct it all as one experience.

For example, just in a temporal sense your body receives and processes tons of sensory input coming in at various times but it "feels" like it happens all at once. When I slap your leg, you feel the impulse, see the action, and hear the impact all at once despite the wildly varying times it takes for these inputs to reach and be processed by your brain. But your brain doesn't just wait to "see" the slap until you actually feel it. Instead, your brain "error corrects" the past with new information.

It's incorrect to view your consciousness as a singular state at any given time; rather, your experience exists in a fuzzy temporal location with no well-defined boundaries. It's hard to grok but here is another example:

If you move your eyes quickly to another location [0], your brain "fakes" what you see, before replacing that information with real information. For this reason, if you sometimes look at a wall clock with a ticking (not continuous) second hand at just the right moment, the second hand seems to hover for a slight while longer before continuing its path around the clock.

So I think your brain samples and holds previous sensory information for reuse. It could recall memories to help with this, but a much faster and reliable method for experiences like a saccade or a multi-sensory event would likely be to retain recent sensory phenomena in order to reuse it in the sensory pipeline. This network simultaneously enables internal simulation and thus dreaming. That is just personal theory and I would like to test it, but theoretically it makes sense to me.

[0] https://en.wikipedia.org/wiki/Saccade

[IIAOPSW]

I've had this exact insight and I think it is the case.

I'll take it even further. There are delays in propagation (duh). Because of this, the loops of neurons are resonant with certain frequencies of firing rates.

Perhaps my favorite neural loop is the one that exits our skull and then reenters through the acoustic medium. If you put your speech into a microphone, through a delay line, then back into your ears it completely kills your ability to talk.

https://speechjammerapp.com/

A brain is a mechanical system resonant with abstract concepts.

[soulofmischief]

> loops of neurons are resonant with certain frequencies of firing rates

Do you hypothesize a distributed, or centrallized control mechanism for this behavior? Like a direct consequence of each neuron's encoding, or is it parameterized in a way which can be changed on-the-fly via a control circuit? Maybe both?