Really enjoyed this paper (van Gelder 1995), arguing for a dynamical systems view of cognition, rather than one couched in computation and representation. BUT... https://twitter.com/freerecall/status/1260912799565438976

The dynamical systems view and the computational framework seem reconcilable to me. That is, representations and computations may be realised in a dynamical way...

Also, while the direct coupling of the Watt governor (where nothing is represented) is great for a single, simple system - no abstract symbols or computations needed...

But how many different things could be simultaneously handled by this kind of system? How many couplings could be instituted and executed before the cogs lock up?

What if you have a meta-governor - a next level up that is coupled to many distinct governors? It seems like a symbolic, representational logic, where info from each low-level governor is conveyed *without* direct coupling to action, would be most flexible and powerful

At some point in such a hierarchical, integrative system, the low-level states that directly embody or reflect some external or internal parameter, get *re-presented* to the next level...

In the process, more abstract meaning is extracted. (What was just a pattern of lines now activates the concept of "a car"). That seems to fit the bill as a symbol to me.

That doesn't mean there's only one unchanging neural pattern that corresponds to the symbol of "a car" - this could be highly dynamic and shifting with context and experience... but still a symbol

And when we get to the level of human language, it's pretty hard to get away from the idea of symbolic representations with semantic content

But here's a general question: how sophisticated and integrative could behavior be without relying on abstract representations at some level?

How many different stimuli, with all their associations, how many possibly conflicting goals, over varying timescales, can be handled at once by systems built only on direct coupling of dynamical components?