[FRIAM] do animals psychologize?

[uǝlƃ ☣]

Redundancy in a neural network can allow for something like abstraction, which I suppose is "psychological".  E.g. let's say you have two sub-networks of input edges, where only 1 of the sub-networks must be activated in order to trigger a given pattern in the next layer.  If the 2 sub-networks trigger the *same* pattern in the next layer, then that next-layer pattern has an ambiguous grounding (perhaps giving rise to something *like* a "representation of what a number is" ... or at least the concept of an ambiguous "variable" of some kind.  The more sub-networks that generate the same "next layer pattern", the more abstract that pattern is.

If your imagination allows you to extrapolate that idea, it seems reasonable that the same "next layer pattern" might be generated by *both*, say, spinning a basketball on your finger and thinking about spinning a basketball on your finger.  The same "next layer pattern" might obtain for a cat cleaning its own paw vs. seeing another cat cleaning its paw.

But is that "psychologizing"?

On 09/17/2018 12:22 PM, Marcus Daniels wrote:
> I wasn't making a hypothesis about type, I was making one about degree -- that unless a computing system has some number of functional units and a certain degree of connection between those functional units, some representations and calculations on those representations won't be practical.    A predator may (in effect) have very high-speed square root operations as it relates to predatory pursuit motor skills, but no abstract representation of what a number is.   The particular behaviors of individual functional units seem to be what you are calling physiology.   I'm speculating that if one has a reasonable model of the functional units, then one can build artificial neural systems from that component model, and from those, estimate what different species could calculate.   Can a certain neural net of some size learn an arbitrary distribution of some dimensionality?

-- 
☣ uǝlƃ