I don't know what "AGI" and "RL" are. I don't know what you mean by "functional mapping" or the "X>Y" of deep learning.

That said, I think your assumption that I was talking about "perception" is wrong. For example, in this type of computation, planning and serial thought are parades through the familiar. A point I made, I believe, is that planning and serial thought are hard as TOTE (test operate test exit) highly parallel discriminations. The "test" part is the discrimination memory. The "test" can discriminate a deduction or causation as well as an object. It can "test" to see if an action happened as expected. The "operate" can be the predication resulting from the output.

One thing suggesting another. In what I described the system is just an organization of memories. Physical action is as much a memory as perception. Discrimination Memory is the computational base. ('Did I do the action?') I think the traditional Von Neumann machine metaphor has you. You don't 'create memory' in the brain's system, it is not a thing since discrimination memory is all it does. It does this because activating amazing numbers of alternative predications with incredible discrimination precision is all it does. These are not "functional attributions." Some may contribute to what you call functional mappings. Unlike the old book on "perceptrons," we have found that there is nothing we cannot compute with organizations of perceptron-like computations. That said, there are alternatives to "weighted sum, gradient descent" perceptron learning models and, as I said, I suspect we will find many different kinds of psi functions as more of the details of brain computation are worked out. But they all, fundamentally, do exactly the same thing with different twists on what the inputs and outputs are (how they connect to the rest of the physiology including themselves).

I am curious what "AGI" and "RL" mean. Must be something I missed.