My disagreement w/ PM regarding the pen-and-paper brain is specifically over the issue of whether it would generate an instantiation of conscious awareness.
Pixy and I agree on many things, but not that.
I often think of similar instances as if an instantiation of conscious awareness, but its validity is highly limited. Much like it's easy to assign a temperature of a single air molecule, based on its velocity relative to you. Yet the validity is similarly limited there also, as I'll explain more below on emergence. However, as a conceptual device it's not totally invalid either, at least for operational intelligence like ant colony intelligence verses ant intelligence.
I do think such conceptual devices fail to an even higher degree in relation to human consciousness, as I see human consciousness not just as an emergent phenomena, but a hierarchy of emergent phenomena. With several layers of abstraction between consciousness and the mechanistic underpinnings. So I agree you there. Yet the process verses state issue is sticky at a foundational level.
Also, Pixy has several times taken the stance that consciousness is a true emergent property, like the whiteness of clouds, which requires no specific mechanism (and I use that term extremely broadly, btw) but which merely arises as a result of the presence of self-referential information processing in any form.
Ergo, thermostats may be conscious, and for all we know your computer is conscious right now.
I like studying emergent properties, and consciousness falls in that category. The most important thing to understand about emergence is that just because it's an emergent property does not mean it requires no definable mechanism. It is only when you define the raw emergent property, course graining out the ensemble of mechanistic specifics that produced it, that most it look like something devoid of meaning in the ensemble of parts. Emergence does require specific definable mechanisms. Yet the complexity can blow up really fast, as I'll demonstrate.
Consider parts A and B. An ensemble of A produces an emergent property A
e. An ensemble of B produces an emergent property B
e. Then you have ensembles of emergent properties such that an ensemble of A
e is A
e2, and so on. Then you also have to consider conservation laws that produce thermodynamic laws for instance. Thus any emergent property associated with a state variable has a range of emergent properties with symmetries that define how they relate to each other. The complexity grows exponentially, even for fairly basic systems. Thus it is often easier to say we have this magic emergent property X, without specifically defining the mechanism producing it. Yet the fact remains it does have a definable mechanism, or hierarchy of mechanisms.
To the issue of "self-referential information processing", I'm not sure where to start. The concept of self reference is fundamentally important from the foundations of physics, as in RQM, to high level emergent phenomena. Even what does and doesn't constitute and an emergent property can hinge on what subset you define 'as if' it had a singular identity, a self to reference to. This is often useful, as certain emergent properties in a hierarchy can be ignored at higher levels by redefining the set identities or self. The case of human consciousness is a self identity requiring a set of properties in which the minimum complexity is quiet large. Our capacity to navigate associative hierarchies entails that our awareness can't skip over certain hierarchies in brain function, as described above, else our awareness would by definition be stuck only at certain levels of that associative hierarchy.
Consider the stunts of Derren Brown, where he depends on people acting on associations they weren't aware of making. Our brains are sorting a complete hierarchy of associations, in which we are only aware of a small subset we pay attention to. Yet the ignored information contained in the full complement of associations is not lost, it plays a role in our actions.
This is why we can't describe the mind in terms of inputs and outputs. Because the process between these inputs and outputs are also inputs and outputs, such that we can reference back to any subset of that process. An abacus cannot, even by any stretch of the principles, do this. However, molds and plants have some limited capacity to do this, even in the absence of neurons. That's another reason, besides being distinctly mechanical, I use metronomes as a toy model, where the synchronization potential becomes the salient feature for the emergent behavior. The RFID ideas is more realistic, as metronomes, even overlooking engineering issues, are strictly limited in dimensionality. The brain neuron maps can be modeled as if it was taking advantage of thousands of dimensions, defined by connectivity.