Dear Ganesh,
thanks for that.
1. > Let us redraw the boxes as the following. For every system S, we will divide the universe into an observable domain O and the rest as B. O is not fixed and can change in time, thus changing the O-B boundary. Any context system C has to be in the observable domain O of S to be able to affect it at that time, and S also has to be in its own observable domain.
- You seem to be talking of the causal domains investigated in relativity theory.
> If O only comprised of C and S in a particular case, then (O,B) or specifically the time evolution (C,S,B) would fall under the physics category now I think. Hope this helps make my argument clearer.
- Not really. How does this work out for (a) sand grains in a desert, (b) rocks on a planet, (c) biomolecules in a cell, (d) cells in a body.
2. Interestingly if we were to study the correlations between C and S even under the physics category, we can show that energy dissipation minimization under finite complexity constraints (alone) is a sufficient condition for emergence of inference and prediction in such systems.
- What is a finite complexity constraint? if you mean existence of complex entities such as biomolecules, I might believe you. The heavy lifting has already been done in creating those molecules, which cannot be brought in to existence by such principles alone, see e.g. the minimal total energy principle.
3. The learning/inference dynamics is in a very particular manner that the implementation requires an hierarchical feedforward feedback model, the type we see in the brain.
- So that complex system must already exist (at the macro level) and be based in appropriate structures such as neurons and biomolecules (at the micro level). They do not come into existence simply via energy minimisation, which is happier with Boltmann gases and salt crystals.
4. It gets even more interesting when you allow for S to have agency (the ability to act but not necessarily with intent or purpose). The optimal solutions to constrained optimization of dissipation for C-S correlations will involve a very nice 'exploitation-exploration' tradeoff.
- is not "agency" as defined here a biological trait? Neither an electron nor the Moon has agency in that sense. This is like what Hartwell et al talk about.
> Not to mention an hierarchical model that realizes these dynamics will necessitate the 'sense of agency' in the system, and we might be able to identify the source of intention in the agency of the system S.
The Earth and the Sun are hierarchical systems. They have no agency. To realise agency you need physiological systems.
5. In addition to the above, I have shown an interesting way to unify individual learning with England's dissipation driven adaptation and how we could explain the brain as a system exhibiting self-organized criticality and its implications of cognition as input mappings.
- I have not understood this idea of dissipative driven adapation. If it is adaptation, there is some selection principle in action which cannot be captured simply by the idea of dissipation. How does dissipation know that a set of eyes or a pair of wings is a good idea? The need is driven top-down, as I discuss in my essay. That is what is missing in England's proposal, as far as I can see. Incidentally Friston has a similar but perhaps more developed proposal, see A theory of cortical responses
6. I would be very interested in your thoughts if you have the chance to read my submission 'Intention is Physical'.
I will take a look.
George