Agent Above, Atom Below: How agents causally emerge from their underlying microphysics by Erik P Hoel

Over on Shtetl-Optimized, you mention "I'm not answering all things people toss out as I don't want to spam the thread." I expect you're not even referring to me, but FWIW I'll repeat my comment there (which reels in a little my comment above that I wouldn't go to the wall over the choice of mapping),

«Hoel works with Markov processes, but specializes to systems that have an exact macro-state separation: that is, having a diagonal block-matrix presentation. Physical systems, however, are only separable in this way for limited lengths of time. There are small probabilities, for example, of interactions between my fingers, subsystems of my body, and my toes; infrequently, I have to cut my toenails. In general, every entry in a Markov matrix is likely to be non-zero and different from other entries.

To identify how to coarse-grain in the general case, we have to consider either the matrix entries or we have to consider whatever information is not encoded in the Markov matrix (physical adjacency, for example). If the former, an algorithm might, for example, compute an eigenvector basis (over the complex numbers) and compute which to discard (but there's likely a better algorithm!); the algorithm is clearly a source of extra information. If the latter, there's an even clearer source of extra information. In both cases, the Markov process is embedded in a larger system of other degrees of freedom, but we can't just move Hoel's argument to that larger system, because the same problem applies to it (perhaps more so, because now either the matrix or the external information is more complex).»

Perhaps even this doesn't engage with or misunderstands your approach too much for it to be fruitful to engage with it? I always thought my reach into Axiomatic QFT (or more generally, any algebra of observables approach) would likely be too much in my own way of thinking for you, but I'm curious whether you think this later thought illuminates your formalism.

Thanks, that seems a good answer. It's rather open-ended, but that's OK.

It felt to me that Scott was talking rather at cross-purposes to you, and particularly that he and other commenters on the thread were mostly too far away from the Markovian process mathematics for the discussion to be very productive, so Kudos for removing yourself from it so cleanly.

As I've mentioned above, my focus is on algebra of observables approaches to Physics, including algebraic QFT but also classical random fields, where the relationship to space-time is firmly specified; Markovian processes are certainly useful as models, but they are rather removed from the QFTs that physicists take seriously as fundamental physical models almost to the exclusion of all other models. Not that I take physicists necessarily to be right about that, but I suppose that if one wants to talk seriously to physicists and to talk about mind supervening on physics some moderately precise link to space-time does have to be made (albeit at the Planck scale totally precise is unlikely). The link between Markov models and consciousness seems also too tenuous to me, but I don't have a mathematical or other framework in which to discuss how it is tenuous.

Perhaps finally, thanks for mentioning the "Markov blanket" idea, which I had not come across before.

I'm going to put my worries about the construction in terms I'm familiar with, quantum mechanics, but which I hope will not be too unfamiliar for you, Erik, instead of in the terms you have been using, which I take to be of stochastic matrices and stochastic vectors representing states.

In the QM context of Hilbert spaces and the representation of states by density matrices, von Neumann entropy, [math]\mathsf{vN}(\hat\rho)=-\mathsf{Tr}[\hat\rho\ln\hat\rho],[/math] is perhaps the commonest measure of information (there are certainly others, but just replace von Neumann by some other measure everywhere in what follows). If we take coarse-graining to be a map, which in general will be nonlinear, from n-by-n density matrices to m-by-m density matrices, [math]X:\mathcal{D}(\mathcal{H}_n)\rightarrow\mathcal{D}(\mathcal{H}_m);\hat\rho\mapsto X(\hat\rho),[/math] then the von Neumann entropy of the coarse-grained system will be [math]\mathsf{vN}(X(\hat\rho))=-\mathsf{Tr}[X(\hat\rho)\ln X(\hat\rho)].[/math] It seems clear(?) that how the von Neumann entropy of the coarse-grained model will be different from the von Neumann entropy of the pre-coarse-grained model will depend both on what choice we make for X and on what density matrices occur as pre-coarse-grained models.

I think my problem is that it's not clear to me that a detailed accounting of where there is more or less information is a specially good way to think about causality or consciousness. In QM, what "causes" what is systematically described by the Hamiltonian, defined as the infinitesimal generator of the evolution of vector states as a function of time; if one works with a Hilbert space of vector states and with density matrices, as Physics now kinda does, that's it.

Now, not very seriously, I'll go off the deep end... There are possibilities for QM to model consciousness insofar as we can construct, not for a toy model of a few dimensions, perhaps, but for a sufficiently large Hilbert space, a measurement operator C that returns "1" if there is at least one conscious agent or "0" if not, as well as more physical measurements such as the intensity of the electric field. The algebraic relationships of C with the Hamiltonian and with whatever coarse-graining operators we use would determine what measurement results would be expected in a given state and at different levels of coarse-graining (which would include, for example, correlations between consciousness and the intensity of the electric field). Ensuring such models are empirically useful is of course not so easy.

I believe it was David Hume who pointed out that murder doesn't exist at the micro level--where at a physics level does an ethical wrong get encoded?

I'm wondering about macro items other than agents--the meaning of a printed word doesn't exist at the micro level, nor does a photograph exist in its pixels. Temperature--or any average or similar calculation--exists at a different "level" than that of the atoms whose root mean square velocity it measures.

I'm not clear on how the exclusion principle can exclude things that can't exist at micro levels, such as an image, a murder, or a feeling.

Maybe I'm completing missing something?

Your article is so beautifully and clearly written.

Thanks!

Sanjay

Thanks for your essay,

1. Fisher proposed how nuclear (quantum) entanglement might be stable over mesoscopic distances in the brain [1].

2. i don't understand causality in quantum mechanics too well, but this is cool. they've demonstrated a computational speedup for a parity task with a single qutrit [2]. it takes one evaluation in the quantum case but two evaluations in the classical computation. it seems like there's a notion of 'quantum causality' going on that's not an averaging over (or function of) the microcausal chains, since they take too long.

[1] https://www.theatlantic.com/science/archive/2016/11/quantum-brain/506768/

[2] https://www.nature.com/articles/srep14671

Paul

Thank you so much for this explanation. I get it. Utter reductionism.

OK, so as cars pull up to a stop sign, and one by one, stop, the exclusion principle is telling us that the configuration of the word STOP, and its processing by each driver's brain is in no way causative of the drivers' stopping, that the word STOP is an epiphenomenon of the underlying physics between the lightwaves reflecting from the stop sign, the drivers' brains, and the brake pads (all of which are epiphenomena too).

Furthermore, if I can reliably predict that certain large groupings of atoms, whom I call competent drivers of functional cars, will "stop" at this other grouping I call a "stop sign," I seem to be encoding a level of information that is real and substantive, as proven by my ability to do it. If not, then proof, math, all human concepts disappear into the exclusion principle as meaningless, leaving us with no "real" basis for our lived sense of meaning in the world.

If you have in fact found a way to bridge this gap, and resurrect the reality of human (and other agents') thoughts and experiences, I'm in awe. I'm re-reading now, trying to better understand.

Hume's argument is in Part I of "A Treatise on Human Nature, 1739:

Take any action allow'd to be vicious: Willful murder, for instance. Examine it in all lights, and see if you can find that matter of fact, or real existence, which you call vice. In which-ever way you take it, you find only certain passions, motives, volitions and thoughts. There is no other matter of fact in the case. The vice entirely escapes you, as long as you consider the object. You never can find it, till you turn your reflexion into your own breast, and find a sentiment of disapprobation, which arises in you, towards this action. Here is a matter of fact; but 'tis the object of feeling, not of reason. It lies in yourself, not in the object. So that when you pronounce any action or character to be vicious, you mean nothing, but that from the constitution of your nature you have a feeling or sentiment of blame from the contemplation of it. Vice and virtue, therefore, may be compar'd to sounds, colours, heat and cold, which, according to modern philosophy, are not qualities in objects, but perceptions in the mind...

Would it change anything if instead of giving the state of group A as {1,0,0,1,1,1,0,1,1,0} or as the coarse-grained {6}, you give it as a function of both, {1,0,0,1,1,1,0,1,1,0,6}?