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

Hi Robin - thanks for commenting. Never apologize for a pun, I loved it.

In terms of purposive behavior and causal emergence; causal emergence can occur without purposive behavior. But I think purposive behavior couldn't exist without there being accompanying causal emergence. I hope it's clear that I think agents causally emerge, assisted by their purposive behavior.

I just read and voted for your essay, and I think it is actually a great overview of some really serious issues. Good to see Smolin, Krakauer, and Braitenberg all tied together in one essay. As to your question about ITI (which I had not heard of until now, so thank you), I remember meeting Krakauer in 2016 and he briefly said it was impossible for there to be any extra information at the macroscale (which, if you're only considering macroscales as zipped compressions is definitely true; however, the theory of causal emergence points out that they can be encodings, not just compressions) so I know he didn't have causal emergence in mind in defining ITI. However, I do think ITI sounds useful for defining the boundaries of systems (another choice is its anagram, IIT: Integrated Information Theory).

Thanks so much for your comment and your essay!

Erik P Hoel

Hi Helder!

Good question about why I'm using discrete and finite systems formalize causal emergence, rather than analog concepts (like feedback, etc). The first reason is that this allows supervening scales to be easily defined and modeled. For instance, one can generate the full space {S} of possible supervening descriptions of any particular system, and then search across that space, as we did in Hoel et al. (2013) "Quantifying causal emergence." Another reason is that information theory, such as mutual information is most often represented as between two finite and discrete variables. A third is that the causal calculus of Pearl is also often represented in terms of Markov chains. So showing how these all can be synthesized is much more direct in these types of systems (applicable to things like cellular automata, etc).

But this doesn't mean linearity / nonlinearity and related concepts doesn't come into play, it just wasn't addressed in this essay. See Hoel (2016) "When the map is better than the territory" of a discussion on how symmetry breaking is critical for causal emergence.

Thanks so much for reading!

Erik P Hoel

Thanks so much Cristi - so glad you found it enjoyable. You immediately hit on one of most interesting questions of this research: how do we related the causal work of the microstates to that of the macrostates. We don't want to multiply entities beyond necessity and have things be overdetermined. There's a few different options - you're right that when it comes to teleological causation (as outlined here) there's less conflict. Just in general causal overfittig and underfitting are nice schemas that outline how it may be non-overlapping in some cases. I give two further options in the endnotes: supercedence (macro entirely constrains or controls micro) or layering (macro contributes what it does above and beyond the micro but micro also contributes). Both of these are viable positions: we argued supersedence in the first paper on causal emergence (Hoel et al 2013) and I argued layering in the second (Hoel 2016).

I just read and greatly enjoyed your own essay - great explanation of how to "zoom" in and out of the different scales and what that means in terms of coarse-grains and thermodynamics. At some point the research on causal emergence should be connected to thermodynamics, given exactly what you're talking about.

Thanks again!

Erik P Hoel

Thanks Daniel, I appreciate it.

I made a comment on your essay so we can have the discussion there - thanks for linking!

Erik P Hoel

Thanks for the read Bill! I'm not sure why those two things would be contradictory in your mind. But I suspect it might have something to do with how I'm using the word "emergence" and what you associate with it.

It's worth noting that emergence can be used semantically in two ways (above you can see George Ellis's comment about this as well). You can use it to say "the patterns emerged from the simple chemical interaction." In this manner it usually means getting something complex from something very simple: it's fundamentally historical. This isn't the usage herein.

There is another way to talk about emergence. For instance, if I had a bunch of NOR logic gates, and I hooked them up to make a complicated circuit that enacts many different kinds of logic functions (like ORs and ANDs and NOTs), you would say that the circuit and other logic functions "emerged" from the underlying NOR gates. This is the way I use emergence in the paper.

Since you asked, here's a brief summary of causal emergence (takes in deep breath): the causal structure of systems can be treated mathematically as a communication channel over which states are sent over time (much like sending messages), and it turns out that describing/observing/intervening upon the system in terms of a higher scale can actually make the channel transmit more information because these higher-scale descriptions are a form of channel coding. Whoof!

Thanks very much Wilhelmus. I didn't explore anything down at the quantum level for my own essay. To me it seems that relying on quantum effects to explain agents is like trying to solve a single mystery by combining two mysteries, which generally just makes everything even more mysterious, but I'm eager to read your essay and find out. I will check it out there.

All the best,

Erik P Hoel

Thanks George. I definitely agree that emergence can be taken in a historical sense (evolution, development, complexity from simplicity) and a level sense (function, scale, causation). It's only in the latter sense that it involves issues of supervenience.

I had a question about your comment: what exactly do you mean by "it is the higher levels that decide what will be done and the lower levels that carry out the work?" Can you explain that a bit more? It may illuminate some of our differences in approach. Because if the lower levels are carrying out the causal work in the system, didn't the lower levels really make that previous higher level decision?

What we are saying is in a sense the opposite of this: causal emergence is only when a macroscale outstrips the microscale in terms of information and causal work, so proving that causal emergence occurs involves directly assessing the causal structure at both the microscale and the macroscale and comparing the two. I think it's really helpful to use simple but well-defined systems like Markov chains for this exact reason: you can derive all the possible supervening scales along with the full causal structure.

All the best,

Erik P Hoel

Thanks so much Robert! Very nice to hear it. Just finished your essay - I greatly enjoyed your breakdown of Maxwell's Demon, and it got me thinking about the long history of cases of trying to get something from nothing. In Maxwell's case, it's a violation of the 2nd law. I think for a long time people have thought of emergence in that way - it's almost like getting something from nothing, because how could you possibly gain any information or causal work going up to a macroscale? It seems like squeezing something from nothing, and I think it is this that's the intuitive force behind the "exclusion argument." But there's a few cases where some people have figured out how to squeeze something from nothing (metaphorically, obviously). One of those is Claude Shannon's noisy-channel theorem. At first it really seems a really noisy channel can only transmit very low amounts of information. Then Shannon showed that through channel coding the information can be radically increased - without altering the channel! By saying that causal emergence comes from treating a system's causal structure as a channel, and that macroscales are encodings for the channel, I'm piggy-backing on Shannon's "something from nothing" proof. So causal emergence is kind of like getting something something from nothing (without altering the system).

Anyways, just wanted to let you know your essay inspired me to think about it with a new analogy.

Thanks so much!

Erik P Hoel

Thanks so much Stefan, I'm so glad you found it intriguing. I definitely plan on investigating it further (when I find the time!). In terms of teleology, I think you're right. However, I'm always wary of those kinds of words, so my own personal stance is to try to explain what looks like teleology (apparent teleology) without coming to overt metaphysical conclusions.

Thanks so much for reading and rating!

Erik P Hoel