Dear Conrad --
Thanks for your remarks here. I want to push on your response a little because my goal was to go as far back as possible: to ask what what's required to get us beyond fundamental physics. If we don't push as far back as we can, we're not truly dealing with an origin problem, but just saying interesting things as amateur sociologists, or evolutionary biologists, biochemists, neuroscientists...
Pushing further back does get you some things that you will want as well. It clears up some technical questions. And it suggests new ways to distinguish derived aspects of life-like behavior from fundamental ones. So for all those reasons it's worth highlighting where we differ in emphasis.
Let me reply to two things of interest:
These are surely important aspects of the naturally emerging technologies through which "systems gain new powers" that lead to new and unpredictable phenomena. But there are others equally important, e.g. anticipation, or communication between systems.
Putting anticipation aside for a moment, and driving back to the fundamental level, our most basic notion of "communication between systems" is correlation. And correlation is insufficient to get what we want. Fundamental physics already has a richly correlated field structure, and there are plenty of interactions in the standard model between the fields. Including self-interaction terms! But you need renormalization to get the memoryful higher-derivative terms. For many reasons, including the instability problems when these are taken are (incorrectly) taken to be fundamental.
Now when you write "communication between systems", you may have in mind boundaries (spatial or otherwise), and all sorts of other interesting phenomena. I'd certainly agree that these will get you very far. But this is something at a later stage. Lurking behind the ability to create boundaries is the ability to create memories (which might be thought of as having a persistent correlation between environment and "description") and self-reference (the creation of dynamical correlations within that description space).
As for anticipation, this to me seem to be a memoryful process as well: the need to compare representations of an expected state with a current state.
Likewise it seems to me that the possibility of "coarse-graining" information arises from the emergence of new kinds of systems, not the other way round.
It is definitely the case that living systems coarse-grain their environment. The recognition of this fact is one of the biggest insights in the information-theoretic study of living systems; I don't know when it first emerged in the literature, and I'd appreciate references if you know of any going further back (take a look at the papers I refer to in the essay).
In any case, when we want to push our explanations all the way back to t=0, we face a problem. Life-like behavior (computation, reproduction, etc.) requires memory. But the fundamental equations do not allow for memory. So we're stuck. Nothing gets off the ground. But we can use the coarse-graining arguments to say, well--we don't see it at the microscopic level. But it does appear at the macroscopic level.
By the way, one alternative that is not in the essay is the possibility that this memory is "baked in" to the initial conditions--that somehow, we get a few random hotspots arranged in such a way as to make a memory system. I haven't thought too deeply about this idea, and it may be possible to make it work in some kind of Darwinian fashion (a very large universe with a vast array of causally-disconnected initial conditions chosen "randomly" to avoid fine-tuning).
One other remark you make:
The "self-reference" at the basis of biology, for example, is specifically self-replication, the ability to create new copies that create more copies. ... Which is all very different from the kind of self-reference we humans discover as kids, as we first learn to talk with other people, and then later gradually being talking with ourselves.
They certainly look different. At the same time, they share something crucial in common. And it's that shared feature, self-reference, that gives their origin problems a similar feel and structure.
Best wishes from 30,000 feet. I look forward to reading your essay in the coming week.