This is great.
The way Glen frames the up-in-the-air parts is IMO the direction to a great
program of questions. Many of them don’t seem to have been well addressed in
general terms, but there is a ton of lovely casework (particularly by
developmental biologists), giving some part of the variety we need to include,
and suggesting how diverse it can get.
On Apr 9, 2026, at 0:03, glen <[email protected]> wrote:
I feel like I'm going to regret this post. But what the hell, eh?
I was down with an unordered:
• collection of features
• measurement method
• successive execution of the method
• relative rates of change
So at the end of EricS' post, including the parse of "cause", I felt like I was stable. But then EricC goes on about
"animals change", "species change", and "organisms". OK, to be fair, EricS did use "generations"
and "reproductive". So there's an implicit ... what? ... unit, atom, ... thing in there that we all agree is doing the
_generating_ [⛧]. And it's common for that sort of thing to be entirely latent, occult. But I took EricS' primary criticism as pointing out
that it *is* occult, and speculation about the generator(s) has to take a particular form (namely relative frequency and rates of change of
the features given the measurement method). I.e. any conjecture about the generator has to be grounded in that.
I think there are two things going on here, one fairly ordinary, and the other
a direction for real novelty.
The ordinary part was my hedgy way of talking about frequency of features. I
did that, in space too brief to say anything clearly, to wave at the fact that,
at the end of the day, all of these statistical estimators of
amplification-rate have to be defined on _sets_ of individuals (or of
whatever). That means we have to commit to some way to partition things, and
to treat members within any set in the partition as equivalent, and those in
different sets as distinct, in defining what it means for the degree of
amplification to differ “by type”. So all that falls to the details of what
signal you are trying to test for and what question you want to answer.
The much more interesting thing is all the rest of Glen’s paragraph above. On
this, I think there is a simple way to sketch a program, which will hold up
even though I don’t have interesting things to say about almost-any of the
solution to it.
0. It seems silly, but like number theorists who always want to start by
defining zero, I think the best place to start is with the partition between
pattern-forming in equilibrium systems, and pattern-forming that _must_ come
from non-equilibrium. Here, “equilibrium” means that the inventory of allowed
elementary events and of whatever governs their probabilities are symmetric
under reversal between forward-time and reverse-time indexing, and moreover,
that all waiting times for any barriers are surpassed. The reason this is a
good division is that it reduces the first step in our ontology to
dimension-counting. If all waiting-times are surpassed, then all the
parameters that set them become unable to impact the collection of possible
patterns formed. (To be more concrete, for example, this means we keep
molecule “free energies of formation” in chemistry, but we discard “kinetics”
describing reaction rates.) Clearly there is less information defining
equilibrium systems, and so less variety available to make distinct patterns.
So, first list all the patterns that _can_ form in such equilibrium settings.
0.a. It is useful to note that the attraction to equilibrium outcomes defines
what one will later need to mean by “things fall apart”. (The usual habit of
the public is to go on and on about maximization of entropy etc., but I think
Yeats keeps us better focused on the very non-occult meaning, whereas if you
say “entropy” half the room goes mystical.)
1. Then, we look at all the other patterns that can _only_ be formed in systems
driven away from equilibrium, where the driving boundary conditions interact
with the remaining barrier-parameters in some ways. Because any
extra-equilibrium patterns are always (by definition) tending to fall apart,
they exist only by the grace of some kind of amplification, and this is where
we get to the need to define the right amplification and attenuation
rate-parameters. Generally, it is the non-uniformity in these sets of
parameters that will make concrete whatever we want to mean by “selection”.
1.a. Now, at this point, there is a fun debate with Michael Lachmann. He wants
_everything_ in this category 1 to be called “evolution through selection”,
because he doesn’t see any single bright line that sets one sub-category within
it apart from the others. I fully understand him, but I won’t choose his
practice, because I think we already have a name for non-equilibrium pattern
formation, which entails everything in the bullet 1 above, and I would rather
reserve any new terms, such as evolution-through-selection, for more restricted
sub-classes, and generally those in the general direction of things we see in
biology, which were the exemplars when the terms were brought into this usage.
Michael and I have no trouble talking across this difference of choice, of
course, and one can see it both ways.
1.b. The things I want to cordon off with my narrower
evolution-through-selection are all those in which the states at single times,
and their connections across times, become characterized by all sorts of other
structure. For this we have lots of wonderful terms, even just within biology:
organism, genome, species, genotype, pan-genome, reproduction, lifecycle, etc.
The new snow, as I see it, is to ask: How is it that this driven-patterning
world brings into existence forms of order that want (meaning, justify the use
of) these names, and how do the different ones come about? That question is
open-ended, and the developmental biologists, ecologists, and others, are our
guides in how to do it.
2. I think, even at this very early level, there is lots of clarity that can be
laid out. Here, some suggested definitions, and these are meant to respond to
Glen’s question about “what? units?…” which I think reflects _exactly_ the
right level of reservedness against pre-registration.
2.a. An “organism” will be roughly whatever we want to describe as something
individual, of which many can be gathered into populations, and which has an
associated life-course and lifecycle. All this, of course, to be formalized
carefully per-case.
2.b. A “genome” is (in our biosphere) an integrated collection of elements
capable of carrying the operating system for the lifecycle of some organism.
It is that integration that makes those elements into “genes” as we will later
call them, so a lot goes into how genomes come into the world and why. A nice
feature is that genomes are governed by things like Chris Kempes’s allometric
scaling laws, somehow connecting them to cell sizes, lifecycle times, domain
type, and so forth.
2.c. A “lifecycle” is whatever sequence the operating system of the genome
takes organisms through, as different forms are built up, shed, recombined or
reworked, etc. May not be a simple “cycle”; the so-called “minor taxa” of
things like brown algae show us what a menagerie of active and
environment-responsive management this can be. “Reproduction” is the
choreographed event-sequence that carries out a lifecycle. It is like the
event-dual to the object-state of the organisms in any of the lifecycle stages
of the corresponding species.
2.d. Fascinatingly, the notation of the pan-genome — which will typically (or
at least often) be about an order of magnitude larger in its gene inventory
than what is found in any actual realized genome — is somehow the
resource-cloud that we want to connect with notions of “genotype” and
“species”. It seems that punting things out of most genomes, so that they are
only carried by somebody rarely-attested in the population, to maintain the
“pan-genome”, and then pulling them back in when needed, is a very active and
regulated process. The “genotype” is the thing that “breeds true” in a
species, and that tells what can or can’t ever be in the genome of some actual
organism of that species. So, roughly, its actively managed part defines the
scope of the pan-genome for the species. Clearly, all this gets cloudy and
overlapping. Genetic elements can be in the pan-genomes of more than one
species, and so forth. The degree of overlap often correlates strongly with
the degree of relatedness of different clades. So we have to learn what is the
right mathematical structure to use to describe the actual world faithfully.
But there do seem to be major modes of organization that justify these terms.
As far as I know, the above draft-meanings are idiosyncratic to me, though I
think they reflect the broad usage by the different biology-groups that own
them. I have spent time talking to people on each of these terms, to check
that they find my picture above concordant with their own sense of the
phenomena, and I think I am not violating anything with this way of putting
things.
I think one could do work within this framework,
Eric
