[I was aiming for legibility to a limited extent only. This post got extracted from a bigger post I've been writing and is meant mostly as a reference, and thus it may make more sense in context than in isolation.]

(Spiritually related: Yes, It's Subjective, But Why All The Crabs?^[1])

(Alternative title: Yes, It's A Spectrum, But Why All The Structure?)

Many important concepts are only partially grasped. For some of those, it seems fruitful to identify certain postulated key/primary characteristics that quantitatively distinguish examples from non-examples, putting the former on one "end" of the multidimensional spectrum and the latter on the other "end".

It might seem that this approach has a significant peril because by constructing a continuous multidimensional spectrum to discuss properties of such phenomena, we also cause them to dissipate into insignificance, as they allow for examples of the phenomena satisfying those properties to a minimal extent to fit into the frame. We tried to clarify the concept — find its "True Name", a "natural" boundary separating it from everything else — but our effort turned against us: we dissolved the boundary.

This, however, is not true. We can intuitively recognize certain "clear"/"unambiguous"/"paradigmatic" examples of the phenomena. It does not necessarily give us that much information about where exactly the boundary is between the paradigmatic examples of the category and other phenomena. It is often probable that the boundary — insofar as it makes sense to conceive of it at all — is actually rather vague.

Nevertheless, certain regions of the phenomena are characterized by scoring high on the primary characteristics in terms of which the space exhibits interesting characteristics as a result of having a certain combination of the primary characteristics.

It is all a spectrum. But look! This region is emptiness, devoid of life. Most that is not void is inert dust. But that little corner over there — even if the coordinates I know are only approximate — is where interesting stuff happens.

I am going to give between two and four (depending on the way of counting) examples to illustrate what I mean by this and why this might be a good way to think about this.

Godfrey-Smith Cubes

In Darwinian Populations and Natural Selection, Peter Godfrey-Smith (PGS) introduces several characteristics of populations of biological organisms that are crucial from the perspective of enabling evolutionary dynamics. Among others, he singles out fidelity of heredity, dependence of evolutionary fitness on intrinsic properties (i.e., those of the organism, rather than contingent facts about the environment), and smoothness of the fitness landscape. "Paradigmatically Darwinian populations", those evolving populations in which significant novelty can emerge and can give rise to complex and adapted structures (to use Godfrey-Smith's terminology), score high on all three, with "less-paradigmatic" populations taking in-between-ish levels.^[2]

Two chapters later, PGS defines collective reproducers as entities capable of self-sufficient reproduction that are composed of entities that themselves are self-sufficient reproducers.^[3] Here again, he introduces three organizing features of collective reproducers: bottleneckishness (B) (the narrowing down of scope/size/number of lower-level units transmitted between generations), germ line sequestration (G) (the degree of reproductive specialization of parts), and integration (I) (division of labor/mutual dependence/loss of autonomy of parts, the maintenance of a boundary between a collective and its outside).

The relevance of those three is that the higher B, G, and I, the clearer the distinction between reproduction and other reproduction-like phenomena, such as growth. This is relevant if we want to talk about the possibility and coherence of phenomena such as group selection or "cultural evolution"^[4]

Sometimes we can define/delineate/[point at] a certain phenomenon in terms of several features (that we take as primary/generator-like/defining/particularly informative), such that, even though this description admits uninteresting, degenerate examples, there is some vague region in this space in which interesting things start to happen, because the combination of high degrees on the relevant characteristics causes an interesting, unique dynamic to emerge.

In From Bacteria to Bach and Back, Daniel Dennett took inspiration from PGS's cubes and created a few of his own to illustrate similar multidimensional spectra. For example, here is one illustrating the spectrum from Darwinian phenomena at (0,0,0) to intelligent design at (1,1,1), which thus warrants gluing it to the (1,1,1) corner of PGS's first cube.

And here is Rosa Cao's from her talk "Agency and giving a damn":

Lyfe

In Defining Lyfe in the Universe, Bartlett and Wong want to … define life, except without anchoring too much on the contingent features of Terran life.

We seek to reframe the definition of life in a more expansive way while recognizing the need to signify the specific kind of life that earthly forms represent. Thus, we have come up with a new term—lyfe^[5]. Henceforth, we will refer to life (as we know it) and lyfe (as it could be, in the most general sense). The two designations are distinguished as follows:

• Life represents life as we know it; it uses the specific disequilibria and classes of components of earthly life.
• Lyfe represents any hypothetical phenomenon in the universe that fulfills the fundamental processes of the living state, regardless of the disequilibria or components that it harnesses or uses.

They propose "four pillars" of lyfe: dissipation, autocatalysis, homeostasis, and learning. All of them are strictly necessary for lyfe, but incomplete combinations also yield interesting categories of phenomena.

The regions labeled as 6, 7, and 8 correspond to "almost lyfe", phenomena missing exactly one of: autocatalysis, homeostasis, or learning.

6. Dissipation, autocatalysis, and learning: A living system that wipes itself out by tragedy of the commons. Examples might include invasive species introduced to an island that destroy their food sources so fast that the food sources are damaged beyond recovery. One might also suggest anthropic climate change as another example. Note that these cases depend critically on where one draws the boundary of the system (e.g., to include humans or not). Indeed, this form of sublyfe or sublife is less likely to occur because if the system is capable of learning, then in principle it could learn how to regulate itself homeostatically (unless it cannot learn fast enough).

7. Dissipation, homeostasis, and learning: A “smart” house thermostat that monitors occupant behavior over time. This system cannot replicate but consumes free energy, is capable of primitive learning, and can regulate its local temperature.

8. Dissipation, autocatalysis, and homeostasis: Thermal Gray–Scott reaction–diffusion spots. Certain nonequilibrium chemical patterns have been shown to grow exponentially and also regulate their local temperature.

Closing remarks

• The instantiation density of the space is far from uniform. Some configurations/[regions in the space] are empty/uninhabited or at least very unlikely to be instantiated for various reasons.
  • For example, lyfe minus homeostasis is unlikely to occur in the first place, because getting to a point when a phenomenon merits being called "lyfe" demands some amount of prior homeostasis, as otherwise the process that eventually led to it would have terminated before (unless it's intelligently designed).
  • Very high heredity (H) stalls evolution and thus makes a given Darwinian population more vulnerable over time to changes of the environment that its genes are not well-prepared for. Over time, we should expect the Darwinian lineages that persist to have evolved sufficient evolvability to prevent an excessive degree of H.
• Directionality of the space. If we think about an evolutionary or developmental lineage as a trajectory (or a tree/branching of trajectories), this gives a certain sort of directionality to the space. You can draw an arrow from region/point A to region/point B if you can plausibly expect A to transform to B.
  • It is, however, plausible that key determiners of this directionality are not necessarily included in your choice of the dimensions. It might then be the case that you need to expand the space to see the directionality, but it might also make it less wieldable.
    • In general, the identification and selection of relevant dimensions is plausibly the greatest difficulty here.
• The questions to ask:
  • What goes on in various regions? Why?
  • Why are the (likely (very) fuzzy) boundaries where they are? Why?
• Expanding the domain of discourse reveals structure already there but hidden. If you cannot see an interesting structure in the space of X type of phenomena, a fruitful move might be to generalize to some Y type of phenomena, of which X is a special case, which collapses some interesting dimensions, but that need to be considered in their entirety if you want to get a glimpse of the full structure.

1. ^{^}

   But see here for a contra to this specific example

2. ^{^}

   PGS introduced more characteristics, but, alas, drawing more-than-3-dimensional cubes is kinda wonky.

3. ^{^}

   See also: Scaffolded Reproducers, Scaffolded Agents.

4. ^{^}

   Both of which the author has opinions on, but I'll let you read the book.

5. ^{^}

   Pronounced "loyf".