nucleation

THEORY-607: The individuals that appear in an evolutionary transition in individuality are a summary description of lower-level processes, not a new level of fact: a partition of particles defines collectives when predicting at the coarse grain matches predicting at the fine grain across many traits, and collectives become indispensable, and so quasi-ontological, only because finite measurement and computation make the fine-grained prediction unaffordable

4 min read

Properties4
StatusProposed
Tagscomplex-systems, individuation, philosophy-of-biology
SourceLIT-049 — A coarse-graining account of individuality: how the emergence of individuals represents a summary of lower-level evolutionary processes
RivalsTHEORY-141 — In a major evolutionary transition, where formerly free-living units come to form a higher-level unit, the collective is stable only if selection between collectives comes to trump selection among the units within them, typically through mechanisms such as policing, kinship and division of labour that align the interests of the units; the levels at which selection acts are then a product of evolution, to be explained rather than assumed

Graph View

Source

Bourrat (2023), LIT-049, Biology & Philosophy 38:33, read in full in NOTE-064.

What was actually shown

The criterion (Eqs. 1–2, pp. 11–14). Take a partition of particles into collectives and compare two routes to the coarse state at t2: the truthful one (evolve the fine state, then coarse-grain) and the projected one (coarse-grain at t1 by averaging, then evolve). Their difference is δC = C_T − C_P. A partition is acceptable if δC < θ, and the minimal one below θ, δ*C, is “the best candidate for defining a level of individuality” (p. 12). It must hold for the same partition across many traits (p. 13). Where low error could come from ecology alone, as when cells align by chemotaxis toward resources, the test is δC|E, required to stay below θ in every environmental state (p. 14).

The pragmatic argument (pp. 15–17). “There is nothing that exists at the higher level that could not be described or predicted, in principle, for the lower level” (p. 15). But with one operation per entity per step, 100 operations predict 10 entities for 10 steps or 2 collectives for 50 (p. 16). Beyond some horizon the coarse description is the only one available, and the paper takes that indispensability as “sufficient to see them as having an ontology” (pp. 16–17).

The reading of Godfrey-Smith & Kerr (pp. 18–20). At their fifth stage a collective fitness parameter appears with no map to particle fitness. Bourrat reads that as forced by the map being “fundamentally unknown”, not measured or too costly, and concludes that “there is no factual change of the level at which selection occurs during an ETI … for pragmatic reasons, it becomes useful to pretend as if this were the case” (p. 19).

What could have come out otherwise. Nothing here could. There is no model, simulation or data. The only numbers are a mis-averaged angle toy and the operation count. The account is argued, not tested.

What this does not say

  • That minimal prediction error finds collectives. It finds the finest partition: if every particle is its own collective, C_P = C_T and δC = 0. Fn. 12 concedes zero error without within-group variation. The cost term that would prevent this is never part of the criterion. δC is also signed, so “δC < θ” accepts any negative error.
  • That conditioning on the environment removes the ecological case. In the paper’s own chemotaxis example, cells are coherent given each fixed resource state. What would exclude it is variation across environments, which makes δC|E an invariance test, not a conditioning one (NOTE-064, C6).
  • That the criterion can be applied where it matters. It needs C_T, a full fine-grained prediction, and the pragmatic argument’s case is exactly where that is unaffordable.
  • That collectives are more than epistemic tools. The abstract says the change of parameters “cannot be dismissed as merely epistemic”. The body’s ground is still pragmatic indispensability under resource limits, and the Conclusion calls individuals “ultimately epistemic tools” (p. 20).
  • That the stage-5 map is in fact unknown in real transitions. No case is examined (C10).

Connections

  • Rival: THEORY-141. Okasha, with Buss and Michod, holds that in a major transition selection between collectives comes to trump selection within them, so the levels at which selection acts are a product of evolution. Bourrat holds there is no factual change of level, only a change in what can be afforded as description, and cites his own work against Okasha’s fitness-decoupling account (C1). Both cannot be right about whether the level of selection changes.
  • The later paper (LIT-005; THEORY-424). Bourrat’s δC|E is the observational, stratify-on-E ancestor of the conditional measure that LIT-005 wrongly equates with the interventional one. Both papers leave scale selection open: here minimal δC prefers parts, there  cannot tell an individual from a part of one.
  • Coarse-graining accounts of emergence. δC = 0 is the commuting-diagram condition, exact lumpability. Dynamical independence (THEORY-257) and effective information (THEORY-222) are criteria that do write down an objective, which is what this account lacks.
  • Collective individuality by organisation (THEORY-418). Griesemer and Shavit put the new level in entrenched interactions and organisational closure, a fact about the collective. Bourrat’s is a fact about the observer’s budget. The two may pick the same collectives, so no rivals edge is declared.
  • Reproducers (THEORY-465), Bourrat’s 2025 framework, inherits the no-worked-example habit both papers show.

Backlinks