Source
Bourrat (2025), LIT-157 (Synthese 205:62), read in full in NOTE-100.
What was actually shown
Reproducee (§2, pp. 4–7). B is a reproducee of A iff B comes into existence after A, A is a cause of B’s existence, and B is of A’s relevant class. The three conditions rule out co-occurrence, mere recurrence and mere production. Choosing the relevant class is pragmatic; there is “no mind-independent way” to fix it (p. 6).
Reproducer (§3, pp. 8–10). A reproducee with reproductive autonomy: “the capacity to reproduce within a range of (relevant) background conditions” (p. 8). Autonomy is graded and comparable only across similar backgrounds. With non-null development every life-cycle stage is “at best a mere reproducee”, so whole life cycles are the reproducers (p. 10).
Darwinian individual (§5, pp. 13–17). A reproducer with high reproductive causal control: heredity that survives as the parent’s description is made finer. Control is distinct from fidelity. A reproducer with 50% colour fidelity whose blue offspring are always azure scores higher than one with 100% colour fidelity and none for shade (p. 16).
Genes (§6, pp. 17–20). Genes and hearts are both mere reproducees on autonomy. They differ in control: a nucleotide switch in a gene reappears in offspring genes, and changes to a heart do not (p. 19). Relative to a cellular background on timescales shorter than cell division, an autonomous retrotransposon is a reproducer and a regular gene is not (p. 18).
Material overlap (§4, pp. 10–13). Any continuous tracking of matter through a reproducer’s life cycle meets hybrid objects holding both future parental and future offspring parts, so overlap is necessary for reproducers. A parent virion causes an offspring virion with no overlap, so it is not necessary for reproducees. Griesemer and Godfrey-Smith partly talk past each other (p. 12).
What could have come out otherwise. A definition could have failed to sort a standard case. The cases were chosen as examples, and the measures that would test the placements were not computed.
What this does not say
- That autonomy and control have been measured. Autonomy is to be measured by conditional mutual information between parent and offspring given the environment, and control by interventional mutual information at finer grains (Griffiths et al.’s INF). Neither is computed. The two autonomy recipes (pp. 8 and 10) are different quantities, and the first can be negative (NOTE-100, C5). The control recipe has no normalisation, so raw mutual information favours finer grains (C13).
- That the autonomy proxy fits the definition. The definition is a range of backgrounds. The proxy measures resemblance given the environment, which is closer to heredity.
- That overlap is necessary in general. The hybrid-object premise is asserted for all reproduction and shown only for retroviruses, via Griesemer (C8). “Material overlap” is never defined.
- That the gene verdict is categorical. It holds relative to a chosen background and timescale, and the paper still states categorical conclusions (C16).
- That control evolves in transitions in individuality. The seeded skim had it backwards: scaffolded collectives get high control “for free” from their cells, and what must evolve is autonomy (p. 21; NOTE-100, corrections).
Connections
- Lloyd’s account of units of selection (LIT-160; THEORY-413) champions Griesemer’s reproducer, which this paper builds on, and objects to “Darwinian individual” as a fusion of roles. Lloyd’s entry predates this paper and is critical of Bourrat’s earlier programme.
- LIT-168 (SEP, Biological Individuals) is seeded as treating genes as Darwinian individuals. This paper denies it, “let alone Darwinian individuals” (p. 21).
- The two formal weaknesses, nothing computed and conditioning on the environment in place of intervention, run through Bourrat’s earlier LIT-005 and LIT-049 as well. This is the one of the three whose main contribution does not depend on its measures.