NOTE-111: Baedke et al. — Unknotting reciprocal causation

Contribution

The paper diagnoses two problems it finds in organism–environment reciprocity theories from Haldane, Uexküll, Meyer-Abich and Waddington through to current niche construction theory (NCT) with its developmental-systems strands: (i) organism and environment are merged, and (ii) experiential and physical reciprocal causation are not integrated. It then offers a single diagrammatic causal model that it claims solves both. What is new after it is a shared notation: ordered O/E/Ex state series with interventionist arrows. In that notation each niche-construction type in the Aaby–Ramsey (2019) and Chiu (2019) taxonomies has a characteristic causal pattern, and multi-species cases that mix types can be drawn in one diagram. The authors explicitly do not propose a new taxonomy of NC types (p. 4).

Key insight

A reciprocal loop between organism and environment looks inextricable only because it is drawn as a loop. Add intrinsic causal paths within each relatum (Fig. 2b) and unroll the loop in time (Fig. 2c–d), and “reciprocal causation” becomes an ordinary acyclic sequence of states, O_n → E_{n+1} → O_{n+2} and so on. Its arrows can be weighted by relevance to a given causal narrative and read as interventionist dependencies. The organism–environment boundary is then not a metaphysical discovery but an “epistemic necessity” (p. 3; p. 24) for apportioning causal contributions. Inseparability views, the authors argue, do not remove the problem of individuating biological units; they only “push it aside” (p. 12).

Assumptions

• Diachronic, Humean causation. Effects follow causes over time. Synchronic co-constitution (Haldane; Walsh 2021’s “the effect x has on y at time t is dependent upon the effect y has on x at time t”) is set aside, not refuted (fn. 3, p. 11; p. 11).
• Interventionism. An arrow between two states means the later state is invariant-dependent on the earlier one under a range of counterfactual interventions (Woodward 2003; p. 14).
• The boundary is epistemic. The model “does not develop a particular theory of biological boundaries” (p. 13, fn. 5). It grants organism and environment are distinct in order to trace causal relations (p. 14). An ontological inseparabilist is not argued out of that position. The authors ask only how the position would be translated into practice (p. 25).
• States are not regular time steps. Organismic and environmental series can run at different time scales (fn. 6, p. 14).
• Sensory specificity of Ex. The experienced environment is the sum of cues that affect an organism through a sensory–cue interface with “a sufficient degree of sensory specificity” for minimally directed functional reaction (p. 20). Not every influence counts as experienced; ionizing radiation that alters DNA without sensory mediation is the counter-example (p. 20).
• Selective highlighting. Unhighlighted arrows still exist; the explanation abstracts from them “to some degree” (p. 14).

Key results

• Historical diagnosis (pp. 4–8). Organism–environment reciprocity is old: Haeckel 1866, J.S. Haldane 1884/1935, Whitehead, Uexküll’s function-circle (Fig. 1), Meyer-Abich’s holobiosis, Waddington’s “exploitive systems”. It declined for three reasons: problem (i), made concrete by Needham’s (1936) objection that without an organism boundary no line can be drawn anywhere; problem (ii), the objection that Uexküll’s sense-physiological environment is too narrow and that subjectivity was suspect (Hartmann 1950, Tinbergen 1963); and a wider gene-centred, externalist turn.
• Recurrence in NCT (pp. 8–12). Early NCT (Odling-Smee et al. 1996, 2003) kept organism and environment separate but ignored experience. Later DST-influenced NCT (Laland and O’Brien 2011’s “cycles of contingency”; Sultan 2015; Walsh 2015, 2021) brought experience in but merged the relata. The authors call these “anti-individualistic positions” that are methodologically problematic because NCT needs the organism as an individuated causal agent (p. 11).
• Model, physical NC (Fig. 2d–e, pp. 13–15). Reef-building corals: CaCO3 secretion (O_{n−1}) → new competitor species in E_n → effects on coral O_{n+1} → a new round of NC.
• Model with Ex (Fig. 3a, pp. 16–17). Ex is a mediating interface. A change in Ex is a change in the organism’s relation to its physical environment without a change in the environment’s intrinsic properties (after Chiu 2019).
• Characteristic patterns (Figs. 3–4, pp. 17–19):
  • Constitutive NC: a lion’s maturation changes Ex of prey.
  • Mediational NC: Arabidopsis root architecture changes Ex from nutrient-poor to nutrient-rich; the transgenerational version is meadow-vole coat thickness.
  • Relational NC: mice piling up to change Ex of a cold nest without changing its temperature.
  • Relocational NC: a break in the E-series, as when anadromous fish migrate through transitional E_τ to river E′.
• Integration (Fig. 5, pp. 20–23):
  • (a) Bumblebees damage eggplant leaves, and the plant flowers early. This is physical NC for the bee but mediational NC for the plant, so the case requires both types plus experience.
  • (b) In ruminant herbivory, microbes do physical NC on the rumen, and the host does mediational NC (it comes to perceive plants as edible). This works only if microbes and host are kept distinct. Chiu and Gilbert’s pair of claims, that each partner is part of the other’s environment and that the holobiont is the unit of mediational NC, is argued to be inconsistent.
• Heuristic claims (pp. 24–25). The diagrams are directed acyclic graphs and could guide the choice of intervention variables. Formalization “elsewhere”.

Claims

id	claim	strength	support
C1	Organism–environment reciprocity theorizing predates Levins & Lewontin by decades (contra Svensson 2018)	strong	primary historical sources quoted: Haeckel 1866, Haldane 1884, Uexküll 1928, Waddington 1959 (pp. 4–6)
C2	Early reciprocity views declined partly because of problems (i) merging and (ii) non-integration	moderate	quotations from contemporaneous critics (Needham 1936, Hartmann 1950, Tinbergen 1963); causal attribution of the decline is interpretive and shares weight with the gene-centred turn (pp. 6–8)
C3	The same two problems recur in current NCT and in DST-influenced accounts	moderate	textual readings of Sultan 2015, Walsh 2015/2021, Laland & O’Brien 2011 (pp. 9–12); selective, and each critic could dispute the reading
C4	Walsh’s (2021) definition of reciprocity is synchronic co-constitution, not reciprocal causation	moderate	informal argument from the definition’s “at time t” (p. 11)
C5	The unrolled O/E/Ex model can represent every NC type in the Aaby–Ramsey and Chiu taxonomies by a characteristic causal pattern	moderate	worked by example for each type (Figs. 2e, 3b–c, 4a–b); no proof of completeness or of uniqueness of patterns
C6	The model integrates physical and experiential NC in multi-species cases	moderate	two worked cases (Fig. 5); illustrative only
C7	Holobiont-as-individual collapses the ruminant case into constitutive NC and loses the microbes’ NC	moderate	argument from the diagram (p. 22)
C8	The organism–environment boundary is an “epistemic necessity” for understanding complex causality in NC	weak	asserted in Introduction and Conclusions; supported only by the illustrative cases and the Needham individuation argument. Inseparabilists are asked to show a practice, not refuted
C9	The model “demystifies” organismic experience and aids experimental design	weak	assertion (pp. 24–25); no experiment or formal model given; formalization deferred

Method

A conceptual and visual model, built in four steps:

1. Draw organism O and environment E as a reciprocal loop (Fig. 2a).
2. Add intrinsic causal loops within each (Fig. 2b).
3. Unroll into a time sequence (Fig. 2c).
4. Represent two ordered state series O_1…O_n and E_1…E_n, with arrows E_n → O_{n+1}, O_n → E_{n+1} and within-series arrows. Each arrow is an interventionist dependency (Fig. 2d).

For a given causal narrative, highlight the explanatorily relevant path and number its steps (Fig. 2e). To add experience, insert an Ex_n node between E_n and O_n (Fig. 3a). For multi-species cases, run parallel O/E/Ex and O′/E′/Ex′ series with cross-links (Figs. 4a, 5). A relocational case appears as a discontinuity in the E-series (Fig. 4b).

Concepts

• Reciprocal causation — a feedback loop between two interacting, yet separate, entities or processes (after Walsh 2015:180; p. 3). It is distinguished from ontological co-constitution, mutual structural fitting (Henderson 1913) and concomitant reaction (pp. 2–3).
• Problem (i) — failure to maintain meaningful, epistemically set boundaries between organism and environment; the two are merged instead.
• Problem (ii) — failure to integrate experiential and physical reciprocal causation. “Integration” means first distinguishing the two, then giving criteria for how they combine (fn. 4, p. 12).
• Experienced environment (Ex) — the sum of environmental cues that can causally affect the organism’s sensory–cue interface, and thereby the organism. A change in Ex changes the organism–environment relation without changing the environment’s intrinsic properties (pp. 16–17).
• Physical / constitutive / mediational / relational / relocational NC — the types from Aaby & Ramsey (2019) and Chiu (2019), summarized in Table 1, p. 10. Here constitutive, mediational, relational and relocational are all treated as forms of experiential NC (p. 17).
• Epistemic boundary — a boundary granted in order to apportion causal contributions, with no claim about metaphysical separability (pp. 13–14, 24–25).

Connections

The paper builds on Aaby & Ramsey (2019) and Chiu (2019) for its NC taxonomy, and on Woodward (2003) for its reading of the arrows. Its targets are inseparability views: J.S. Haldane, Uexküll, Levins & Lewontin, Griffiths & Gray / DST, Sultan 2015, and Walsh 2015/2021. In the record it pairs with Hazelwood (LIT-156). Hazelwood’s close reading found that his target is Laland and colleagues’ selection–niche-construction reciprocity, and this reading confirms that the two papers do not collide. Baedke et al. stay with organism–environment arrows and do not model selection as a relatum, so they neither take nor answer Hazelwood’s dilemma. Combined, they suggest a division: organism–environment reciprocal causation, unrolled diachronically, is unproblematic; selection-involving reciprocity is where the metaphysical trouble lies. The paper’s individuation argument (Needham; the holobiont case) connects to the record’s organism and individuality works: Baedke’s own The Organism (LIT-167), Griesemer on scaffolding individuality (LIT-188), and the individuality items the dossier names (LIT-157, LIT-168). Trappes (LIT-169) thanks these authors, and they thank her; the two papers come from the same Bielefeld/Bochum NC3 milieu. Her analysis of what the niche is complements their analysis of how niche construction is causally structured.

Bearing on the philosophy-of-science tag. The work addresses causation and explanation in a scientific theory: how to represent and apportion causal contributions in a feedback system that evolutionary biology claims is reciprocal. It also touches individuation of the theory’s relata. Its position is epistemic and interventionist. Reciprocal causation is real but diachronic, and it is tractable once unrolled. The organism–environment boundary is a methodological necessity rather than a metaphysical thesis, and the co-constitution metaphysics is set aside. The tag is justified, and philosophy-of-science is the right primary topic, with natural-sciences second.

Bearing on the record

It supports reading LIT-156 as a critique of selection-involving reciprocity, not of organism–environment reciprocity. NOTE-111’s Open questions should drop “the main pro-reciprocal-causation paper that Hazelwood argues against”, as LIT-110 v2 already did. No THEORY codes are affected in nucleation. For ML practice: nothing. The O/E unrolling is structurally the same move as unrolling a recurrent system into a DAG, but the paper carries no instruction for machine-learning practice, and none belongs in the Anthology of the SOTA.

Note of 2026-10-16: THEORY-446 now cites LIT-110 as a source.

Limitations

• The model is illustrative. The paper gives no formal semantics, no completeness result for the “characteristic patterns”, and no empirical test. Formalization, as DAGs or path analysis, is deferred (p. 25).
• Unrolling a loop into a DAG is standard in causal modelling. The novelty lies in the application and the Ex node, not in the technique. The paper does not engage with the causal-modelling literature on cyclic models or equilibrium feedback beyond citations (Pearl 2009; Otsuka 2015).
• Synchronic co-constitution is set aside, so a Walsh-style inseparabilist is not answered, only redescribed. The claim that the boundary is an “epistemic necessity” (C8) is stronger than the argument, which shows only that the model can draw a boundary usefully.
• What counts as “experience” in plants and microbes rests on an unanalysed notion of a sensory–cue interface with “sufficient” specificity (p. 20).
• The selection side of NC is not modelled: how Ex changes selection pressures is named as an open problem (p. 16) and left open.

Open questions

• Does the promised formalization (DAG or path analysis) exist, and does it deliver identifiability of the causal contributions it says the diagram apportions?
• Can selection be added to the O/E/Ex diagram as a node or arrow without falling into Hazelwood’s composition problem (LIT-156)? A combined diagram would settle whether the two papers are compatible.
• How does a change in Ex alter selection pressures or establish individualized niches (p. 16)? This links to the individualized-niche work of Takola & Schielzeth (LIT-127).
• Is there a non-arbitrary criterion for which arrows to highlight, or is the explanatory narrative doing all the work?

Corrections to the seeded skim

• The dossier’s and NOTE-111’s Open questions still call this “the main pro-reciprocal-causation paper that Hazelwood (LIT-156) argues against”. The LIT’s v2 history already corrects this, but the NOTE does not. The full text supports the correction: the paper stays entirely within organism–environment reciprocal causation and never claims that natural selection and niche construction cause each other. Its only claim about selection is the general NCT one, that niche construction “may alter selection pressures” (abstract; pp. 8–9). So it does not take the horn Hazelwood rejects, and it does not answer his composition-versus-causation worry either. That worry is simply outside its scope.
• The dossier gives Fig. 5 as the bumblebee case alone. Fig. 5 has two panels: (a) Bombus terrestris damaging Solanum melongena leaves to hasten flowering (Pashalidou et al. 2020), and (b) the transition to herbivory in ruminants through rumen microbes (Chiu and Gilbert 2020). Panel (b) carries the paper’s argument against the “holobiont-as-individual” view (pp. 21–22): collapse microbe O′ into host O and the case becomes indistinguishable from constitutive NC.
• The skim reads the decline as “early organism-centred reciprocity was marginalised as externalism”. More exactly, it was marginalised in favour of an externalist explanatory logic (Brandon’s shared selective environment; Falconer’s environment-as-noise) and of reciprocity between other relata (gene–environment, gene–population, frequency dependence), which the authors say mostly do not encompass organism–environment reciprocal causation (pp. 7–8).
• The skim omits a scope restriction the authors state outright. The model addresses only diachronic reciprocal causation, “within a Humean view of causation” (fn. 3, p. 11). It explicitly does not address synchronic ontological co-constitution, and it reclassifies Walsh’s (2021) definition of reciprocity as co-constitution rather than reciprocal causation (p. 11).
• Constitutive NC (the lion case, Fig. 3b) is treated here as a form of experiential NC (p. 17), whereas Aaby and Ramsey list it separately from Chiu’s experiential types. The skim’s list does not show this regrouping.
• philosophy-of-science tag: justified. Primary topic stays philosophy-of-science.