THEORY-1288: If a mechanism's behaviour supervenes on its constituents' behaviours and constitution is not causal, every cause of that behaviour is also a cause of some constituent along another path, so no intervention can change the whole without separately changing a part; mutual manipulability by ideal interventions is then unsatisfiable, and once interventions are relaxed to admit these common causes, every model with constitutive dependence has a model without it that implies the same correlations under every manipulation, even in ideal experimental conditions

Source

Baumgartner and Gebharter (2016), LIT-1934, §§3–5, pp. 10–20 of the penultimate draft (NOTE-1493); Baumgartner and Casini (2017), LIT-1939, §§2–3, pp. 6–11 (NOTE-1490); Baumgartner, Casini and Krickel (2020), LIT-1924, §2, p. 6 (NOTE-1494). Romero 2015 (LIT-1933, unread) is cited by all three for the same result.

What was actually shown

The premises. (Sup): every change in a mechanism’s macro behaviour Ψ is necessarily accompanied by a change in some constituent behaviour Φᵢ. (NC): Ψ and the Φᵢ are not causally related; Baumgartner and Casini derive this from spatiotemporal overlap, and Baumgartner, Casini and Krickel from Simultaneity (“Simultaneous changes cannot be causally related”).

Every cause of the whole is a common cause. Let I change Ψ. By (Sup), I’s change in Ψ comes with a change in some Φⱼ, so on a difference-making account I causes Φⱼ as well. One variable causing two can do so along one path or two; (NC) rules out the single path I → Ψ → Φⱼ. So I is a common cause of Ψ and Φⱼ (B&G pp. 11–12; B&C pp. 7–8; BCK p. 6).

Mutual manipulability is unsatisfiable. Craver’s MM needs, for each constituent, a possible ideal intervention on Ψ that changes Φᵢ. An ideal intervention’s paths to Φᵢ must all pass through Ψ (Woodward’s IV.iii). A common cause violates that. So no such intervention exists, and MM entails that nothing constitutes anything (B&G’s INEX, p. 13). Dropping (NC) instead makes constitution an instantaneous, bidirectional causal loop (Leuridan’s conclusion, granted on p. 14).

Relaxing interventions leaves the data uninformative. Woodward’s later definitions let a common cause count as an intervention if its other path runs through a supervenience base. MM is then satisfiable. But each mutual manipulation is a common cause, and “For every model featuring constitution there exists a pure common-cause model that entails the very same correlations under manipulations” (B&C p. 10; B&G Fig. 5, p. 19). The confounding is introduced by the manipulation itself, so it survives an ideal experiment, which in causal discovery would remove it (B&C pp. 10–11).

What could have failed. A whole could have had a cause that left every part unchanged. (Sup) rules that out by definition. Or the single path through Ψ could have been allowed, which (NC) forbids. Each step is deductive. I checked them against the definitions as the papers state them.

The asymmetry that survives. The micro level can be changed without changing the macro level, since several micro states may realize one macro state. So bottom-up interventions that leave the whole fixed can be surgical (B&G footnote 8, p. 13; B&C p. 12). What cannot exist is a cause of the whole that leaves all parts alone.

What this does not say

• That constitution cannot be known. It says first-order intervention data cannot show it. B&G and B&C each propose a second-order, abductive route (THEORY-1294), and BCK a sufficient condition under further principles (THEORY-1293).
• That interventions inside a mechanism are uninformative. The result concerns a whole and its parts changing together. A diachronic cause inside a system, such as a production pathway whose knockout changes a maintained quantity only over time, is a cause, and can be intervened on surgically. By B&G’s own temporal clause (M**) (p. 17), it is not a constituent of what it later changes.
• That the premises are agreed. Leuridan rejects (NC); Krickel 2018 (LIT-1925, unread) proposes reading mutual manipulability causally, between a component and temporal parts of the phenomenon. The result is conditional on (Sup) and (NC).
• Anything about where a system’s boundary lies. All three papers take parthood as given (B&G p. 6; B&C footnote 1; BCK p. 5).
• That simultaneity is sharp. (NC) via Simultaneity draws the line between causation and constitution at zero delay. In a dynamical model a part’s response takes time, and whether it counts as simultaneous depends on the time scale of the description (CLAIM-528).

Connections

• CLAIM-522 is left standing: its separating interventions act on production pathways whose effects on the capacities are delayed, and its question is where a producer sits relative to a declared boundary, not what constitutes the organization (ARG-050).
• CLAIM-515’s third question, “which mechanisms realize those capabilities, shown by interventions that discriminate implementations”, is answerable by intervention only where “realize” is read diachronically (CLAIM-528).
• QUESTION-284 asks whether any evidence tells constitution from causal modification. For part–whole constitution, no finite set of manipulations does; a second-order pattern may.
• CLAIM-150 says identification underdetermines constitution for communicative acts. This is a proved instance of that form for mechanisms.
• CLAIM-388, CLAIM-386. Tests of a collective variable’s closure use micro changes that leave the aggregate fixed, the surgical bottom-up interventions the asymmetry above allows.