Baumgartner (2013) showed that the modified interventionist theory of causation (M*)–(IV*) proposed by Woodward is compatible with non-reductive physicalism and even entails the possibility of downward mental causation, but that this compatibility is hollow: reading interventionist correlations as evidence of irreducible mental causal powers requires an unargued methodological principle ((†), to the effect that non-reductive supervenience excludes data-confounding between variables that cannot be independently manipulated); epiphenomenalist structures that generate all observed correlations via purely physical paths are empirically indistinguishable from genuinely downward-causing ones, so epiphenomenalism cannot be empirically refuted; mental causal autonomy is restricted to the first physical link of any causal chain leaving the supervenience base; and mental-to-physical edges enter causal graphs only through a stipulation ((‡)). The problem is to determine whether any manipulability-based (interventionist or causal-modeling) framework can ground non-reductive mental-to-physical causation on empirical grounds: establish or replace principle (†), settle whether epiphenomenalist alternatives can be ruled out a priori or discriminated empirically, assess the adequacy of (M*)–(IV*) as a theory of causation, and determine whether any alternative formal modification can both restore compatibility without (†) and (‡) and render mental causation empirically distinguishable from epiphenomenalist physical alternatives. The subsequent literature (causal Bayes-net reconstructions, anti-reductionist interventionism, hybrid causal models, probabilistic and stochastic supervenience, through 2026) has produced candidate frameworks that restore compatibility, but none has been shown to deliver the required empirical distinguishability; the problem remains open.

Background

Non-reductive physicalism holds three tenets: (i) the physical domain is causally closed (every physical effect that has a cause has a sufficient physical cause); (ii) mental properties supervene on physical properties without being reducible to them — supervenience meaning that no two possible situations can differ mentally while being physically identical, while the mental is nevertheless not identical to, or definable in terms of, its physical base; and (iii) mental properties may cause physical effects, including effects of their own physical supervenience bases ('downward' mental causation). Kim's causal exclusion argument (Kim 2005) contends that (iii) is untenable given (i) and (ii): a physical effect already has a sufficient physical cause, and a mental cause that supervenes on a physical base seems causally redundant with respect to that base, so the mental is either excluded or trivially overdetermines the effect.

Interventionism (Woodward 2003) analyzes causation in terms of manipulability: roughly, a variable X is a (type-level, direct) cause of a variable Y iff there possibly exists an intervention on X — a change in some variable I that alters X while being statistically independent of every other cause of Y located on a path to Y that does not go through X — such that Y changes when other suitable variables are held fixed. Several philosophers hoped that this manipulability-based theory could empirically vindicate downward causation, but Baumgartner (2009, 2010) established an incompatibility result: because of non-reductive supervenience, any manipulation of a mental variable is necessarily accompanied by a change in its physical supervenience base, which is itself a cause of the physical effect not located on a path through the mental variable; hence no variable can count as an intervention variable for a mental variable, and the original interventionist definitions plus tenets (i)–(ii) entail that mental variables cause nothing — that is, they entail epiphenomenalism (the view that mental properties are causally inert). Woodward responded by modifying the definitional core: the weakened intervention condition (IV*) exempts causal paths running through a target's supervenience base from the independence requirements on interventions, and the modified cause-condition (M*) exempts variables related by supervenience (or definition) to X or Y from the hold-fixed conditions. On the resulting (M*)–(IV*) interventionism, a mental variable counts as a cause of a physical effect whenever interventions on it are associated with changes in the effect, even though those interventions inevitably also wiggle the supervenience base. Baumgartner (2013) showed that this variant is not only compatible with non-reductive physicalism but conceptually entails the possibility of downward causation — and simultaneously showed that this 'victory' is hollow, for three reasons which, together with a fourth open issue they raise, jointly constitute the present problem. First, treating such interventionist correlations as evidence of irreducible mental causal powers presupposes a methodological principle for which no argument exists: (†) if two causally unrelated variables X and Y cannot be independently manipulated because X non-reductively supervenes on Y, then the correlation of X and Y cannot give rise to data-confounding (i.e., observed dependence that is an artifact of a common physical path rather than evidence of a causal link). The standard analogies — definitional relations such as total cholesterol versus its components, or determinable versus determinate properties such as red versus a specific shade — fail to support (†), because those relations are reductive, whereas non-reductive supervenience is precisely not. Second, to every causal structure featuring downward causation there corresponds an epiphenomenalist structure, in which all observed interventionist correlations are generated entirely by physical-to-physical paths through the supervenience base, that behaves identically in every possible intervention test: the single test that would distinguish them — holding the supervenience base fixed while manipulating the mental variable — is metaphysically impossible. Hence (M*)–(IV*) discards epiphenomenalist structures only through principle (‡), a stipulation preferring graphs with mental-to-physical edges, which on its metaphysical reading begs the question against the epiphenomenalist and on its representational reading reduces non-reductive physicalism to a claim about the proper representation of causal structures rather than about causal powers in the world. Third, (M*)–(IV*) restricts the 'causal autonomy' of a mental property to the first physical link of any causal chain leaving its supervenience base; beyond that link, the mental property's causal impact coincides with its base's. Fourth, it is open whether any alternative formal modification of interventionism could both avoid the original incompatibility between interventionism and non-reductive physicalism without invoking (†)- and (‡)-style principles and render mental-to-physical causation empirically distinguishable from epiphenomenalist physical alternatives. A related non-interventionist route is the counterfactual difference-making account of List and Menzies (2009), on which the exclusion argument fails when mental properties are difference-makers while their physical realizers are not; the present problem concerns the manipulability-based strand, which asks for an explicitly interventionist grounding.

Problem Statement

Determine whether manipulability-based (interventionist) causal theory can ground non-reductive mental-to-physical causation on empirical grounds. Notation: 'supervenience' is the non-reductive dependence relation defined in the Background — no two possible situations can differ mentally while being physically identical, while the mental is not reducible to, or definable in terms of, its physical base; '(M*)–(IV*)' is Woodward's modified interventionist theory, whose cause-condition (M*) exempts variables related by supervenience or definition to a cause or to its effect from the hold-fixed requirements and whose intervention condition (IV*) exempts causal paths running through a target's supervenience base from the independence requirements on interventions; 'principle (‡)' is the principle implied by (M*) that whenever an (IV*)-intervention on a mental variable M_i is associated with changes in a physical effect Z of its supervenience base SB(M_i), M_i is a cause of Z notwithstanding that the intervention is connected to Z also through SB(M_i) — on its metaphysical reading, accepting (M*)–(IV*) thereby presupposes the nonexistence of epiphenomenalist structures, while on its representational reading it stipulates that representations of the underlying causal structure must feature a path from M_i to Z; and 'data-confounding' means observed dependence between variables that is an artifact of a common physical path rather than evidence of a causal link between them. The source (Baumgartner 2013) identifies four constituent issues that jointly bear on this single objective:

(1) Establish, refute, or replace methodological principle (†): if two causally unrelated variables X and Y cannot be independently manipulated because X non-reductively supervenes on Y, then the correlation of X and Y cannot give rise to data-confounding. Either supply an argument showing that non-reductive supervenience (as opposed to reductive dependence such as definition or the determinable–determinate relation) removes the confounding danger when a supervenience base is allowed to vary during interventions on the supervening variable, or provide an alternative warranted inferential bridge from (M*)–(IV*)-interventionist correlations to irreducible mental causal powers.

(2) Determine whether epiphenomenalist structures — structures in which all interventionist correlations between mental variables and physical effects are generated entirely via physical-to-physical paths through supervenience bases — can be ruled out on a priori metaphysical grounds; or, failing that, exhibit legitimate empirical means (available intervention tests or other evidence) that discriminate structures featuring genuine downward causation from their epiphenomenalist behavior-equivalents, given that the one discriminating test (holding the supervenience base fixed while manipulating the mental variable) is metaphysically impossible.

(3) Assess the adequacy of (M*)–(IV*) interventionism as a theory of causation: in particular, whether its restriction of mental causal autonomy to the first physical link of a causal chain leaving the supervenience base, and its reliance on principle (‡) (preferring graphs with mental-to-physical edges, whether read metaphysically or as a representational convention), are compatible with it being a theory of causal relations in the world rather than of their representation.

(4) Determine whether any alternative formal modification of interventionism can both (A) avoid the original incompatibility between interventionism and non-reductive physicalism without invoking principles like (†) and (‡), and (B) render mental-to-physical causation empirically distinguishable from epiphenomenalist physical alternatives.

Issues (1)–(4) are facets, identified in the source, of the single question whether manipulability-based methods can substantively ground non-reductive downward causation; they are jointly posed, and the headline question is answered — affirmatively or negatively — by what (1)–(4) establish.

References

  1. Baumgartner, Michael (2009). 'Interventionist causal exclusion and non-reductive physicalism', International Studies in the Philosophy of Science 23(2):161–178. DOI 10.1080/02698590903006909. https://doi.org/10.1080/02698590903006909
  2. Baumgartner, Michael (2010). 'Interventionism and epiphenomenalism', Canadian Journal of Philosophy 40(3):359–383. DOI 10.1080/00455091.2010.10716727. https://doi.org/10.1080/00455091.2010.10716727
  3. Baumgartner, Michael (2013). 'Rendering Interventionism and Non-Reductive Physicalism Compatible', Dialectica 67(1):1–27. DOI 10.1111/1746-8361.12008. https://doi.org/10.1111/1746-8361.12008 (source of the open question)
  4. Kim, Jaegwon (2005). Physicalism, or Something Near Enough. Princeton University Press. ISBN 978-0-691-11385-7 (ebook ISBN 978-1-4008-4084-7, DOI 10.1515/9781400840847). https://doi.org/10.1515/9781400840847
  5. Woodward, James (2003). Making Things Happen: A Theory of Causal Explanation. Oxford University Press. ISBN 978-0-19-515527-3, DOI 10.1093/0195155270.001.0001. https://doi.org/10.1093/0195155270.001.0001
  6. List, Christian, and Peter Menzies (2009). 'Nonreductive Physicalism and the Limits of the Exclusion Principle', Journal of Philosophy 106(9):475–502. DOI 10.5840/jphil2009106936. https://doi.org/10.5840/jphil2009106936
  7. Eronen, Markus I., and Daniel S. Brooks (2014). 'Interventionism and Supervenience: A New Problem and Provisional Solution', International Studies in the Philosophy of Science 28(2):185–202. DOI 10.1080/02698595.2014.932529. https://doi.org/10.1080/02698595.2014.932529
  8. Woodward, James (2015). 'Interventionism and Causal Exclusion', Philosophy and Phenomenological Research 91(2):303–347. DOI 10.1111/phpr.12095. https://doi.org/10.1111/phpr.12095
  9. Baumgartner, Michael, and Alexander Gebharter (2016). 'Constitutive Relevance, Mutual Manipulability, and Fat-Handedness', The British Journal for the Philosophy of Science 67(3):731–756. DOI 10.1093/bjps/axv003. https://doi.org/10.1093/bjps/axv003
  10. Gebharter, Alexander (2017). 'Causal Exclusion and Causal Bayes Nets', Philosophy and Phenomenological Research 95(2):353–375. DOI 10.1111/phpr.12247. https://doi.org/10.1111/phpr.12247
  11. Baumgartner, Michael (2018). 'The Inherent Empirical Underdetermination of Mental Causation', Australasian Journal of Philosophy 96(2):335–350. DOI 10.1080/00048402.2017.1328451. https://doi.org/10.1080/00048402.2017.1328451
  12. Hoffmann-Kolss, Vera (2022). 'Interventionism and Non-Causal Dependence Relations: New Work for a Theory of Supervenience', Australasian Journal of Philosophy 100(4):679–694. DOI 10.1080/00048402.2021.1941154. https://doi.org/10.1080/00048402.2021.1941154
  13. Kinney, David (2023). 'Bayesian Networks and Causal Ecumenism', Erkenntnis 88(1):147–172. DOI 10.1007/s10670-020-00343-z. https://doi.org/10.1007/s10670-020-00343-z
  14. Stern, Reuben, and Benjamin Eva (2023). 'Anti-reductionist Interventionism', The British Journal for the Philosophy of Science 74(1):241–267. DOI 10.1086/714792. https://doi.org/10.1086/714792
  15. Blanchard, Thomas (2023). 'The causal efficacy of composites: a dilemma for interventionism', Philosophical Studies 180(9):2685–2706. DOI 10.1007/s11098-023-02001-w. https://doi.org/10.1007/s11098-023-02001-w
  16. Gebharter, Alexander, and Maria Sekatskaya (2024). 'Mental causation, interventionism, and probabilistic supervenience', Synthese 203(6):206. DOI 10.1007/s11229-024-04608-w. https://doi.org/10.1007/s11229-024-04608-w
  17. Blanchard, Thomas, and Andreas Hüttemann (2024). 'Causal modeling in multilevel settings: A new proposal', Philosophy and Phenomenological Research 109(2):433–457. DOI 10.1111/phpr.13043. https://doi.org/10.1111/phpr.13043
  18. McDonald, Jennifer (2025). 'Causal models and causal relativism', Synthese 205(3):108. DOI 10.1007/s11229-024-04893-5. https://doi.org/10.1007/s11229-024-04893-5
  19. Gebharter, Alexander, and Maria Sekatskaya (2025). 'Probabilistic supervenience and agential possibilities', Philosophical Explorations 28(3):245–264. DOI 10.1080/13869795.2025.2502373. https://doi.org/10.1080/13869795.2025.2502373
  20. Zhang, Youheng (2026). 'Stochastic supervenience: a conservative extension of classical deterministic supervenience', European Journal for Philosophy of Science 16(2):32. DOI 10.1007/s13194-026-00739-8. https://doi.org/10.1007/s13194-026-00739-8