Skip to main content

Intervening on the Causal Exclusion Problem for Integrated Information Theory


In this paper, we examine the causal framework within which integrated information theory (IIT) of consciousness makes it claims. We argue that, in its current formulation, IIT is threatened by the causal exclusion problem. Some proponents of IIT have attempted to thwart the causal exclusion problem by arguing that IIT has the resources to demonstrate genuine causal emergence at macro scales. In contrast, we argue that their proposed solution to the problem is damagingly circular as a result of inter-defining information and causation. As a solution, we propose that IIT should adopt the specific interventionist causal framework that we offer and show how IIT can harness this interventionist framework to avoid the causal exclusion problem. We demonstrate how our argument remains fully compatible with the methodology, empirical data, and conceptual aims of the theory.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7


  1. Tononi writes, “Integrated information is indicated with the symbol Φ (the vertical ‘I’ stands for information, the circle ‘O’ for integration)” (Tononi 2008, p. 220).

  2. ‘Cut’ refers to a partitioning of elements to determine which forms integrated information subsets.

  3. For the sake of completeness, IIT technically defines a concept as “[a] mechanism and the maximally-irreducible cause-effect repertoire it specifies, with its associated value of integrated information φmax. The concept expresses the cause-effect power of a mechanism within a complex” (Tononi and Koch, 2015, p. 6).

  4. Here one might wonder, how can information be objective, if such objectivity is not grounded in causation? A possible solution could be to adopt Floridi’s (2011) Information Structural Realism (ISR) in which information is objective in virtue of providing the interpretation of basic objects of reality—that is, structural objects. This is merely one possible avenue of investigation to see how IIT might rescue the objectivity of information. Suffice it to say, that if the arguments we present hold in this essay, IIT shouldn’t look for the objectivity of information in causation, but something else. What that something else is we leave open for further discussion and development, as it is outside the scope of this essay.

  5. Technically, a directed graph is an ordered pair <V, E> with V being a set of vertices that represent the causal relata of the graph and E a set of directed edges connecting the vertices. Less technically, it’s a graph with variables in it connected by arrows that stand for causal relations. A directed path is a route from one variable to at least one other via at least one directed edge, see Woodward (2003, p. 42 for more fine-grained details). All the graphs we work with in this paper are acyclic and we remain neutral on whether directed graphs have to be acyclic generally speaking.

  6. This is, in fact, a toy graph. Like a toy model, it is designed with a specific function in mind—here, to accessibly illustrate the conditions of IV. We have added some non-standard representative elements to do this work. The square-ended arrow is a preventer, illustrating that X is no longer causally dependent on U because of I, we feel this nicely captures Woodward’s (2003, p. 100) sentiment that, ““any variable U (distinct from I) that was previously a cause of X is no longer such a cause,” but with the proviso that it is not the only way the target variable, X, could be fully controlled by an intervention and thus not the only way to meet I2. Similarly, the solid line blocking the arrow exiting T, represents the idea that there could be paths for I to reach Y not via X (I3). However, in the present case that state-of-affairs does not occur, thus I3 is met.

  7. Woodward (2015b, pp. 332–335) explores (2) and Baumgartner (2013, pp. 16–24) provides a response. We will not be detailing that branch of the dialectic as we argue that option (1) has sufficient resources to avoid the exclusion problem.

  8. For detailed presentations of the expansion argument see Burge (2003) as well as Baker (2003). For a response to the expansion argument see Kim (1998, p. 77ff) and for rejoinders to Kim’s response see Bontly (2002), Ladyman et al. (2007), and Marras (2000).

  9. To wit, recently Gebharter (2015) has demonstrated that from within a ‘causal Bayes net theory of causation’ (CBN), mental and physical properties can not only be modelled on the same graph, but in such situations, mental properties always fail to meet a productivity requirement, giving rise to an analogous causal exclusion problem. Gebaharter argues that because CBN can give a more detailed account of why the exclusion problem is valid, we ought to accept the exclusion problem. This argument paves the way for a more in-depth discussion of the conceptual and empirical reasons for, and against, choosing to model supervenience relations as causal relations. Whilst Gebharter’s argument shows that CBNs can coherently choose to do so, our present purpose has been to defend the claim that interventionism can coherently chose not to.

  10. To be clear here, we are not arguing that it is conceptually incoherent for interventionist causal claims to be grounded in causal powers. Rather, we think that avoiding metaphysically loaded semantics—like cause-effect power—will help proponents of IIT to make their position clear, without any loss of content. This is in keeping with our condition (3). We interpret the strategy utilised by proponents of IIT in avoiding the causal exclusion argument to be metaphysically neutral, rather than relying on metaphysical arguments that stem directly from adopting a causal powers view (e.g., Gibb 2013, 2015; Mumford and Anjum 2011). Given that interventionism is a framework that also strives to remain metaphysically neutral (Woodward 2015b) we are merely suggesting that IIT’s causal semantics are adjusted to reflect this alignment of interests. Thanks to an anonymous reviewer for raising this objection.


Download references


We would like to thank Larissa Albantakis, Philip Goff, Anna Kocsis, Michele Luchetti, and Carlos Montemayor for helpful comments and advice on previous drafts. Thank you to the audience of the 5th WFAP conference at the University of Vienna, at which we presented an earlier version of this paper. Thanks to three anonymous reviewers for constructive comments and suggestions. The author ordering is solely alphabetical, all work towards the development of this paper was equally shared.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Matthew Baxendale.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Baxendale, M., Mindt, G. Intervening on the Causal Exclusion Problem for Integrated Information Theory. Minds & Machines 28, 331–351 (2018).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


  • Integrated information theory
  • Interventionism
  • Active counterfactuals
  • Consciousness
  • Causal exclusion
  • Downward exclusion