A Note on Constitutive Relevance in Mechanisms

Carl Craver (Explaining the Brain, 2007) argues that what it is for an object doing X (micro variable) to be a working component of a mechanism doing Y (macro variable) is (i) for the former to be a part of the latter and (ii) for the two of them to be mutually manipulable. The second clause means that if you ideally intervene on the micro variable you can generate changes in the macro variable and vice versa. This is the mutual manipulability (MM) account of constitutive relevance.

In a 2017 paper, Baumgartner and Casini argue that the kind of mutual manipulation posited by MM is impossible, because one of the two variables is part of the other, so they can’t be manipulated independently as the notion of ideal intervention requires.

But given that scientists perform mutual manipulations all the time, there is something wrong with Baumgartner and Casini’s critique.

Here is how I see it. Micro variables are part of macro variables, which means that micro variables also enter causal relations with other micro variables that are also parts of the same macro variable. Thus, in bottom-up interventions, intervening on a micro variable alters the output of the macro variable via affecting other micro variables that, together, constitute the macro variable and produce its output. Mutatis mutandis for top-down interventions: you can intervene on the input to a macro variable, which in turn affects its micro constituents including the one of interest, where you then detect a change. Thus there is a mixture of constitutive and causal relations that must be captured by an adequate MM account. The way I am articulating MM departs from the way Craver 2007 put it. Craver insists that the relation between micro and macro is solely a matter of constitution not causation, whereas I am claiming that a mixture of causation (between micro constituents) and constitution is involved. Still, I think I’m just articulating what Craver was after and what I’m saying is consistent with his diagrams, in which multiple micro variables that constitute a macro variable are causally connected to one another. So there is room for slight refinement in formulating an adequate MM account but the substance is the same. The important point is that Baumgartner and Casini’s critique does not go through.

Baumgartner and Casini go on to offer an alternative to MM, which they call No De-Coupling account (NDC). NDC claims that micro and macro variables, when one partially constitutes the other, can only be manipulated via interventions that cause changes in both variables via separate causal paths. Problem: since the micro variable partially constitutes the macro variable, there can’t be separate causal paths from an intervention to the micro and the macro variables. Any causal path from the intervention to the micro variable will also affect the macro variable and vice versa. They are not (wholly) distinct variables, so they cannot enter separate causal paths. So NDC doesn’t work. I’m sort of turning Baumgartner and Casini’s critique of MM against NDC; the critique does not work against MM (when properly formulated), but it does work against NDC.

Or am I missing something?

67 Comments

  1. At SMS last week, I argued that, among other things, (a) Craver is providing an account of constitutive explanatory relevance rather than constitution itself, so the alleged problem has less force, (b) Baumgartner’s and Casini’s (and Krickel’s) solution has some of the same defects they attribute to the (MM) view when it is defended by others, and (c) one might nevertheless use their insights about simultaneity to repair problems with interlevel “interventions” so that the manipulationist account gives the answers that it is supposed to give.

  2. Ken

    I think you are missing something. So, the B&C point is that the kind of thing you describe as MM does not satisfy the conditions of an ideal intervention. Here is abstract: “The first part of this article finds Craver’s mutual manipulability theory (MM) of constitution inadequate, as it definitionally ties constitution to the feasibility of ideal experiments, which, however, are unrealizable in principle.” That may not be the best way to put the point, but that is the point. One might put the problem differently. Compositional explanations never satisfy the conditions for ideal interventions, so the conditions are vacuous.

    Or, her is another way to put the point. What you describe as MM does not satisfy the conditions of an ideal intervention.

    So, I don’t think you get to the heart of the B&C point until you speak to ideal interventions.

    Contrary to Tom’s view, I don’t see how treating Craver’s account of MM as an “account of constitutive explanatory relevance rather than constitution itself” will help much. In either case, it looks like the account will be vacuous.

  3. Gualtiero Piccinini

    Thanks, Ken! You are absolutely right that B&C’s objection rests on the notion of ideal intervention. Unfortunately they rely on Woodward’s notion of ideal intervention, which is obviously inapplicable to the constitutive relevance case. That is why Craver (2007, p. 154) introduces a distinct notion of ideal intervention for the constitutive relevance case, which B&C do not even mention.

    As I mentioned in my post, I still think that Craver’s formulation of MM (and thus of ideal interventions with respect to constitutive relevance) can be refined by taking into account that the micro variable causally interacts with other micro variables that, collectively, constitute the macro variable. Still, as far as I can tell, B&C’s objection does not go through.

    What do you think?

  4. Ken

    Well, I think that B&C do take on Craver’s version of Woodward when they write,
    “Correspondingly, he proposes a theory of constitution that adds a parthood and a mutuality tweak to Woodward’s interventionist theory of causation. Subject to Craver’s (2007, 153) mutual manipulability theory (MM), the behavior of a spatiotemporal part X of S constitutes S’s – ing iff it is possible to (ideally) intervene (from the bottom up) on X’s -ing such that S’s -ing changes, and (from the top down) on S’s -ing such that X’s -ing changes.”

    See p. 2 of B &C “An abductive theory of constitution”. What’s wrong with this?

    I’ve seen Casini and Emily Prychitcko go through the objection, taking on the point that the explanation of processes involves both constitutive and causal connections, so I think the other side gets this and at least makes the case that it does not help. I am about to start working through these papers in more detail, but I don’t see an obvious reply to this and I don’t see that you have brought up a consideration that they have not at least thought about.

  5. Gualtiero Piccinini

    Thanks. There is nothing too wrong with the passage you cite. It summarizes MM quickly. It’s slightly misleading in that in order to do justice to MM, they should also mention that MM relies on a specialized notion of ideal intervention (different from Woodward’s) designed to suit the constitutive relevance case, but they do not mention that. And I’m not aware of any passage in B&C that discusses this. They simply treat MM as if it were adopting Woodward’s notion of ideal intervention:

    “The core notion in its definiens is the modal notion of manipulability, which Craver (2007, sec. 4.8.3) cashes out in terms of the existence of a possible ideal intervention as defined by Woodward (2003, 98).”

    But in that very section (2007, 4.8.3), Craver explicitly departs from Woodward’s notion of ideal intervention to define a novel notion of ideal intervention for the purpose of formulating MM. So at least in their published paper, B&C simply misrepresent MM and raise an irrelevant objection based on Woodward’s notion of ideal intervention.

    If they make the case that Craver’s alternative notion of ideal intervention does not help, I’d like to know where they make it.

  6. Ken

    Well, I’m not going to get into the finer points of textual exegesis here, since I have not read this literature carefully. But here is what you need to show, rather than merely assert, “So at least in their published paper, B&C simply misrepresent MM and raise an irrelevant objection based on Woodward’s notion of ideal intervention.” In your original post, you don’t even mention ideal interventions.

  7. Ken

    Here is another way of making the last point. The Baumgartner, Gebharter, Casini, et al., project is to come up with a way to use some version of an ideal intervention to articulate the informal notion of MM for some (debatable) end. It is a bold conjecture to say that they have all missed a viable version that is in plain sight in Craver, 2007. That could be right, but that is what is up for grabs. So, you could just post up that version and show how it works.

  8. Gualtiero Piccinini

    Ken, I’m not sure what else you want me to do. I quoted their paper where they incorrectly state that MM is based on Woodward’s notion ideal intervention. If you look at their objection to MM, it is explicitly framed in terms of Woodward’s notion, which is irrelevant because it’s not the one Craver uses. I gave you the page (p. 154) where Craver defines an alternative notion of ideal intervention that is designed to fit MM. In my original post, I even granted that Craver’s formulation of MM could be refined to take into account some additional subtleties having to do with the interplay of causation and constitution and I sketched how I would go about that. If you think I am still missing something, please let me know what that is. Thanks.

    Also, FWIW, my original post does refer to ideal interventions.

    • Ken

      “Also, FWIW, my original post does refer to ideal interventions.”
      Right. I should have been more careful. My point is that in your “Here is how I see it.” paragraph you do not get the ideal part into the story.

      But, here is what more I think you can do, in outline. Present what you take to be Craver’s version of an ideal intervention, then show how it avoids the B&C problem. This is what I take Beate, below, to be driving at when she writes, “Second, I am not sure whether Carl’s notion of an ideal intervention differs from Woodward’s in a way such that the problem is avoided “

      • Gualtiero Piccinini

        Thanks. As I said, I’m not defending Craver’s exact formulation. But since you are asking: B&C’s objection to Craver’s MM hinges on the fact that ideal interventions in Woodward’s sense require that if you are trying to show that X causes Y, you must intervene on X without independently intervening on Y. As they point out, if Y partially constitutes X, you can’t do that. Any intervention on X will also be an intervention on Y. Of course! Craver never claimed or presupposed otherwise. In Craver’s specialized notion of ideal intervention (p. 154), there is no requirement that you intervene on X without independently intervening on Y (and vice versa), for the obvious reason that you are not trying to show that X causes Y; you are trying to show that Y is a constituent of X.

        To repeat: I do think that Craver’s account can be refined, and I tried to sketch how to go about it, and it sound like both Tom and Beate are pushing in roughly the same direction (as is Prychitcko, based on a talk by her that I witnessed). My point is simply that B&C’s objection is unfair to say the least.

        IMHO, all these people are really talking about how to discover whether something is a constituent of a mechanism, not what constitutive relevance is. But that is beside the point.

        • I think you are right when speaking about causation as a relation between objects/events in the world. But interventionism takes causation to be a relation between variables. Hence, “causal path” is defined as a path between variables (or at least, this is how I interpret the notion). Therefore, the individuation criteria of causal paths are not metaphysical but in terms of variables and causal models. Moreover, Baumgartner & Gebharter & Casini, as well as Romero have an independent argument for the “common cause” claim. For example, Romero argues that according to Reichenbach’s criterion, a structure where an intervention changes two variables X and Y, can either be of the form I -> X -> Y or I -> Y -> X or X Y. Since constitutive relevance is supposed to be non-causal, the first two interpretation cannot be true in the case of interventions into phenomena/mechanisms. Hence, the intervention must be a common cause of X and Y. Baumgartner and Gebharter get the same result simply by applying interventionism (the argument is a bit more complex).

  9. Gualtiero Piccinini

    PS: I do appreciate that you guys (Tom and Ken) are taking the time to respond to my post. I’m in the process of refereeing a paper that is based on B&C’s paper and I want to be as fair as possible, so this discussion is very helpful.

  10. Really interesting discussion! Let me add some points:

    First, a side remark : different formulations of the same problem have been developed by Lena Kästner (unpublished) and Felipe Romero (2016) (both were Carl’s PhD-students, not that this matters…).

    Second, I am not sure whether Carl’s notion of an ideal intervention differs from Woodward’s in a way such that the problem is avoided – he still requires that the intervention does not change both variables directly (I1). He “solves” the problem by stating (in brackets) that the component-variable and the phenomenon-variable are related by supervenience. But I do not see how he can do this. Sure, the component is not the complete supervenience base of the phenomenon. But it is surely part of the supervenience base (or if you do not want to call it “supervenience” call it “constitution”) such that a change in the component-variable non-causally leads to a change in the phenomenon-variable.

    Third, Baumgartner & Gebharter (2016) as well as Baumgartner & Casini (2017) solve the problem (interventions into phenomena cannot be ideal) by allowing for a modified notion of an ideal intervention along the lines of Woodward (2014). According to Woodward (2014), ideal interventions can change the putative effect variable via two different causal paths only if the two paths are related by supervenience (or any other non-causal dependence relation). In this sense, we can have ideal interventions into phenomena – ideal interventions that directly influence the phenomenon and its supervenience base. Also interventions into the input of a phenomenon, as you suggest Gualtierro, will be ideal in this modified sense (call it ideal*).

    The problem now is that if you use this modified notion of an ideal intervention alone to spell out constitutive relevance, you get an empty account: the notion of an ideal* intervention presupposes that you know that the putative constituent is a constituent (otherwise the intervention would not be ideal*). Baumgartner, Gebharter, Casini (& Romero) suggest different solutions for this problem. All of them (so far) seem to be abductive: every intervention into the phenomenon has to be ideal*… One way to spell this idea out in more detail is the coupling idea.

    I am not convinced that the coupling idea works or is without problems. But I am convinced that we have a real problem here. I suggested a possible solution (forthcoming; a draft can be found here: https://www.academia.edu/33205180/Saving_the_Mutual_Manipulability_Account_of_Constitutive_Relevance).

    The suggestion is based on the metaphysical framework of interlevel causation I defend here: https://link.springer.com/article/10.1007/s10838-017-9373-0?wt_mc=Internal.Event.1.SEM.ArticleAuthorAssignedToIssue.

    One implication of my suggestion is that changes in the phenomenon and changes in the components occur at different times (such as changes in the engine and changes in the car’s moving). This might not capture all cases of constitution (e.g., changes in the wheel’s turning and changes in the car’s moving).

    Another suggestion is to introduce a new notion of an intervention in terms of horizontal surgicality (that’s what Lorenzo presented at SMS). This suggestion implies that changes have to occur at different times. (it suffers from the opposite problems that my suggestion suffers from).

    I compare both suggestion here (these are just slides, sorry): https://www.academia.edu/34098962/Mechanistic_Constitution_-_How_many_relations

    • Gualtiero Piccinini

      Thanks, Beate! It sounds like your solution is along the lines of what I was suggesting, which is very reassuring. The one thing I don’t understand is this:

      “According to Woodward (2014), ideal interventions can change the putative effect variable via two different causal paths only if the two paths are related by supervenience (or any other non-causal dependence relation).”

      Could you please give me the full reference to Woodward 2014 so I can look into it? (If you know the page number that would be even better.)

      • Sure! The title of the paper is “Interventionism and Causal Exclusion” and it is from 2015 not 14, sorry. The page is 333 where he says “Consistently with this, when an intervention occurs on X, its supervenience base SB(X) should not be regarded as one of those “off route causes” in IV that one needs to control for or hold fixed in intervening on X”

        • Gualtiero Piccinini

          Thanks. The way Woodward puts the point makes sense. Woodward says or implies that if X supervenes on Y, any causal path from I to X is the same as the causal path from I to Y. That makes sense, and it contradicts what B&M say (“separate causal paths”).

          • Not sure whether Woodward argues that there is only one causal path or two. One could say that there are two but one supervenes on the other. It might depend on what the individuation criteria for causal paths are. Since, according to Woodward, all there is to causation is manipulability, I would think that two causal paths different iff they involve different variables. Hence, the two causal paths would be different. Two variables are affected via the same causal path iff one is the cause of the other, or if they are indeed identical variables. Since variables are individuated intensionally, the latter does not make much sense. Hence, I would think that Baumgartner & Co. are right.

          • Uups: this should be here:

            One additional remark: I think for Baumgartner & Co’s argument it is not too important whether the causal path connecting the intervention and the phenomenon-variable is identical with the causal path connecting the intervention with the component-variable. All they need is that the intervention is necessarily fat-handed, i.e., necessarily changes two variables at the same time without these variables being causally connected.

  11. Regarding why it matters that Craver is talking about constitutive explanatory relevance rather than constitution itself: because if MM [or a corrected version] is a condition on constitution and it cannot be satisfied, then the account of constitution is wrong. (Or there is no such thing as mechanistic constitution.) But if something like mutual manipulability–leaving it open whether thesis MM is the right way to state it–is a criteria for explanation, then if there is a problem with applying the criteria the result is just that you can’t draw inferences about explanation in those cases. Because the criteria is best treated as sufficient rathe than necessary, there could be other criteria. Or we could accept that the inferences are less justified insofar as the manipulation departs from ideal conditions, etc. This is a much less damaging problem.

    • I agree that MM does not give us a metaphysical notion of constitution. Analogously to interventionism as a theory of causation, MM provides an operational definition of what constitutive relevance is, and how we detect it. Not sure whether this is in line with the metaphysics/explanation distinction. But this might be a more general problem afflicting interventionism. I would be happy with simply assuming that interventionism and MM deal only with causal/constitutive explanation. What the metaphysics of causation and constitution is, is a different question.

      • One additional remark: I think for Baumgartner & Co’s argument it is not too important whether the causal path connecting the intervention and the phenomenon-variable is identical with the causal path connecting the intervention with the component-variable. All they need is that the intervention is necessarily fat-handed, i.e., necessarily changes two variables at the same time without these variables being causally connected.

    • Ken

      Ok, Tom, now I think I see what is going on. I think you are conflating two distinctions.

      So, the Craver account could be an account constitutive explanatory relevance or an account of constitution itself. But, then, it could be an account that offers sufficiency conditions versus, let us say, necessary and sufficient conditions. So, there are four options.

      1. MM is an account of what constitutive explanatory relevance is.
      2. MM is an account of constitution itself.
      3. MM provides sufficiency conditions for constitutive explanatory relevance.
      4. MM provides sufficiency conditions for constitution.

      The B&C objection to 1. and 2. amounts to both of them being “wrong”. The B&C objection to 3. and 4. amounts to them both being vacuous sufficiency conditions. So, it seems to me that the distinction between 1 and 2 and between 3 and 4 doesn’t help much.

      • I read B&C as targeting 2 & 4. They also thinking that undermines 1 & 3 because they think providing an account of constitutive explanatory relevance requires providing an account of constitution. But I do not think that follows. One does not always have to ground out at the foundations.

        • Ken

          Grant the B&C are targeting 2. & 4. My point is that you can run essentially the same argument against 1. & 3. You don’t have to argue that undermining 2. & 4. *thereby* undermines 1. & 3. It’s that once you see how the objection goes to 2. & 4., you can carry it over directly to 1. & 3.

          So, for example, if the problem for 4. is that the conditions are vacuous. Then the problem for 3 is also going to be that the conditions are vacuous.

  12. Gualtiero Piccinini

    Beate, thanks for your comments. I’m not sure what you think B&C are right about. If I push a cart full of groceries, the groceries get pushed too. How could there be two different causal paths between my intervention and the whole cart-including-groceries (macro variable) vs. my intervention and the groceries (micro variables)? It’s one and the same push; it reaches the groceries because they are part of the cart-including-groceries. By the same token, suppose that instead of pushing the whole cart I reach with my hand inside the cart and push against a jug of milk; the whole cart will move as well. How could there be two different causal paths? The push is one and the same; the whole cart move because the jug of milk is part of it and connected to the rest of it in the right way. (Of course, the jug pushes the rest of the cart, but that does not create a separate causal path either.) So, yes, you can easily define two different labels: the “push-to-milk-jug” causal path and the “push-to-cart” causal path. This is analogous to giving two names to the same person. Labeling something twice does not change the fact that it’s one and the same thing.

    • I think you are right when speaking about causation as a relation between objects/events in the world. But interventionism takes causation to be a relation between variables. Hence, “causal path” is defined as a path between variables (or at least, this is how I interpret the notion). Therefore, the individuation criteria of causal paths are not metaphysical but in terms of variables and causal models. Moreover, Baumgartner & Gebharter & Casini, as well as Romero have an independent argument for the “common cause” claim. For example, Romero argues that according to Reichenbach’s criterion, a structure where an intervention changes two variables X and Y, can either be of the form I -> X -> Y or I -> Y -> X or I is a common cause of X and Y. Since constitutive relevance is supposed to be non-causal, the first two interpretation cannot be true in the case of interventions into phenomena/mechanisms. Hence, the intervention must be a common cause of X and Y. Baumgartner and Gebharter get the same result simply by applying interventionism (the argument is a bit more complex).

  13. The worry that I have about this whole debate, including my own contributions to it, is that its unclear that it is advancing our understanding of the phenomena in the world, either metaphysical or explanatory. If B, et al. are right, do we really understand either “constitution” (here used as a generic term for composition, constitution, realization, etc.) or constitutive (generic) explanation better than we did before? Are we left thinking that Craver and Woodward are completely misguided, as is sometimes implied; or just that they haven’t got all the details right yet? Getting the details right is important, I suppose. But for most philosophers who are trying to make use of Woodward or Craver, it’s the general approach that matters and the details usually make no difference. Per Aristotle, “It is the mark of an educated [person] to look for precision in each class of things just so far as the nature of the subject admits.”

    • I think this debate produced a lot of interesting ideas and thoughts regarding our understanding of constitution and constitutive explanation. For example, Baumgartner, Gebharter, Casini, and Romero have interesting ideas with regard to the question what kind of inferences underly constitution claims. Is is true that scientists can only abductively infer to constitution (where this is a consequence due to the very nature of constitution and experiments/interventions, rather than due to the fact that we live in a non-ideal world)?

      Lena Kästner (her book is published now: https://www.degruyter.com/view/product/482159) has interesting ideas on non-interventionist studies and how scientists make use of them to infer to constitution in order to solve the problems of MM.

      I myself think the problems of the MM account show that we have to think more about the metaphysics behind all of that. Phenomena are processes/process-like, there can be causation (light) between levels, etc.

      These points were not just indirectly motivated by the problems of the MM account but where suggested as solutions to them (in terms of modifications of the MM account or in terms of alternative accounts). Hence, I think Carl’s idea was brilliant and motivated many new ideas.

  14. Ken

    Gillett and I have been pointing out certain limitations of the MM approach in various talks for the last few years. Tom and Beate, you have heard both Carl and me giving talks on these issues.

      • Ken

        Well, Gillett and I are interested in more than just teasing these apart. Gillett has a multi-point characterization of compositional relations (implementation, realization, and constitution). Chapter 2 of his book on Reduction and Emergence sets this out.

        • Ken

          And, you know, much of Gillett’s work has been directed toward showing why one or another postulated relation, e.g. subset realization, Ground, identity, is not the on in play in compositional explanations in the sciences. I’m hoping that when folks get done wrestling with MM approaches, they will look at the Gillett alternative to the metaphysics of composition.

          • For compositional relations, I favor composition, viz., the relation between parts and wholes. But I also think there are probably many composition relations for objects, and then that brings us to Aizawa and Gillett on the “building” (cf. Karen Bennett) relations among events, properties, or processes.

    • I took your talks to address more general issues regarding the mechanistic approach (e.g., the role of properties in mechanistic explanation/ontology) rather than particular problems of the combination of MM and interventionism. But I might not remember that correctly. Sorry. I mentioned only Baumgartner & Co., Lena, and my own work due to a recency effect, I guess. Surely, this list is not complete.

      • Ken

        Hi, Beate,
        Gillett has been making the case against MM as a species of “neo-causalism”. It ties to take equipment designed to deal with causation and tweak it to apply to a wholly distinct relation that he calls “implementation”. He tries to show how this does not work. MM fails as a descriptive account of what compositional relations are. That invites the reply that MM is a method of testing for compositional relations.

        So, in my talk, inspired by discussions with Gillett, I take up that idea. I mention three foci in New Mechanism 1) the focus on processes as the explanandum, 2) the focus on entities and activities as explanantia, and 3) interlevel interventions as “crucial” for testing for compositional relations. It’s my third point wherein I suggest an alternative to interlevel interventions, namely, the intercalation of independent results. Compositional relations are tested by a combination of counting and measuring.

          • Ken

            Hi, Beate,
            One version Gillett has is section 1.2 of his chapter in our edited collection:
            Gillett, C. (2016) The Metaphysics of Nature, Science, and the Rules of Engagement. In Scientific Composition and Metaphysical Ground.

            In essence the book attempts to be about different approaches to the ontology of interlevel mechanistic explanation.

            I’ll send you a text for my paper.

  15. Gualtiero Piccinini

    Beate wrote: “Yes, that might be true. But what do you think of, for example, Baumgartner & Gebharter’s argument in “Constitutive Relevance, Mutual Manipulability, and Fat-Handedness” on pages 741/742?”

    It has more bells and whistles but it’s the same fallacious argument, implicitly based on Reichenbach’s criterion, which doesn’t apply because we are dealing with variables that are not (wholly) distinct.

  16. Ken

    One of the things that Carl has been working on lately, I think, is what *I* would describe as the following. Talk of “abstraction” (the omission of details) and “explanatory relevance” (what figures in an explanation) are blanket terms that cover two distinct sorts of processes. The first is ontological, the second pragmatic. The first has to do with whether or not things stand in the appropriate compositional relations; the second has to do with whether or not scientists are concerned to mention all of the things that stand in the appropriate compositional relations. Carl was talking about this at a little pre-SMS workshop we had in Newark.

  17. Dear all,

    stumbling across this debate has been quite interesting for me. Thanks to all of you for contributing!

    I feel the urge to enter the debate but am unsure as to what would be the most suitable entry point. Our argument in B&C and B&G turns on a number of assumptions. Gualtiero rejects the argument’s conclusion. In consequence, he either has to reject the deductive validity of our argument or reject one of its assumptions. It does not become entirely clear to me what his preferred move is, but I suspect he wants to reject one (or more) of the assumptions. So, here’s a list of the relevant assumptions:

    1. Constitution is a non-causal dependence relation (it expresses a synchronic dependence, whereas causation is diachronic).

    2. Macro properties/phenomena non-reductively supervene on their constituents (relative to the latter’s organization), meaning that macro properties are non-identical to their constituents, and correspondingly that variables representing macro properties are different from variables representing their constituents.

    3. Causation is to be understood in terms of Woodward’s interventionist theory of causation.

    4. An intervention on X w.r.t. Y is a cause of X that is not connected to Y on a causal path that does not go through X (that is a common requirement both in Woodward’s and Craver’s notion of an intervention, cf. condition I1_c of Craver (2007, p. 154); hence, the difference between Craver’s and Woodward’s notions of an intervention is irrelevant for our purposes.)

    5. A directed causal path is an ordered n-tuple of variables, to the effect that two paths are identical iff they feature the same variables in the same ordering.

    Now, from (2) and (3), it follows that all candidate interventions I on a macro variable \Psi are causes of both \Psi and at least one of its constituents \Phi.
    There are three logically possible causal structures accommodating the fact the I causes both \Phi and \Psi, viz. I -> \Psi -> \Phi, I -> \Phi -> Psi, \Psi \Phi.
    (1) excludes the first two of these possibilities. It follows that the structure over these three variables is this one:
    \Psi \Phi.
    From (2) and (5), it follows that the paths and are non-identical, i.e. different, because \Psi is a different variable from \Phi.
    This, in turn, entails that I is a common cause of \Psi and \Phi.
    From (4), it follows that I is not an intervention on \Psi w.r.t. \Phi. (this holds also according to Craver’s notion of an intervention, due to a violation of condition I1_c)
    Nothing in this argument hinges on I, \Phi and \Psi being our target variables, hence, the argument generalizes: there cannot exist intervention variables on macro variables w.r.t. to their constituents.
    But according to MM, the possible existence of such intervention variables is necessary for constitution. Therefore, there does not a exist a single case of constitution.

    Gualtiero interprets this conclusion as a reductio of the corresponding argument. Fair enough. I would then invite him to pinpoint the assumption(s) he wants to reject (or if he’s rejecting the argument’s deductive validity, please, identify the inferential step that is invalid). My suspicion is that he is targeting either (2) or (5), or both, or maybe also (1)… He says that macro variables are “not (wholly) different” from their constituents. What does that mean? Is \Phi identical to \Psi? Are macro properties reducible to their constituents? Yes or no? If yes, Gualtiero rejects (2) and the problem clearly goes away. We can close the debate. But then constitution reduces to identity (for which I do not think we need a separate theory).

    If he does not reject (2), what about (5)? He insists that the causal paths and are not different. But according to the standard notion of a causal path used in the causal modeling literature (cf. Spirtes, Glymour, Scheines, Pearl, Woodward etc.), these paths clearly are different because, assuming we do not reject (2), \Phi is non-identical to \Psi. If (5) is rejected, I would ask for an alternative definition of the notion of a causal path that, on the one hand, does not have the undesired consequences and, on the other, does the theoretical work that notion is required to do in the relevant literature.
    In her comments, Beate correctly points out that we simply use the standard notion of a causal path in our arguments, but Gualtiero does not really respond. Instead, he brings up an example that, in light of assumption (1), is rather besides the point: “If I push a cart full of groceries, the groceries get pushed too.” Of course, there is only one path here, viz. a causal chain from the push to the movement of the cart on to the movement of the groceries. The groceries and the cart are not excluded to be causally related, viz. (1) does not hold for them. In case of macro properties and their constituents, however, we exclude a causal relation based on assumption (1). So, is it maybe assumption (1) that Gualtiero is rejecting? Is constitution a form for causation?

    It is fair to reject the conclusion of our argument against MM. But doing so requires explicitly identifying the argument’s assumption(s) that are false. Once we have that clearly on the table, we can think about the consequences such rejections have for constitutional and causal modeling, and for the New Mechanistic project more generally.

    Thanks again for that interesting debate,

    Michael

    • Gualtiero Piccinini

      Hi Michael, thanks so much for chiming in! I reject this: “There are three logically possible causal structures accommodating the fact the I causes both \Phi and \Psi”. I don’t know how you get there but none of your options accurately describes the situation. The correct representation is not a causal path in your sense but a mixed causal-constitution path.

      • Gualtiero Piccinini

        PS: FWIW, I offered my cart and groceries example is to illustrate my point in an easily understandable way. I’m sorry if I wasn’t as clear as I hoped to be. There is a causal arrow from me to the cart-including-groceries, and a constitution link from the cart-including-groceries to the groceries. On the flip side, there is a causal arrow from me to the milk carton, and a constitution link from the carton to the cart-including-groceries. Either way, the correct representation of the situation is not a causal path but a mixed causal-constitution path, as it were. Of course we can also further analyze the mechanism by looking at the causal connection between the milk carton and the rest of the cart, etc. But that goes beyond Craver’s formulation of MM towards the more refined formulation that I sketched in my post.

        • I am afraid that is all too quick for me. On the face of it, you seem to question the validity of our argument. You are saying that the three ways for the variable I to be a cause of both \Phi and \Psi we list in the paper are, in fact, not all logically possible ways for I to be a cause of both \Phi and \Psi. In addition to our three structures, you say that the variable I can also be connected to both \Phi and \Psi via a “mixed causal-constitution path”. The immediate follow-up question, of course, is: What is that?? What is the definition of the notion of a “mixed causal-constitution path”? What are identity criteria for “mixed causal-constitution paths”? Does a “mixed causal-constitution path” represent a third dependence relation? So, do we not only have causation and constitution, but also mixed-constitution-causation, call it MCC for short? What is that? To me this seems to be an entirely novel suggestion. What is the difference between causation and MCC, on the one hand, and constitution and MCC, on the other? [Or are you saying that the variable I is BOTH a cause and a constituent of \Phi and \Psi? (I put this option in brackets because I think that is not what you are saying, right?)] I hope you will understand that before I can take your solution to the problem seriously, I will want to see answers to all these questions. Simply injecting the label of a “mixed causal-constitution path” is not going to do it.

          What is more, I doubt that introducing the notion of MCC is going to solve the problem without additionally rejecting one of our assumptions. For the fact remains that Woodward’s theory of causation ENTAILS that I is cause of both \Phi and \Psi — not an MCC, but a CAUSE, plain and simple. And a variable I can only be a cause (simpliciter) of two variables \Phi and \Psi on one CAUSAL path or on multiple CAUSAL paths. There simply is no further logical possibility. Clearly, there may be other relations holding among (I, \Phi, \Psi), and MCC might be such a further relation, and there may be further paths connecting I, \Phi, \Psi, and a “mixed causal-constitution path” might be such a path. But that does not change the fact that Woodward’s theory implies that CAUSATION (simpliciter) holds as well. And for the question whether the variable I is an intervention variable, causation is all that matters. So, if you want to claim that the variable I is NOT a cause SIMPLICITER but and MCC of \Phi and \Psi, you have to reject Woodward’s theory of causation (and provide some alternative theory of MCC), meaning you have to reject our assumption (3).

          Let me end with a more general observation. In recent years, interventionism has been implemented in all sorts of contexts to “solve” all sorts of problems: just to name a few, it has been used to solve the problem of causal exclusion, the problem of constitution, or the problem of extended cognition. My impression is that in all of these contexts authors make reference to Woodward’s explicit definitions but do not adhere to their wording, meaning they do not work with the technical definition of an intervention variable but with some pre-theoretic understanding of intervening. (And if I may say so, I have the impression that this is also the approach you intend to take.) I think this is a very unfortunate state of affairs. Woodward has spelled out in detail the requirements we must impose on the notion of an intervention in order for it to do the theoretical and methodological work we want it to do (and as you know, Woodward has not been the only one doing so — Craver being somebody else, or Pearl, or Sprites, or Eberhardt…, and with respect to the feature of interventions that is crucial for our argument, viz. assumption (4), all of these technical notions of an intervention agree). Accordingly, if that notion is employed in any context to solve any kind of problem, there are very good reasons to adhere to the definitional and technical details. That does not mean, of course, that these details cannot (and should not) be changed, nor does it mean that we might not introduce further types of paths or further types of dependence relations apart from causation and constitution. But if the definitions are changed or new paths or dependence relations are introduced, the new definitional details should be made explicit.

          • Gualtiero Piccinini

            Michael, thanks for your queries. I’m not saying anything especially new; just the sort of considerations that Jaegwon Kim has been making for 40 years. Woodward sees it too (“Interventionism and Causal Exclusion,” PPR 2015).

            Maybe I don’t understand what you mean by “causal path”. The way you represent causal paths, a causal path seems to be a path all of whose links (between variables) are causal. If so, then patently there are other kinds of path besides the three kinds of causal path you list through which an intervention can be a cause of a variable. Woodward mentions causal-definitional paths, and Kim has made a career of discussing causal-supervenience paths and their implications. I called it a causal-constitutional path only because I was using the language that you and Craver use. It’s a path some of whose links are causal and some of whose links are constitutional.

            If you want to read more about how I think about these and related matters, here is my latest effort:
            https://www.academia.edu/34634632/Levels_of_Being

  18. Sorry guys. This interface does not properly display angle brackets apparently. Whenever it elliptically says “the paths and ” above, what it should say is “the paths (I, \Phi) and (I, \Psi)”. These paths are different because \Phi is not identical to \Psi.

  19. hmm, also arrows do not seem to be displayed properly. It should say:

    There are three logically possible causal structures accommodating the fact the I causes both \Phi and \Psi, viz.
    I \rightarrow \Psi \rightarrow \Phi,
    I \rightarrow \Phi \rightarrow Psi
    \Psi \leftarrow I \rightarrow \Phi.

  20. M. Couch

    Hi, I’m just stumbling into this discussion. For my own criticisms of Craver’s mutual manipulability account see my paper “Mechanisms and Constitutive Relevance” (2011) where I argue that the MM account is only epistemic. This means that it is not an account of constitutive relevance relations, but an account of how to find constitutive relevance relations. This may relate to Tom’s point that the MM account is about “explanation”, though I haven’t read his paper. FWIW Carl mentioned to me in person that the paper makes a good point.

    • Gualtiero Piccinini

      Thanks, Mark! I agree 100%. The whole debate discussed here would be better characterized as a debate about how to discover constitutive relevance relations.

      • Ken

        “The whole debate discussed here would be better characterized as a debate about how to discover constitutive relevance relations.” Except that some version MM seems to be the only option discussed. Are there other options that have been examined? Maybe Kastner has some alternative. Gillett and I have been working on an alternative.

        • Jens Harbecke (similar to M. Couch) account for constitution in terms of regularity (which can be interpreted as a metaphysical notion, I guess). I myself have developed an account of constitution in terms of part-whole relationships and causation (counterfactual dependence). Surely, Gillett’s dimensioned realization might be the most metaphysically robust interpretation of constitution.

          • Ken

            Hi, Beate,

            I think that Gillett (and I) would make the case the Gillett has the best account of what constitution/composition is (like). There are plenty of rivals out there. Subset realization. Ground. Other theories of realization, e.g. maybe Polger’s or Melnyk’s. All the non-reductive physicalists have to have some account or another. By contrast, it seems to me that the philosophers of science working on New Mechanism tend to be metaphysics averse. Maybe Glennan is the least averse of the New Mechanists. One of the points of the book Carl and I co-edited on Scientific Composition and Metaphysical Ground was to try to draw attention to the common interests in compositional explanations.

          • Hi Ken,

            I agree that the mechanistic debate and the debate(s) on the metaphysics of non-causal dependency relations in philosophy of mind, general metaphysics, and the metaphysics of science have many things in common and should interact more. Maybe the disagreement is about the direction of fit. At least those of the new mechanists that are interested in metaphysics seem to prefer a top-down approach: start with the new mechanistic approach, mutual manipulability, mechanistic modeling, etc. and then think about the metaphysics of it. What are the relata? How can we account for mutual manipulability (given that we have a clear understanding of what mutual manipulability is)? Can we account for the metaphysics of constitution purely in terms of entities and activities? Gillett and you seem to prefer a bottom-up strategy: first account for the metaphysics of all potential making-up relations, then show how it can make sense of mechanisms and constitution.

          • Ken

            Hi, Beate,
            “Maybe the disagreement is about the direction of fit.”
            Maybe.

            “At least those of the new mechanists that are interested in metaphysics seem to prefer a top-down approach: start with the new mechanistic approach, mutual manipulability, mechanistic modeling, etc. and then think about the metaphysics of it. What are the relata? How can we account for mutual manipulability (given that we have a clear understanding of what mutual manipulability is)? Can we account for the metaphysics of constitution purely in terms of entities and activities? ”

            I would have thought that the NMists begin with the science and see what is going on there. Use the science to figure out what constitution/composition is.
            But, that is what we think we are doing. Granted, Gillett is more of a metaphysician, but I have much the same background as does Craver. We both graduated in HPS having many of the same professors. And, much of what I write is very “sciency”. My choices of scientific examples are driven by trying to meet challenges to multiple realization. So, “Autonomy of Psychology in the Age of Neuroscience” is about how science works. In particular, it brings in the part-whole relation with normal color vision.

            “Gillett and you seem to prefer a bottom-up strategy: first account for the metaphysics of all potential making-up relations, then show how it can make sense of mechanisms and constitution.” Gillett does know a lot of metaphysics, but he is interested in the metaphysics of science. He helped found a society concerned with that. But, he started his theory with properties, then moved to individuals, and is now working on processes. We don’t yet have a theory of kinds. So, we are not beginning with a metaphysics of all potential making-up relations. (That seems to be the kind of thing Polger and Shapiro, 2016, might have in mind. A realization relation that includes all ontological categories.) But, Gillett was working on this while the New Mechanism was getting up steam. So, it was a kind of parallel development.

  21. Ken

    And Carl and I have been trying to engage the New Mechanists and Grounders on these issues for a while now. Harbecke contributed to our volume. Casini, Glennan, Gillett, and I did a session at the PSA last time. Casini was at a workshop we had in Newark a couple of weeks back. We wanted you to join us, Beate, but we only had a low event for folks already in town for the SMS. I’m working on an Author meets Critics session for Glennan’s new book.

  22. Ken

    I’m sort of surprised that folks do not see where the Aizawa/Gillett project fits into things. So, for example, Craver and Tabery write in their SEP article: “Because the framework concept of a mechanism is so useful for thinking about levels and explanation in the sciences, some scholars have sought in the notion of mechanism a way of fleshing out the ontological relationship of realization.” Then they proceed to suggest that Gillett is up to this. But, in truth, Gillett and I see our project as going in the opposite direction. We are trying to characterize the ontological determination relations among individuals, properties, and processes that we take to be in play in mechanisms.

  23. Ken

    But, Beate, I guess that my sense is that the New Mechanists are working on very much the same problems that Gillett and I are, only we are offering different answers.

  24. This is an interesting discussion. Given that Beate mentioned my 2015 “why there isn’t inter-level causation” paper above, I wanted to make a brief clarification. In that paper, I argue that using Woodward’s notion of ideal intervention in the way Craver does to formalize mutual manipulability (MM) leads to problems. BUT, this is an argument against *one* way of characterizing MM, not at all a proposal to throw the MM baby out (i.e., its important epistemological insight) with the inadequate modeling bathwater.

    I don’t provide an MM refinement (that’s not my goal in the paper), but I explicitly say that I want to preserve it. I also say that to preserve MM, the MM theorist does not have to be strictly committed to the interventionist notion of ideal intervention and could perhaps use a weaker notion (see e.g., introduction and p. 3749). Hence, I’m pretty much in favor of MM refinements. And I agree with Gualtiero that given that mutual manipulations happen all the time in practice, saying that they are impossible is not be the right way of explaining what’s wrong and what needs to be done.

    Here is a question about the post, just out of curiosity. Gualtiero, you say “The way I am articulating MM departs from the way Craver 2007 put it. Craver insists that the relation between micro and macro is solely a matter of constitution not causation, whereas I am claiming that a mixture of causation (between micro constituents) and constitution is involved.” Are you ruling out macro-to-micro and micro-to-macro causation?

    Thanks!
    -Felipe

Comments are closed.

Back to Top