Corey Maley & Gualtiero Piccinini --- The Metaphysics of Mind and the Multiple Sources of Multiple Realizability

[New Waves in Philosophy of Mind]

The Metaphysics of Mind and the Multiple Sources of Multiple Realizability

Corey Maley & Gualtiero Piccinini

     

 Multiple Realizability (MR) has been at the center of controversies on the metaphysics of mind.  On one hand, anti-reductionists have invoked MR to support their view.  On the other hand, reductionists have questioned MR.  In this paper, we aim to show how an appropriate and independently motivated ontology sheds light on MR and allows us to make progress on the metaphysics of mind.  An egalitarian ontology is such that neither the parts are prior to the whole nor the whole is prior to the parts.  Within such an egalitarian framework, we will argue that there are several sources of MR.  Nevertheless, both reductionists and anti-reductionists have to give something up.

A PDF of the paper is ready to view and download in the attachment below. 

A direct link to the PDF: http://goo.gl/G35jz

[Kenneth Aizawa]

Hi, Corey, Gualtiero,

As you know, there are many points of contact between what you are up to and what Carl Gillett and I (and Tom Polger and Larry Shapiro) are up to.  But, one point you might help me understand is what is going on when you claim that the Dimension view of realization "makes it too easy to find cases of multiple realization".  I am not sure how easy it should be to find cases of multiple realization.  How easy or hard should it be?  What do you mean by saying that MR is trivialized?

Let me try to make the point a little sharper.  You sometimes talk about realization in terms of a subset model, but then again sometimes on a model of mechanistic explanation.  Let me set aside the subset stuff to focus on the mechanistic account.  On p. 11 of your paper you write,

N and M are distinct realizations of T when and only when:

1.     both N and M realize T;

2.     N’s realizing T is mechanistically explained by some subset of N’s components (call this C(n)) and their organization (call this O(n));

3.     M’s realizing T is mechanistically explained by some subset of M’s components (call this C(m)) and their organization (call this O(m)); and one of the following is satisfied:

4.     [MR₁] C(n) = C(m) but O(n) ≠ O(m),

5.     [MR₂] C(n) ≠ C(m) but O(n) = O(m), or

6.     [MR₃] C(n) ≠ C(m) and O(n) ≠ O(m).

This, however, looks a lot like the view that Carl and I have, only weaker:

A property G is multiply realized if and only if 

(i) under condition $, an individual s has an instance of property G in virtue of the powers contributed by instances of properties/relations F₁-F_n to s, or s’s constituents, but not vice versa; 

(ii) under condition $* (which may or may not be identical to $), an individual s* (which may or may not be identical to s) has an instance of property G in virtue of the powers contributed by instances of properties/relations F*₁-F*_m of s* or s*’s constituents, but not vice versa; 

(iii) F₁-F_n ≠ F*₁-F*_m and 

(iv), under conditions $ and $*, F₁-F_n of s and F*₁-F*_m of s* are at the same scientific level of properties.

(See, for example, The Autonomy of Psychology in the Age of Neuroscience (with Carl Gillett).  (2011). In Illari, P.M., Russo, F., and Williamson, J.  (eds.) Causality in the Sciences.  Oxford University Press. (p. 207).).

 

Your condition 1. looks like our conditions (i) and (ii).  Your conditions 2. and 3. seem redundant (or superfluous) to me, but ok.  Then your conditions 4.-6. looks like an explicitation of our condition (iii).  But, you don't have anything that corresponds to our clause (iv).  Our clause (iv) matters, of one does not count the realization of, say, pain at the neuronal level and at the biochemical level as multiple realizations of pain.

So, here is how things seem to me: you charge us with having too permissive a theory of MR, but then your theory of MR is more permissive.

I think this is worth noting, since Carl and I get this "trivialization" objection a lot and I am not sure what to make of it.

Ken

 

[Gualtiero Piccinini]

Ken,

 

thanks so much for your thoughtful comment. i'll need to think about it. here is what i can say for now.

 

i think we need to add something like your condition (iv). but...

 

i also think that our conditions 4-6, as articulated in the text of our paper, are stronger than your condition (iii). you let any difference between sets of lower level properties count as different realizers, whereas we don't (or at least we don't want to). maybe this isn't explicit enough in the formulation you cite. i'll have to think about it.

[Kevin Ryan]

Hi Corey and Gualtiero,

Thank you for the paper - I'm particularly interested to see how you develop the egalitarian ontology in future works.  At the moment, I have one quick, minor question about the notion of causality you use, as well as a thought about the triviality problem Ken raised.  In terms of the first point, does your account assume some form of casual dispositionalism or realism to hold in the world?  I don't think the presence or lack of causal glue would count for or against your theory, but it might have an impact on how we would deal with certain instances of MR (such as MR in purported cases of extended cognition). 

In regards to triviality, is it possible that one of the key problems being face, here, is a tension in the two main audiences for an account of realization? Craver and Wilson, in their article for The Philosophy of Psychology and Cognitive Science (2007), examine the issue concerning differences between what's expected from realization by "metaphysicians of mind" and "cognitive scientists".  Along these lines my intuition is that. for the latter group, a condition akin or identical to what Ken suggests may be necessary, while the former may find it unnecessary or, perhaps, even have problems with it (it's too strict if we're considering pain or cognition via the Martian intuition, for instance).

[Eric Funkhouser]

I am sympathetic to the mechanistic aspect of your account of MR. But, I have concerns about the egalitarian ontology and the subset account of higher-level properties. Well, mainly I have problems with the latter.

I've long had problems with the subset view. It seems to me that higher-level properties typically have causal powers not found at the lower-level. I don't have in mind any form of strong emergentism -- the causal powers of the higher-level properties are still determined by the lower-level properties. I think a large part of the appeal of the subset view comes from mistakenly thinking of properties as obeying the same kind of mereological principles that hold for concrete objects. But, be that as it may...

There are additional problems, it seems, when we try to combine the subset account with an egalitarian ontology. Given that, on the subset account, lower-level properties can do everything that higher-level properties can do, but not vice versa, it seems that they aren't on equal footing at all.

[Paul Schweizer]

Hi Corey and Gualtiero, thanks for your very interesting paper, and apologies for posting a reply rather late in the day. It seems to me that 'multiple realizability' is a very broad term that applies to a range of different types of phenomena, and hence that the same analysis might not uniformaly apply to all these different types. Your analysis is geared towards multiple realization in terms of functional characterization, such as being a corkscrew, and this will naturally invoke causal powers. But there is also multiple realization in terms of abstract structure, and this seems quite different. In the latter case it's isomorphism at some relevant level of description which unites diverse instances, and this might have nothing to with causal powers. For example, being a knight in a game of chess is determined by abstract structural role in the game. And multiple realization of abstract computational formalisms seems to belong to this latter type. In section 5 you discuss computation, but I wasn't sure exactly how it fits with the analysis of being a corkscrew (and I may well just be missing something here!). Abstract mathematical functions seem very different than biological or mechanically specified functions, and I was unclear about the overarching unity of their treatment.

 

One small remark on the points brought out at the end of section 2 concerning 'observer-relativity'. Is the difference between water and corkscrews not also a difference of natural versus 'conventional' kinds? So even though membership in either kind might be based on criteria whose satisfaction (or not) is a matter of objective truth, still the criteria for conventional kinds are themselves less 'objective' because they stem from human practices and stipulation rather than fundamental microstructure?

 

And on an entirely different point, I found your suggestion that the higher level properties are 'subtractions rather than additions of being' from their lower level realizers a very interesting perspective.

 

Best,

Paul 

 

       

[Dan Weiskopf]

Corey & Gualtiero:

I agree that the case against MR made by Shapiro and others doesn't go through. If I can put in a plug for some of my own work, I made a similar argument in 'The Functional Unity of Special Science Kinds' (BJPS, 2011). There I argued that there are genuinely MR functional categories that meet (what I called) the Really Distinct Realizers condition. Your various forms of MR strike me as more precise ways of spelling out what I was trying to capture with that term. 

I just have two quick comments. One is directed towards getting you to sharpen your discussion of when MR properties are apt for forming kinds. In your Sect. 6 you rightly noted that laws are rarely central in many sciences. I also agree that being functionally specified doesn't (contra Shapiro) mean that all generalizations involving a category are 'merely' analytic. But we still need some sort of positive story about the significance of MR categories. I'd suggest that a positive account of functional kindhood can be found if we look to various modeling practices that involve these functional categories. Roughly, to the extent that a category plays a significant role in a well-confirmed explanatory model, it should be considered a kind. This moves us away from the nomic conception by giving a positive account of kindhood to put in place of the inductivist conception that dominated post-positivist thinking about kinds.

Second, I've recently begun to suspect that MR is much less important than has been advertised. For one thing, whether a property or phenomenon is MR depends (as you point out, and as I also discuss) on how we specify it. Choosing a more generic or specific property may change the verdict about MR. But what does this verdict in and of itself tell us? Concerns about MR have often focused on reductionism. However, reductionism seems to me to be a dead letter. Even if some property is not MR, it hardly follows that it is reducible (certainly it doesn't follow once we abandon the nomic conception of interlevel relations). So MR is not needed to secure this particular form of autonomy. This isn't to say that we don't have any interest in learning that there are many possible ways to build a functionally specified system, or many ways to map a system's properties onto other of its properties. Certainly we might. So MR might be interesting just in itself. Its role in resolving philosophical controversies about reduction, however, seems to have been bypassed once we realize that there are few to no reductions out there even if MR is generally false. I'd be curious to know whether you think MR has any significance, apart from its being leftover from these philosophical debates?

[Gualtiero Piccinini]

Hi  Kevin,

 

Thanks so much for your comments and encouragement. I am not sure what you  mean by "causal dispositionalism." In principle I am inclined to stay neutral, for the purposes of this paper, between the view that properties are causal powers (perhaps identical to qualities, a la C. B. Martin and  John Heil), categorical bases of causal powers, or even natural classes of resemblances between particulars (though I really dislike the latter view). we probably say that properties are causal powers in the paper, though, don't we? If so, that's must be my view :-)

 

I'd be interested to hear more about how "causal dispositionalism" might affect the way we deal with MR in the case of extended cognition.

 

I don't like to keep metaphysics and philosophy of science separate a la Craver and Wilson, though perhaps I should re-read their paper. I  think metaphysics and philosophy of science should work together. And I'm inclined to accept that Corey and I need Ken's clause (iv). 

[Gualtiero Piccinini]

Eric,

 

Thanks so much for your interesting question. You are right that according to us, lower level properties can do more than higher level ones. But that doesn't mean that they are ontologically less fundamental because in fact they are just (proper) subsets of them. A (proper) subset is neither more nor less fundamental than its superset. It's just a partial aspect of its superset, as it were.

 

Things would be different with other versions of the subset view. For example, Shoemaker says that properties "bestow" powers, and adds that  higher level properties bestow powers that are a subset than those bestowed by lower level properties. This "bestowing" business generates serious problems, including that it suggests the kind of hierarchy between higher and lower properties that you mention as well as a version of Kim's causal exclusion argument (if the samel powers are already bestowed by the lower level properties, what do we need to higher level ones for?).

 

Can you give me examples of powers that are found at a higher level but not at a lower level? (as opposed to descriptions that apply at the higher level but not [easily] at a lower level?)

[Gualtiero Piccinini]

Dan,

 

Thanks so much for your comments and  for reminding us to look at your paper.

 

As to the significance of MR, that's a very useful question and I'm very glad you asked. Yes, I think MR is significant: it's significant precisely in helping us identify the correct account of inter-level property relations, namely, the version of the subset view that is sketched in our paper. What MR shows is that the same set of causal powers (= higher level property) can be a subset of different supersets (= lower level properties). Corey and I should make this explicit in the paper!

[Gualtiero Piccinini]

Paul,

 

Thanks so much  for your insightful and encouraging comments.

 

What you call "conventional kinds" (which are a subset of functional kinds; or kinds defined by their teleological functions; the other subset being physiological kinds in biology) do require a more sophisticated account than non-biological "natural kinds". Corey and I are almost done writing a paper that gives a general account of teleological function in biological systems and artifacts. We argue that teleological functions are contributions to the goals of organisms. (Obviously the devil is in the detail.)

 

Knights in chess are an example of a functional kind, and so are computers. I have written a lot about computation in the physical world and articulated and defended a mechanistic account of computation. So I think computation fits right in with all the other examples of MR, except that as we mention in the paper, it's more than just multiply realizable; it's medium independent.

[Kenneth Aizawa]

Dan, 

Does this mean what it seems to mean: "whether a property or phenomenon is MR depends ... on how we specify it".  So, do you think that you can take a given property and, say, stop it from being MR by describing it differently?  Do you want to say that realization is not a relation among properties, but a relation among properties under a description?  Isn't it rather, that by changing descriptions you maybe change the property you are talking about?  This seems like a big move that merits articulation and defense.  Cory and Gualtiero don't say very much about this.  Bechtel and Mundale might have something like this in mind in that late section of their paper, but they more or less just throw it out there.

Ken

[Kenneth Aizawa]

A bread knife has the power to cut bread, but the handle of a bread knife does not have the power to cut bread.

An electric iron has the power to iron clothes, but the power cord of an electric iron does not have the power to iron clothes.

Craver gets at this idea in his picture of S-phi-ing by way of a bunch of R's psi-ing.

[Thomas Polger]

Corey and Gualtiero,

I enjoyed looking at this.  Lately I have been thinking a great deal about how to situate the work that Shapiro and I have done on MR in a broader context. I think MR is important because it has a job to do in justifying the taxonomies and practices of the "special" sciences. And I think this means that only some kinds of variation in nature count as MR, viz., the kinds that could (at least potentially) do that job if they are actual.  I think you agree.

With that in mind, let me note a few things you say that raised questions for me from this "view from 30,000 feet."

(1) You write: "Many different substances can be used to build corkscrews, but their only insightful, predictive, explanatory, non-wildly disjunctive, specification is that they must lift corks out of bottles (or do so in such and such a way)."  This is a familiar claim, but not very egalitarian.  Why "only"? Why should it not be good enough that that is one of many specifications? In the familiar jargon: Why think the autonomy of psychology is secured only if psychological explanation is mandatory?  In my "Some Metaphysical Anxieties of Reductionism" I argue that the reduction/antireduction debate is confused because both sides accept a strong version of the autonomy claim, and they should not. (Yeah for explanatory pluralism!)

(2) I'm not convinced by your substitution test, even as a first approximation.  Suppose, in your example, instead of replacing the wooden table leg with a steel one, I replace the legs with cables that suspend the top from the ceiling, or with electromagnetic supports that float it on a magnetic field.  These seem like genuinely different ways of doing the table leg job, and thus seem like prima facie cases of MR to me. (In contrast, the gallium arsenide circuit fails to do the circuit job in that system, so I would say is not a case of MR.)  But I admit that I am applying my criteria, and maybe our criteria differ after all.

(3) Speaking of which: Two things about the six part recipe you provide. First, for simplicity, the conjunction of 4-6 is equivalent to ~[(C(n)=C(m)) & (O(n)=O(m))], right? But I think you really want distinctness, not just non-identity; for as you note later, non-identity is not sufficient for distinctness.  (See also  Daniel Stoljar on distinctness, and David Sanford, "Distinctness and Non-identity", Analysis, vol. 65 no. 4 (October, 2005), pp. 269-74)

(4) Anyhow, I am a pluralist about explanation yet I don't think I need to deny some metaphysical priority claim. In that way, it seems, my pluralism differs fro your egalitarianism.

I hope these are helpful observations. My guess is that our views are not very far apart, really.  Cheers!  

[Eric Funkhouser]

"Can you give me examples of powers that are found at a higher level but not at a lower level? (as opposed to descriptions that apply at the higher level but not [easily] at a lower level?)"

On my view properties are abstractions, somewhat in the Lockean sense that involves subtracting away detail. We normally think of this subtraction as applying to higher-level properties, as your subset model does. Namely, on your view higher-level properties just subtract away some of the causal powers distinctive to their realizers. But this subtraction can go both ways – and I think it should in a truly egalitarian metaphysic. So just like higher-level properties subtract away lower-level detail, lower-level properties subtract away higher-level detail. Of course, given realization or supervenience theses, the lower-level properties necessitate this higher-level detail (so, say, some kind of exclusion worry could still arise). Nevertheless, if we think of properties as abstractions the lower-level properties need not contain the details distinctive of the higher-level. On the causal subset model (which I do not favor, by the way), these “details” would be causal powers.

I take these abstractions with metaphysical seriousness, so it’s not merely a matter of different descriptions applying to the same system. And I don’t think there are abstractions for every possible description. Nor do the descriptions need to perfectly match the world in order to successfully refer. Normal standards for successful reference apply, and there is room for revision. To give a concrete case, suppose folk psychology is true and there really are beliefs and desires that approximate our folk psychological descriptions. Also suppose that beliefs and desires are realized in the microphysical. Microphysical properties, as abstractions, realize beliefs and desires but they do not themselves contain that higher-level detail. Microphysical properties only contain details like charge, mass, and the like. Such microphysical properties – not even collections of them – do not contain the propositional content and rationalizing details that we find with beliefs and desires. So, they do not cause intentional actions, say. I think that many dual explananda theorists at least tacitly accept an account of properties like this. (And several property theorists, of course, more explicitly accept an account of properties as abstractions like this.) This point generalizes to all non-reductive, higher-level properties, and there is nothing special about psychology in this regard.

[Gualtiero Piccinini]

Tom,

 

Thanks so much for your thoughtful comments.

 

(1) If I understand correctly, I can say that in no way are we motivated by trying to save the autonomy of psychology. In fact we deny that psychology is autonomous in most interesting senses of the term. It's just that when we look for relevant explanations of phenomena, we identify relevant causal factors. Those are the "higher level properties." Thanks for the reference to your paper. We should look at that.

 

(2) Thanks for the example. I'll think about that. We were aware that our "first approximation" test was not going to handle every case,  but maybe we need to say more about this.

 

(3) Thanks for that too. We probably need to say something about that.

 

(4) Thank God we still disagree about something :-)

[Gualtiero Piccinini]

Eric,

 

Have you written about the view of properties you describe? (Pardon my ignorance.)

 

I agree that properties are abstractions in Locke's sense. But I'm not sure I understand what you mean by lower level properties subtracting details away from the higher level properties while still necessitating the higher level properties. I don't see how you avoid causal exclusion.

 

I'd say that the lower level properties "drown" the  higher level details in so much additional details that the higher level properties become invisible ("implicit" in the lower level descriptions).