Jessica Wilson

Jessica Wilson (University of Toronto), "Non-reductive Physicalism and Degrees of Freedom," with commentary by Michael Strevens (New York University).  The paper, commentary, and response can be found here.

Comments

[Justin]

I like this paper a lot, but I've got a few questions about the proper subset strategy and unique realization scenarios.

Let M be some special science state and P be a physical state which either is identical to M or which uniquely realizes M (exactly which relation obtains will be what is at issue). P has the following set of causal powers: {a, b, c}. Now consider two different proposals about M’s set of causal powers: on Proposal #1, M has the set {a, b}; on Proposal #2, M has the set {a, b, c}. Persuaded by what you say in your paper, I accept that if Proposal #1 is correct then M is irreducible to P even though it's uniquely realized. I also assume that if Proposal #2 is correct then M is reducible to P. My question is, how do we tell which proposal is correct? I am interested here in both the epistemology – What possible evidence could support one proposal against the other? – and the metaphysics – What are the underlying facts which make true whichever proposal is true?

To feel the force of these questions, it may be helpful to consider a pair of scenarios (roughly: possible worlds) which are as much alike as possible except that Proposal #1 is true at one while Proposal #2 is true at the other. The question then is what further differences between these scenarios must obtain, given this particular difference between them regarding M’s causal profile?

I take it that this difference in M’s causal profile is compatible with these scenarios being perfectly alike w/r/t the distribution of all physical and special science states within them. So for instance, in Scenario #2 (i.e., where Proposal #2 is true) an entity will be in P iff it's in M, since P is identical to M; in Scenario #1, it will also be the case that an entity is in P iff it's in M, but now this is due to the supervenience of M on P (I’m assuming here that where there’s realization there’s supervenience) together with the fact that M is uniquely realized. As a second respect in which these scenarios can be alike, I take it that they can be perfectly alike w/r/t all causal facts occurring at the physical level within them -- they will be alike w/r/t the causal powers of various physical states, the specific causal interactions involving physical state tokens, etc.

So again, how will these scenarios differ? One thought here is that the difference between them in M's causal profile will be bare: that M has the causal power c in scenario #2 but not in scenario #1 will be unaccompanied by any further difference. Here are two concerns with this move though. First, can a physicalist happily allow that there can be this sort of brute difference in the causal powers of special science entities which is not grounded in any difference at the physical level (e.g., any difference in the distribution of physical states, causal powers, laws, etc.)? Second, supposing a physicalist can allow this, what possible evidence could we ever get which would tell us which scenario we are actually in? The two scenarios are physically indiscernible, and so there could be no physically detectable evidence which would allow us to discern between them.

Just so it’s clear then: this is meant as a challenge to any attempt to establish irreducibility when there’s unique realizability – it’s not supposed to only apply to the case I’ve set up. Incidentally, this is a place where multiple realizability would seem to be potentially helpful (at least as a source of evidence). If in addition to P there were a physical state P* with the causal powers {a, b, z}, and entities in state P* were also in M, this would tell against Proposal #2 and in favor of Proposal #1.

Let me now try to relate my above concerns to your example involving E (a molecule) and s (a relational state involving rigidly bonded atoms) from p. 24. You say that s will have all of E’s causal powers and more besides: s will also powers to produce effects in circumstances involving temperatures or energies in which atoms can exist but molecules can’t. But if E can’t exist in those circumstances, then s can't either (assuming supervenience). The individual atoms which figure in s can perhaps exist in those circumstances, but they can't be related in the way s requires without there being an s token present; and if an s token is present, an E token will be present too. If S supervenes on e and e uniquely realizes S, then e and S will always be co-tokened. But then, given this co-tokening, I'm not perfectly clear on what sorts of considerations can be used against the reductionist who maintains that S and e have matching causal profiles.

[Jessica Wilson]

Hi Justin,
Thanks for your good question. Let me start by making some general remarks about how I presently understand the underlying metaphysics and epistemology in cases of irreducibility involving reductions in DOF. I'll then apply these to the case where a special science entity is only singly realized.

What I claim is that when we have reason to think that a special science entity M composed of some e_i exists, this reflects (at least for uncontroversially physically acceptable special science entities) that M has a reduced set of DOF compared to the set of DOF of the system of its e_i when not subject to these constraints, and relatedly, that the special science treating of M has been "extracted" from the more fundamental science treating of the e_i by imposing these (and perhaps other) constraints on goings-on involving the e_i.

What makes it the case that the imposition of some constraints leads to an entity's having such a reduced set of DOF, and more generally enables the extraction of a set of special science laws governing such entities? In general, this will be a contingent matter, that depends on the details of how the constraints affect the lower-level goings-on so as to render certain lower-level details irrelevant to certain goings-on involving the composed entity. As it happens, at relatively cool temperatures, relatively stable structures form whose properties and behavior may be characterized without attending to certain details relevant to the structure's components. As it happens, at relatively warm temperatures, unstructured aggregates exist with properties and behavior that may be characterized without attending to certain details relevant to the system's components. One of the advantages of attending to the role of reductions of DOF in the special sciences is that such reductions provide a basis for unifying the diversity of ways in which details relevant to characterizing lower-level goings-on cease to be relevant at higher-levels.

Ultimately this irrelevance of certain lower-level details to higher-level goings-on, frequently (always?) reflected in reductions in DOF of the special science entities I consider, is the metaphysical basis for the irreducibility of these entities. Supposing this is correct, then a sufficient epistemological basis for the claim that a given special science entity M is irreducible would consist in providing evidence that M was weakly emergent, in the DOF sense.

But now what if M is only singly realized?

I want to start by noting that we shouldn't be surprised at the suggestion that the requisite ontological irrelevance could be in place in cases of single realization. Consider a shape made up of some pattern of strangely shaped tiles. We could focus on the shape, or focus on the detailed pattern of the tiles, even if there was only one way of arranging the tiles so as to make the shape. This provides one illustration of how causal phenomena (involving perception) may be tuned to grains that are more or less fine, in a way that seems completely independent of whether multiple realizability is in place. So it shouldn't be surprising if the phenomenon of causal grain as manifested in there being various special sciences turned out to be independent of whether special science entities were multiply realized.

To continue, let's consider the simple case of a molecule M composed of e_i standing in rigid bonds, that is singly realized by relational atomic state s, consisting of atoms standing in particular atomic relations.

Now, everyone agrees that M will have a reduced set of DOF relative to the system of its e_i when these are not subject to the compositional constraints. I argue that s does not have a reduced set of DOF relative to the system of e_i when not standing in the specific atomic relations at issue. Hence, M will not have the same DOF as s, and so by Leibniz's Law M is not identical with s.

Why think s has the same DOF as the system of e_i when not standing in the specific atomic relations associated with s? Ultimately this is because s is not subject to the constraints that M is subject to; since it is the imposition of these constraints that gives rise to the reduction of DOF, s's DOF aren't reduced. I have two reasons for thinking that s is not subject to the constraints to which M is subject.

First is that (or so it seems to me) s could exist even when the constraints are not in place (in temperatures, for example, where M couldn't exist). Why couldn't a state of the type of s be instanced in high temperatures, but yet, due to the instability of the bonds, this token of s not realize any M? As Justin notes, this would violate supervenience of M on (just) s, in that there could be a change in M (it could exist or not) without a change in s. But this doesn't bother me overmuch, since we can just build the constraints into the supervenience base (which is not to say that we should build the constraints into s!). In any case the second reason doesn't require that we assume that s can exist without M's existing.

Second is that s is an entity governed by the laws of atomic physics, which laws (unlike the laws of molecular physics) do not presuppose the imposition of the constraints to which M is subject. Since s is governed by laws that don't impose the constraints, s is not subject to the constraints.

The fact that s is an entity governed by laws that do not impose the constraints to which M is subject provides us with another route to M's irreducibility, even supposing M is singly realized. I say that what causal powers an entity has is a matter of what it can do; and I say that what an entity can do is specified by the laws that govern it. Since s is governed by laws that do not involve the constraints to which M is subject, and given that the laws governing M are extracted from the laws governing s, s will be able to do all the things that M can do (e.g., bring about effects in relatively low temperatures) and more besides: s will also powers to produce effects in circumstances involving temperatures or energies in which atoms can exist but molecules can’t. Justin worries that the claim that s has more causal powers than M requires that s be able to exist without M (in particular, be able to exist in circumstances where molecules don't exist), but this is incorrect: for s to have causal powers to bring about effects in circumstances not permitted by the constraints on M only requires that s be connected by (atomic) law to atomic goings-on that can exist in such circumstances. s is so connected; M isn't; hence s has more causal powers than M.

[Brad Weslake]

I really like this paper. Independently of whether the degrees of freedom approach to the relationship between theories yields the metaphysical conclusions argued for, it certainly is an excellent framework within which to address the questions. And I agree with Strevens that the account has a nice upshot for an independently attractive account of explanatory relevance. Besides praise, I have a few questions.

Firstly, I am unclear what argument is being hinted at in this sentence (pp. 18-19) -- "Moreover, assuming that token instances of E might have been otherwise realized, neither is it appropriate to do as Kim (2001) suggests, and take instances of E to be token identical to whatever physical realizer realizes it on a given occasion (as per Boyd 1980)". Could you say a little more about which line of argument you have in mind? I would have thought a lot more work would be needed to establish this claim.

Secondly, concerning the argument in 4.1.1, I don't see how Ockham's razor cuts against reduction -- from the standpoint of the reductionist, we already have all the properties (I'll use properties interchangeably with your "details") we need and so none at all are being added at the level of characterising E. The reductionist is going to claim that the simplicity gained by neglecting some of the irrelevant base properties in characterising the higher-level properties is purely explanatory, not ontological -- we are throwing properties away for explanatory purposes; bringing them back in wouldn't be multiplying entities, just ceasing to ignore them. Maybe this sort of argument is just an instance of your first objection (p. 24), that "The arguments of the last section are unlikely to convince a hard-core reductionist that E is not ontologically reducible". But for someone approaching the issue from an explanatory perspective, gaining all of the explanatory virtues of your account without the additional metaphysics is going to look pretty attractive.

[Jessica Wilson]

Thanks for your good questions/remarks, Brad. I'll respond to each in turn.

1. In the section you're referring to, I'm sketching the dialectic of the debate between reductionists and non-reductionists over whether multiple realizability motivates irreducibility. Let me lay out the broader dialectic, then say a bit more about the role the Boydian argument plays.

We start with a reductionist who wants to identify a type of realized entity E (say, a type of mental state) with a type of realizer entity R (say, a type of brain state). The non-reductionist responds that in cases where E is multiply realizable by entities of different types R1, R2, ... (say, a type of brain state, a type of silicon state, ...), it isn't appropriate to identify the type of E with any one of the types that can realize it. The reductionist then has a couple of responses. (1) Take the proposed identity to hold between the type of E and the disjunction of types of E's realizers; to which the non-reductionist will respond that the type of E is a natural kind, whereas the disjunction is not, as indicated by 'being E''s being projectible, whereas being such as to satisfy the disjunctive predicate is not; to which the reductionist might respond either that such explanatory considerations as projectibility are irrelevant to issues of ontological reduction, or that characterizations in terms of disjunctions are characterizations in terms of what is common to the disjuncts, so that the disjunction should count as picking out a natural kind/being projectible. (2) Take the proposed identity to hold between a given token of E and the token of whatever type realizes it on a given occasion.

It is in response to this latter proposal that the Boydian move comes into play. The reductionist moves to token identity presumably because they find the arguments from multiple realizability against type identity compelling. But if so, Boyd argues, then they should also reject token identity, since any given token of E *might* have been otherwise realized. You are right that more needs to be said here -- in particular, about the identity conditions of the tokens of E at issue (events, states, states of affairs, or whatever they might be). Hence it is that the Boydian response isn't decisive. I don't get into the details in my paper because my goal isn't to come down on one or other side of the debate over whether multiple realizability is sufficient for irreducibility, but rather just to show (by reference to the ongoing multitracked and undecisive dialectic) that it isn't obvious that this is so.

2. Yes, I agree that reductionists will generally think that they have Ockham's razor on their side, precisely because they "already have all the properties ... we need and so none at all are being added at the level of characterising E". I am attempting to wield Ockham's razor in a different way than usual, by suggesting that it can be also read as calling for not positing any entities beyond necessity in ontologically characterizing an entity. Recall that all parties agree that the special science entities E at issue exist; what I argue is that given that E is weakly emergent, the associated reductions in DOF pertaining to E's occurrence, properties and behavior indicate that E's ontological characterization should not include any irrelevant ('beyond necessity') details (namely, those associated with DOF that E doesn't have).

I also agree with you (as an application of my p. 24 remarks) that this implementation of Ockham's razor needn't force a reductionist to switch horses, since they may privilege their application of Ockham's razor over mine. Ultimately (as I go on to say) my goal is to provide a principled way for non-reductionists to stay on their horse.

Still, how might considerations of explanatory virtue legislate between these options? Following Strevens, I am inclined to think that considerations of explanatory relevance weigh in favor of an account on which E's ontological characterization not citing unnecessary details over one that does. The question to be asked is: Why does it make sense (as it so often does) to (as you put it) throw away or ignore properties for explanatory purposes? That what is being explained involves entities whose ontological characterization does not include the thrown-away or ignored properties gives a straightforward answer, that moreover provides an ontological ground for our epistemology---a desirable result for the sciences, at least. But it is at least unclear both what answer the reductionist can give, and whether this answer will provide the desired ontological ground; hence I would need to hear more about how "from an explanatory perspective" the reductionist will be able to gain "all the explanatory virtues of my account without the additional metaphysics".

A related legislative consideration may also be worth noting. I think many reductionists have thought that they had to buy reducibility, even if it meant some sacrifice in making straightforward ontological sense of (among other things) explanatory relevance, in order to gain physical acceptability (as per one leg of the dilemma Kim raises for non-reductive physicalists). Insofar as (for the cases I consider) the special science entities E are, while irreducible, nonetheless clearly physically acceptable, this source of motivation for reductionism is lessened to a considerable extent: we can have our physical acceptability and our ontologically grounded explanatory relevance, too.

[noa latham]

Hi Jessica,

I like the idea of finding ways of looking at reduction that go beyond the issue of multiple realisability. But I haven’t yet understood your basic notion of weak emergence. In your 2.1.5 you say that a state s that is identical to physical entities in a purely physical relation is not weakly emergent because giving the configuration of s will involve giving the configuration of all the physical entities composing s. Going back to your system of two point particles rigidly bonded in 1.3.1, I understood you as showing that such an entity has a reduction of configuration DOF from 6 to 5, so is weakly emergent. Isn’t such a system identical to physical entities in a purely physical relation? Why doesn’t this show that weakly emergent entities needn’t be irreducible?

In the second part of 2.1.5 you say that states identical to disjunctive combinations of entities are not weakly emergent because they have the same DOF had by a system of the parts prior to combination. I can see why this might be true if the parts here are the individual disjuncts. But this would mean that a lot of interesting high-level entities such as jade, along with heterogeneously disjunctive entities, do not count as weakly emergent. So I don’t yet have a clear idea of what kinds of entities you take to be irreducible.

[Jessica Wilson]

Response

Thanks for your questions, Noa. I could be clearer in my paper on both the points you mention.

1. With respect to the system E of two point particles that are rigidly bonded: yes, E is weakly emergent (assuming we have reason to think it exists), and no, by my lights E is *not* identical to physical entities standing in a purely physical relation. E is realized by, but not identical with, a state s consisting of physical entities e_i standing in purely physical relation. My reasons for thinking that E is not identical with s are as noted in my response to the first comment in this thread---basically, s is not subject to the constraints that E is subject to---for one thing, s could exist at very high temperatures, where the bonds at issue would not be rigid; for another, s is governed by laws that don't impose the constraints at issue. Hence s doesn't have fewer DOF than the system of e_i when not constrained to stand in rigid bonds; E does have fewer DOF than the system of e_i when not constrained to stand in rigid bonds; hence s and E aren't identical.

2. On my account, disjunctions of e_i or states consisting of e_i standing in relation turn out not be weakly emergent, since such disjunctions turn out to have the same DOF as the unconstrained e_i (see section 3 of my response to Stevens). That disjunctions are not weakly emergent is compatible with jade's being weakly emergent (hence irreducible), supposing that we do not identify jade with a disjunction.

There is some split in intuitions concerning whether boolean combinations of entities in a given reduction base should also be seen as part of the reduction base. I am inclined to think so, on grounds that the reductionist gets to help themselves to resources (like logic) that very plausibly do not invoke additional ontology. Some (e.g., Strevens) suggest that to characterize an entity in disjunctive terms is already to characterize it at a higher level of abstraction (hence in some sense 'irreducible'). I think this is incorrect, but I also think that in many cases an irreducible entity is to be found in the vicinity; namely, one that is characterized only in terms of what is common to all the disjuncts (of the sort that would be associated with a reduction in DOF).

[noa latham]

Hi Jessica,

Thanks for your response. If we have a state s consisting of e_i in a physical relation R, then I still don’t see why R doesn’t give s a lower DOF than that of the system of e_i that isn’t subject to R. I was thinking that a rigid bonding relation (in some imaginary physics) could be an example of such an R. This would make s weakly emergent and reducible.

I get the impression that you want to say that E or M is not identical to s because it is subject to constraints that can’t be built into s because they can’t be specified in the language of physics (narrowly construed). But then it seems as though in saying that a special science entity is irreducible, the work is being done by the claim that what it is to be such an entity cannot be expressed in the language of physics, rather than that it has a reduced DOF.

[Jessica Wilson]

Hi Noa,
Thanks for the very helpful followup. I think I see the concern, and here's how I presently want to respond.

First, I do want to say that it is sufficient for a realized entity E not to be identical to its realizing state s that E be subject to constraints that s is not subject to, but I take this to be a metaphysical matter, not anything having to do with the language of physics (or whatever science treats of entities and relations constituting s). In particular, I take it to be a metaphysical matter that, when certain constraints are imposed on phenomena at a lower level, a distinct level of causal/interactive grain may become manifest, which is reflected in the reductions in DOF associated with the various special science entities I discuss in my paper.

However, your remarks indicate that I need to be more specific in my formulation of Weak Emergence (DOF) about the fact that the relevant reductions in DOF come about as a result of imposing constraints not operative at the lower level. Let me illustrate what I have in mind using the case of a molecule E composed of two rigidly bonded atoms e_i.

As I was thinking about this case, E is a molecular special science entity (understood, contrary to fact, as involving rigid bonds) realized by an atomic state s consisting of atomic e_i standing in atomic relation. Here the atomic relations at issue are not themselves appropriately understood as "rigid", both because states involving the atoms standing in these relations (involving electromagnetic forces of attraction) may be present even at high temperatures and more generally, because s, being an atomic state, occurs at a level of causal/interactive grain that doesn't impose the constraints associated with rigid bonding. It is rather in virtue of the atomic phenomena being constrained to occur in only certain circumstances that phenomena associated with a different level of causal/interactive grain (including E and E's properties and behavior) occur. Hence it is that E has a reduced set of DOF compared to a system of e_i not constrained to stand in rigid bonds, whereas s does not have a reduced set of DOF compared to a system of e_i when not standing in non-rigid atomic relations; and hence it is that E, but not s, is weakly emergent (relative to atomic phenomena).

You raise the concern that reductions in DOF are not sufficient for weak emergence, by consideration of an alternative way of understanding the molecule case, on which there can be atomic bonding relations that have their rigidity built into them, as it were. If there were such "rigid" atomic relations, the state s consisting of atoms in these rigid relations would have a reduced set of DOF relative to the atoms when not so related. But in this case s is clearly reducible to atomic phenomena, and so weak emergence can't just be a matter of reductions in DOF.

This seems right. In my paper I already acknowledge that not every reduction in DOF will correspond to a weakly emergent entity E, since some such reductions (e.g., that associated with the center of mass of a system of particles) needn't be associated with any entity at all, much less a weakly emergent one. Your case indicates that another way in which a reduction in DOF need not indicate the weak emergence of an entity is where a composed entity has a reduced set of DOF (as in the "molecule" in your case), but not as a result of constraints being imposed on phenomena operative at the level of causal/interactive grain associated with the composing entities. This response needs tightening up, but that's the idea (to be developed in the next draft). Thanks again.

[noa latham]

Hi Jessica,
I get a much clearer idea now of how you understand your molecule example. Thanks. A couple more thoughts: You might want to say more about how a system of e_i constrained to stand in the relation R can have the same DOF as when the e_i are not so constrained. You want to avoid the trivial sense in which the e_i lose some freedom in virtue of being no longer permitted to stand in not-R. Perhaps the place for this is where you are refining your notion of DOF. Once it’s clear that the notion of reduction of DOF is not so weak that it applies to all but completely unconstrained systems, then the question arises how this helps us understand why some systems have a higher level of causal/interactive grain or are irreducible. It can’t simply be because they have a reduced DOF.