First of all, the approach Craver takes for defining levels is notable in many ways. What I find especially important, however, is that he develops his definition in an interdisciplinary framework, paying close attention to compatibility with or even application in neuroscience. So what are levels as used in such a scientific context? In his article of this volume, Craver extends his original definition of mechanisms a little to accommodate for the existence of lower-level mechanisms that take part in the realization of higher-level mechanisms.
I use the term ‘mechanism’ permissively to describe non-aggregative compositional systems in which the parts interact and collectively realize the behavior or property of the whole. Mechanisms are by definition more than the sums of their parts: they have properties their parts do not have, and they engage in activities that their parts cannot accomplish on their own. (Craver this collection, p. 16)
Mechanisms as construed here are entities and activities organized in non-aggregative compositional systems, such that they are productive of regular changes from start or set-up to finish or termination conditions, and the properties of the whole mechanism are produced collectively through the interaction of its component mechanisms. This establishes the basis for Craver’s introduction of levels of mechanisms.
Craver’s three defining questions—namely about the relata, the relations, and their placement—constitute a valuable contribution to the conceptual clarification of the term “level” helping to capture the criteria for the correct usage of the term. This already hints at the descriptive pluralism Craver promotes, meaning that there is a set of equally correct ways to use the term depending on the respective answers to these three key questions. In the case of mechanisms, levels are best individuated, according to Craver, in terms of a part–whole relationship between the property “ψ-ing” of S, given that S is a mechanism as a whole, and the property “ϕ-ing” of X, given that X is a mechanism component that is a constitutively relevant spatiotemporal part of S.
Summing up Craver’s position on levels and levels of mechanisms as I understand it, they:
are metaphors with multiple distinct conditions of correct usage dependent on the relata, relations, and their placement and
refer to part–whole relationships.
Levels of mechanisms have properties of mechanisms and properties of their parts as their relata (as opposed to levels of size, which have objects as their relata),
are always non-aggregative (though aggregative levels do also exist),
are not monolithic, but constitute a local organizational part–whole hierarchy, while
the part–whole relationship must satisfy the constitutive relevance condition;[1] while
there is no causal relation between them,
they bear explanatory potential,
and, finally, the placement of entities on levels of mechanisms is weak in the sense that for all entities that are not related as part and whole it can be said that they are on the same level.
What might the difficulties with this account of levels be? As a minor point, first, there might be some implications of using the concept “metaphor” in connection with levels. By definition, a “metaphor” identifies two things—a primary and secondary subject—with one another, such that one of the two can be captured in description more powerfully (Hills 2010). What could the primary and secondary subject in a level-metaphor be? The primary subject would probably be a level in the sense of a level of mechanisms, the secondary subject could maybe be a plane, a stage, or a degree. But how would that help elucidate what the primary subject levels actually are? As far as I can see, using “level” as a metaphor would more effectively describe what a level is, but not actually define it and, thereby, simply capture what our intuition about levels is in the first place—namely that it is somewhat analogous to a level in the secondary subject sense. Also, it seems that conceiving of “levels” as a metaphor would already somewhat negatively answer the question of what levels actually are, as solely existent as a figure of speech—an analogy that could be eliminated without any ramifications. Craver’s descriptive pluralist approach is formulated specifically to counteract elimination of levels and thereby to sustain their application in science. However, descriptive pluralism obviously does still act on the assumption that levels are metaphors, and only describes conditions that fit their application better than others. Therefore, this approach does not seem to be particularly helpful for our intended and certainly desirable goal.
Another issue I would like to raise is that in Craver’s account of levels as presented here, the key defining feature of levels seems to be a relation condition[2] between certain entities. This already becomes apparent with the three questions mentioned above, which aim to help us adequately describe specific instances of levels. Levels of mechanisms in particular are specified as a part–whole relationship between properties of mechanisms that are located on different levels. But does this really capture what levels actually are? Or does it rather render the relation condition between levels more precise, instead? While helpful to set the criteria for conditions under which the term “level” is correctly employed, and highlighting distinguishing features of different levels, saying that levels are essentially relations between sets of entities is at best an indirect or descriptive definition of levels and does not seem sufficient for a complete definition. It leaves open how levels come into existence, what their ontological status is, and why we posit certain entities on the same level in the first place, that is, what the commonalities of entities are that lead them to be on the same level.
From the key aspect of a part–whole relationship in Craver’s account stems the notion that levels are local and non-monolithic. This means that only those entities that are involved in such part–whole relationships can intelligibly be said to be on different levels. What are our theoretical options for conceptually covering all other entities? Since they don't fulfil the part–whole criterion they cannot be on distinct levels and, therefore, in a sense they are all on the same level. However, according to the definition of levels at hand, to be on any level means that there are other levels, which are distinguished from the first level by the part–whole relationship of the entities involved. Since the entities under consideration do not exhibit this kind of relationship, they are on no level at all. In general, this seems like a reasonable option. But let's consider the case that entities that are not in such a part–whole relationship do, nevertheless, share some features—for example a hedge and a fence both one meter high. In accordance with levels not being monolithic and the previous considerations, these would not be on the same level, but on no level at all. What exactly is wrong, however, with saying that two entities sharing the feature of being one meter high but which are not related as part and whole are on the same level? One could, of course, simply invoke the account of levels of mechanisms and argue that it is designed such as not to warrant such a level. But does this limitation really procure us a better understanding of “level”, or could it rather be too restrictive for that purpose? Its consequence, at any rate, is a very strong focus on the vertical dimension, namely the relation between levels, whereas the horizontal dimension, that is, entities related qua being on the same level, is somewhat neglected. So, let us ask, what are the criteria for two entities being on the same level? It is exactly this relation between entities on the same level that the concept is primarily supposed to capture, and yet which seems to be underspecified by the definition provided. And how similar do two part–whole relationship units have to be in order for it to be correct to say that their respective wholes and parts are on the same level? Or does it even follow that two things that are not part of one and the same part–whole relationship cannot even be on the same level at all?
As a third point, finally, there remains the issue of the extent to which levels of mechanisms are similar or distinct from levels of realization and emergence, respectively. All three kinds of levels share that an application of the concept of inter-level causation is not feasible in their case, since they do not fit classical assumptions about causation such as non-synchronicity of cause and effect—a very substantial point Craver emphasizes in the target article. But what are the differences between these kinds of levels?
According to Craver, levels of mechanisms and levels of realization seem to differ in that the former exhibit a relationship between wholes and parts, whereas the latter exhibit a relationship between wholes and sets of realizers. But this distinction seems rather frail. How are the parts involved in levels of mechanisms different from the set of realizers involved in levels of realization, such as to warrant this distinction? At least in levels of mechanisms, as Craver envisages them, the parts are several mechanisms that together form the whole, which is comprised of all the particular “part-mechanisms”. If mechanisms in general realize certain phenomena, this suggests that all “part-mechanisms” are also realizers in the same way. Now, if the "part-mechanisms" on the "part-levels" are all realizers, it is reasonable to say that the "whole-mechanism" on the "whole-level" is realized by the organized coaction of its parts and that the "part-levels" are the realizers of the "whole-level". Thus, the distinction between levels of mechanisms and levels of realizers conflates.
The difference between levels of mechanisms and levels of emergence, on the other hand, is based on the unpredictability, unexplainability, and metaphysical novelty of higher-level properties, as opposed to lower-level ones. Craver’s point here is that levels of mechanisms, while they can be unpredictable, do not have to be so necessarily, that they are always explanatory, and the novelty of higher-level properties is a trivial fact. But why think that the opposite must hold in the case of emergence? Of course, “spooky emergence”, “[…] the existence of higher-level properties that have no explanation in terms of the parts, activities, and organizational features of the system in the relevant conditions” (Craver this collection, p. 21), is spooky by definition—that much is clear. Also, admittedly, the way emergence was construed historically by the British Emergentists perfectly fits this view and deliberately opposes mechanisms as it can be found, for example, as per Broad (1925). However, why should we prematurely accept this view of emergence as given and eliminate any possibility of further development towards a notion of emergence that is perfectly commensurable with modern science? In fact, I think the formidable way in which Craver develops his account of levels of mechanisms is perfectly suited to facilitate development in this direction. So what could a definition of emergence be, and how can it be united with mechanisms? The following definition of emergence by Evan Thompson (2007) already seems compatible with Craver’s framing of the way properties of higher-level mechanisms are constituted by properties of lower-level ones.
A network, N, of interrelated components exhibits an emergent process, E, with emergent properties, P, if and only if:
(1) E is a global process that instantiates P, and arises from the coupling of N’s components and the nonlinear dynamics, D, of their local interactions.
(2) E and P have a global-to-local (“downward”) determinative influence on the dynamics D of the components of N.
And possibly:
(3) E and P are not exhaustively determined by the intrinsic properties of the components of N, that is, they exhibit “relational holism.” (p. 418)
This definition is compatible with Craver’s characterization of levels of mechanisms in the following respects: properties of higher-level mechanisms, global emergent properties, are realized by properties of lower level mechanisms; and there is a part–whole relationship between those relata, as well as a non-causal influence between the levels. What Thompson's definition additionally contributes is a point about predictability. For many phenomena in nature, the interactions of lower-level components are so complex that they can only be described by non-linear dynamics. A precise predictability of the higher-level phenomena might not always be possible at present due to there being too many factors involved in the underlying processes—it simply exceeds current computational tractability. Craver acknowledges this point—so it can also be said to be consistent with his account—but worries that this might have ontological ramifications: “[i]f that explanatory relationship is severed, then the sense in which emergent properties are at a ‘higher-level’ must be altogether different than the compositional notion of levels in levels of mechanisms” (Craver this collection, p. 21). More precisely, he suspects that in emergence ontological novelty arises through the epistemological limitations just mentioned; otherwise ontological novelty would simply be a banal fact already expressed by his account.
There are, however, several problems with this view: first, one can make a distinction between the epistemological (e.g., “predictability”) and ontological (e.g., “novelty”) dimension of emergence (O'Conner & Wong 2009). There is, in principle, no reason to assume that the ontological dimension is dependent on the epistemological; rather, they seem to be fully dissociable.
Second, his criticism backfires with regard to the banality of the properties that higher-level mechanisms exhibit in his own account of levels. For there to be “higher” levels of mechanisms, these mechanisms must show new properties, that is, in order for them to qualifiedly be on that level. Hence, his account cannot go without ontological novelty of some kind. To now say that this ontological novelty would be only a banal fact undermines his very own striving for mechanistic explanation, which certainly is not banal. In fact, it is still interesting how “higher-level” properties come into existence, what it means to say that they are new, and how the concept “level” might be connected to this. A successful reconciliation of mechanisms and emergence in the form of mechanistic emergence could provide a solution to this problem.
Third, the dissociability of the epistemological and ontological dimension of emergence does not contradict the possibility of their mere coexistence. Once we dismiss the idea that ontological novelty follows from epistemological intractability, overcoming the restraints of historical accounts of emergence, the fact that the coming into existence of new properties on a higher level is not tractable at the moment does not mean that it is not so in principle. The reason why we call this coming-into-existence “emergence”, as might be conceived by a revised account, is not based on the fact that it is epistemically intractable in principle, but rather that it shows novel properties on a higher level that appear to be epistemically intractable in principle, while they might at some point be very well explained in a mechanistic framework combined with a proper theory of property instantiation.
Thompson's definition leaves room for local components to be part of a mechanism. A mechanistic explanation of the emergent phenomenon, it seems, would not be incompatible with an account of emergence, but rather contribute to its explanation by elaborating on how the organization of the parts is essential for emergent properties to arise. As for the ontological novelty of higher-level phenomena, this is certainly a crucial point in emergence: there are new properties coming into existence on the higher levels that are somehow realized by processes of components on the lower level, which in isolation do not show the same properties as the whole. For this to happen, however, contrary to what Thompson's definition implies, the underlying interactions of the components or the emergent properties themselves must not necessarily be unpredictable in principle. But I anticipate an objection: this form of emergence would again only be very weak or banal, but not ontologically new. As already mentioned above, unpredictability does not have a bearing on ontological novelty and is therefore not crucial for emergence. What levels of mechanisms and emergent levels share is that on higher levels there are new properties, which means that there is a notable ontological difference. How extensive such an ontological difference must be in order not to be banal remains a matter of debate. Still, in the case of mechanisms, as well as in the case of emergence, it is very likely that there is a significant phenomenon making up the higher level—otherwise it probably would not be of such interest for enquiry as it clearly is.
As a result of these thoughts, in what follows I will try to reconcile levels of mechanisms and levels of emergence as two interconnected forms of realization. This alone, of course, does not solve the problem of what the levels involved actually are. So in fact there are two problems to be solved for an account of mechanistic emergence: (a.) what it means exactly for a higher-level phenomenon to exhibit new properties, and (b.) what exactly constitutes a level. As a route to a possible solution, in the next section I will sketch a definition of properties that can be implemented in a definition of mechanistic emergence and that at the same time provides a positive account of levels.