3.1. The Argument from Context

An important consequence of how the mechanistic framework defines the criteria of being at the same level and being at different levels (see sec. 2) is that levels of mechanisms are local: they are well defined only within a given compositional hierarchy. If two entities are not working parts of the same mechanism, then there is no meaningful way to address the question whether they are at the same level (Craver Reference Craver2007, 192; Bechtel Reference Bechtel2008, 147). Because of this locality, lower levels are never causally closed, since lower levels are never extended beyond the set of entities that together constitute a mechanism, and thus they are always restricted and partial. So within the mechanistic framework there are no comprehensive lower levels that are causally complete and closed (Bechtel Reference Bechtel2008, 148). Local levels restricted to the constituents of a mechanism typically lack resources to account for effects exerted from outside the mechanism, that is, from the context in which the higher-level whole is embedded (Bechtel Reference Bechtel, Schouten and Looren de Jong2007, 183; Reference Bechtel2008, 152). As a consequence, such lower-level effects will have only higher-level causes.

3.2. The Argument from Organization

The spatial arrangement of the entities forming a mechanism and the temporal organization of their activities are crucial determinants of a mechanism. However, just like resources necessary to account for contextual effects, resources required to account for arrangement and organization are also unavailable at the lower level of the parts. The spatial and temporal structure of the parts is independent of their behavior: parts conforming to the same laws and producing identical behavioral repertoire in isolation can nevertheless be organized into very different structures (Bechtel Reference Bechtel, Schouten and Looren de Jong2007, 183). This organization is imposed on the parts by the higher-level whole (Bechtel Reference Bechtel2008, 150; Reference Bechtel2009, 554–57) and thus is a manifestation of the unique causal powers of higher levels.

3.3. The Argument from Constraints

The spatial and temporal organization of the parts constrains the behavior of the parts: modes of organization imposed on the constituents restrict how they can interact with each other (Bechtel Reference Bechtel2009, 555–57). Similarly, the functioning of the mechanism as a whole in its higher-level context impinges specific conditions on the mechanism that, at least partly, dictate how the parts can operate (Bechtel Reference Bechtel2008, 240; Reference Bechtel2009, 557–59). So by imposing a structure on the parts and by interacting with its environment, the higher-level whole constrains the behavior of the parts—which is an extra causal influence on what happens inside the mechanism exerted by higher-level entities and thus a further source of the causal autonomy of higher levels (Bechtel Reference Bechtel2017a, 271).

4.1. Identifying Higher- and Lower-Level Causal Roles

It is a fundamental tenet of the mechanistic framework that the organized activity of the constituent parts of a mechanism produces the very behavior that characterizes the target phenomenon since this is the requirement that ensures the success of explaining a higher-level phenomenon mechanistically in terms of the behavior and organization of certain lower-level entities. If the behavior produced by the organized activity of the lower-level entities was not identical to the characteristic behavior of the target phenomenon, then the mechanistic story would clearly not be an account of the target phenomenon (Fazekas and Kertész Reference Fazekas and Kertész2011; Soom Reference Soom2012; Rosenberg Reference Rosenberg2018). That is, the mechanistic framework is inherently committed to identifying the causal roles a whole plays at the higher level with the causal roles the organized activity of the parts plays at the lower level. So, even if causal processes at different levels look different, they are not different—in fact, the corresponding ones must be identical (Bechtel Reference Bechtel2008, Reference Bechtel2009; Fazekas and Kertész Reference Fazekas and Kertész2011).

Note that this picture is not eliminativist, for higher-level entities are token identical with sets of lower-level ones. It is not epiphenomenalist either, for higher-level entities do possess causal powers—the same causal powers that are possessed by certain spatially and temporally structured sets of lower-level entities (see Fazekas and Kertész Reference Fazekas and Kertész2011; Soom Reference Soom2012). This latter point needs special emphasis, given that Bechtel interprets his opponents as if they were arguing for the unjustified extreme view, which is “highly reductionist,” “represents all activity at one lowest level” (Bechtel Reference Bechtel2017a, 269), and renders “higher levels epiphenomenal” (262; for such a view, see Rosenberg [Reference Rosenberg2018]). Contrary to this, the picture advocated here is compatible with the usefulness of higher-level inquiries (as the only sources of higher-level descriptions) and the importance and significance of higher-level accounts (as the only sources of certain generalizations, as say, in the case of multiple realizability); that is, it is compatible with the epistemic or explanatory autonomy of higher levels. What it is incompatible with is the view Bechtel wants to argue for the causal autonomy of higher levels.

4.2. Answering the Argument from Context

Another commitment explicitly endorsed by proponents of the mechanistic framework is that causation is an intralevel phenomenon: causal links do not span between different levels, they are confined to single levels. As Craver and Bechtel put it: “there are no causal interactions beyond those at a level” (Reference Craver and Bechtel2007, 561; see also Craver Reference Craver2007; Bechtel Reference Bechtel2008, Reference Bechtel2017a; Fazekas and Kertész Reference Fazekas and Kertész2011). Note, however, that this commitment to causation as an exclusively intralevel phenomenon is in tension with the claim that levels are local. Because of locality, even if it is possible to analyze two entities that are interacting, causally connected parts of a mechanism in terms of lower-level submechanisms, it is not possible to decide whether the lower-level parts constituting the submechanism responsible for the behavior of one of the entities are at the same level as the lower-level parts constituting the other submechanism (Craver Reference Craver2007).

The problem with this consequence of locality is as follows: (i) in accordance with the logic of mechanistic explanations, the organized activity of the parts constituting the submechanisms produces the very behaviors that are characteristic of the entities analyzed; (ii) if an interaction between the two entities is part of this behavior, then there will be an interaction between the two submechanisms as well; (iii) and if causality is strictly an intralevel phenomenon, then the causal connection between the two submechanisms entails that the two submechanisms are at the same level (Fazekas and Kertész Reference Fazekas and Kertész2011; Eronen Reference Eronen2015). That is, lower levels can be extended beyond the original scope of mechanistic inquiry. Consequently, lower levels are not necessarily restricted and incomplete, and thus there is no principled reason to think that they are causally not closed. In particular, note that the higher-level context that plays a crucial role in the ‘argument from context’ is the higher-level environment of the entities targeted by mechanistic decomposition. This environment consists of those higher-level entities that interact with the entity under scrutiny. Since lower levels can be extended exactly along the lower-level counterparts of such higher-level connections, contrary to the claim of the ‘argument from context’, resources are very much available at the lower level to account for the effects exerted by the context of the mechanism (Fazekas and Kertész Reference Fazekas and Kertész2011; Soom Reference Soom2012).

4.3. Answering the Argument from Organization

Bechtel explicitly argues that how the components are spatially, temporally, and relationally organized goes beyond the account of the parts and their operations. However, a lower-level account is not restricted to the (intrinsic) characteristics of the parts and the behavioral repertoire they produce in isolation. Thinking so would simply misidentify the supervenience base of the higher-level entity under scrutiny. The organization of the lower-level parts crucially determines what kind of behavior the overall mechanism produces. So a lower-level account must include information about the spatial and temporal structure of the parts and their interactions. In fact, the organization of parts can only be accounted for at the lower level, since it is the lower-level methodology that can target lower-level entities and thus can uncover their organization, and it is the lower-level terminology that is apt for describing the parts and their structure. Note that these are criteria that Bechtel himself proposes as ways to characterize higher and lower levels: he claims that at higher and lower levels different experimental and explanatory strategies and different vocabularies are needed (Bechtel Reference Bechtel, Schouten and Looren de Jong2007, 185; Reference Bechtel2008, 155–57). That is, since spatial, temporal, and relational facts of lower-level entities are discovered by lower-level inquiries, and are characterized by lower-level vocabularies, by Bechtel’s own standards they are very much integral parts of lower-level accounts (Fazekas and Kertész Reference Fazekas and Kertész2011).

5.1. A Case Study: The Circadian Clock Mechanism

Bechtel’s own case study to support the argument from constraints is about the intracellular molecular ‘clock mechanism’ that is responsible for circadian rhythmicity—cyclic oscillations of daily behaviors and physiological functions. The molecular clock mechanism partly consists of a transcriptional-translational feedback loop involving so-called clock genes Per and Cry and their protein products PER and CRY. Inside the nucleus, transcription factors BMAL1 and CLOCK drive the transcription of the genes Per and Cry to RNAs, which then are decoded by ribosomes to produce proteins PER and CRY. After a translocation back to the nucleus, PER and CRY inhibit the transcriptional effects of BMAL1 and CLOCK through direct protein-protein interactions. Via this negative feedback loop, PER and CRY autoregulate their own transcription, which results in a periodic increase and decrease in their concentration and defines the phases of the oscillation of the clock mechanism. During the night, PER and CRY concentration is high, while it is low during the day such that it peaks in the middle of the night after which it decreases and bottoms out in the middle of the day. Correspondingly, the level of Per and Cry transcription is low during the night, increases at dawn as PER and CRY concentration falls, reaches its maximum during the day, and then decreases at dusk (the presentation here is simplified; for full details, see Rosenwasser and Turek [Reference Rosenwasser, Turek, Kryger, Roth and Dement2017]). That is, the transcription of Per and Cry, which is a component of the mechanism, depends on the phase of the oscillation and thus is sensitive to the actual state of the mechanism as a whole (Bechtel Reference Bechtel2017a, 257). The constraining effects of the higher-level whole on the behavior of the components can most clearly be seen, Bechtel argues, in the different effects of light exposure on Per transcription at different times of the day: while light input has no effect during daytime, at dusk it delays, whereas at dawn it advances the phase of the oscillation (267).

Bechtel’s claim here is that characteristics of the lower-level mechanism, like the behavior of certain parts (the level of transcription of Per and Cry), and how the organized activity of the parts processes a given input (exposure to light) are determined by the actual state of the higher-level whole (the phase of the oscillation). Recall that Bechtel uses this example to support his argument from constraints against the causal closure of the lower level and for the causal autonomy of the higher level (see, e.g., Bechtel Reference Bechtel2017a, 272). So the claim that he really needs is that the way in which the higher-level whole determines the unfolding of lower-level events is via a causal influence and that no lower-level entity exerts the same causal influence. Does Bechtel’s own example support this claim?

5.2. Constraining Effects Are Exerted by Lower-Level Entities

Bechtel crucially relies on the phase of the oscillation of the circadian clock mechanism as the determinant of how lower-level processes unfold. The phase of the oscillation, however, is not a characteristic of some higher-level whole that has no counterpart at the lower level. On the contrary, the molecular description that characterizes the clock mechanism can and does define the phases in purely lower-level (molecular) terms: as the periodic changes in the concentrations of PER and CRY (see above, also Rosenwasser and Turek Reference Rosenwasser, Turek, Kryger, Roth and Dement2017, 352–55). So the phase of the oscillation is not some higher-level factor that causes changes in the level of transcription of Per and Cry in a way such that the relevant causal power is not possessed by any lower-level entity. It is the changes in the concentrations of PER and CRY that cause changes in the level of transcription of Per and Cry—via the direct protein-protein interactions with BMAL1 and CLOCK inhibiting their drive of the transcription of Per and Cry. Similarly, it is not the phase of the oscillation, as a higher-level entity per se, that constrains how the parts of the mechanism change their behavior as a response to a light signal but the actual level of concentrations of PER and CRY (see Bechtel Reference Bechtel2017a, 267–68).

Interestingly, Bechtel is in agreement with us with regard to the presence and importance of these lower-level interactions. He acknowledges that “the phase of the oscillator at a time just is the concentrations of PER, CRY” (Bechtel Reference Bechtel2017a, 257, our emphasis). And he also acknowledges that “as a result of the interconnectivity of the parts, especially the feedback loops, the … mechanism functions as a unit, with the operations of the individual parts of the mechanism determined by other parts of the mechanism” (268, our emphasis). Nevertheless, he thinks that these interactions are compatible with the causal autonomy of higher levels (272). But they are not. These interactions, by ensuring that the lower-level effects in question have lower-level causes, causally close the lower level with regard to the behaviors under scrutiny and leave no room for the causal autonomy of the higher level.

Per, Cry, PER, CRY, BMAL1, CLOCK, and so on, are interacting working parts of the clock mechanism, and as such—by the standards of Bechtel’s own definition that puts working parts of a mechanism at a lower level than the whole, the behavior of which is produced by the mechanism (Bechtel Reference Bechtel2008; see also Craver Reference Craver2007)—they are lower-level entities. Furthermore, they, their interactions, their spatial arrangement, and the temporal organization of their activities are all studied by lower-level methodologies and characterized by lower-level vocabularies (see sec. 4.3). Is there contextual information that cannot be captured at this level? No: the relevant contextual information concerns the effect of light on the phase of the clock mechanisms. But this is also defined in terms of the lower level, as the boosting effect of light on Per and Cry transcription. What is left? Nothing. All information is there at the lower level, and all causal influences can be exerted by lower-level entities. The higher level possesses no unique causal powers that could not be attributed to some lower-level factor. Therefore, the higher level is not causally autonomous.

Bechtel is right that feedback loops are important. But they are important as parts of lower-level interactions. Having an account of how feedback loops actually work reveals how lower-level processes impose complex constraints on their own unfolding, and thus, instead of disproving, it contributes to appreciating the causal closedness of the lower level.