Limiting Cases for the New Mechanists: Intralevel Causation is Insufficient for Modeling Complex Biological Systems Abstract Mechanistic accounts of explanation pervade the philosophy of science. Typically, however, such accounts struggle to explain complex biological systems. Craver and Bechtel offer one suggestion along these lines: levels of mechanisms are related by constitution, and causation only occurs intralevel (there is no interlevel causation; ‘bottom-up’ or ‘top-down’). We argue that this suggestion leads to a dilemma when we take seriously the modeling practices involved in studying contemporary complex biological systems. Our suggestion is that the new mechanists should go the route of allowing for interlevel causation, because it better reflects modeling practices and better accommodates non-reductive mechanistic explanations. I.

Mechanistically Mediated Effects and Modeling Practices Within philosophy of science, mechanistic explanations are popular. However, some philosophers have argued that there are limits to the new mechanists’ accounts1. One concern is that mechanistic accounts may not be able to explain complex systems, understood as multilevel and nonlinear systems. Simply put, the new mechanists have struggled to fit dynamic systems or multilevel systems, which often involve feedback loops or nonlinear dynamics, into the current mechanistic account. We focus our discussion on a particular suggestion made by Craver and Bechtel about how to think of multilevel systems in a mechanistic framework. We take their suggestion to be a flagship example for the new mechanists. So, although our criticism focuses on their particular suggestion - mechanistically mediated effects - our concern may be generalizable to other suggestions that are relevantly similar. According to Craver and Bechtel (and other mechanists), what makes for a mechanistic level is constitution; a higher mechanistic level is related to a lower mechanistic level when parts (or ‘entities’) and operations (or ‘activities’) are organized together to constitute a whole.2 A typical example is a car: the engine, frame, doors, wheels, etc. are parts with operations at the lowerlevel that, when organized together, make up the whole car at the higher-level. One of the assumptions that Craver and Bechtel explicitly make about higher-level causes is that the parts of the mechanism taken singularly cannot be affected by or have effects on things that are at the same mechanistic level of the whole. For example, cars don’t cause doors or bumpers to be dented. Rather, it’s the parts that make up the car that can cause such things. This is what Craver and Bechtel refer to as ‘mechanistically mediated effects’ (Craver and Bechtel [2007], 547). Top-down ‘causes’ are really just instances where the whole takes the lower-level ‘along

1

See Machamer, Darden and Craver 2000, Glennan 1996 & 2002, Bechtel and Abrahamsen 2005, and Craver 2007. Such levels are not to be confused with levels of scale –E.coli bacteria and a host do not count as being on different levels merely in virtue of differences in size. 2

Page 1 of 14

for the ride’ in virtue of the constitution relation and the intralevel causes at the lower mechanistic level (Craver and Bechtel [2007], 558-561). However, scientists and philosophers looking at complex biological systems often appeal to topdown or bottom-up causes in their explanations. Craver and Bechtel argue that the notion of topdown causes is incoherent (Craver and Bechtel [2007]). They argue that putative cases of topdown causes (and bottom-up causes) are instances of mechanistically mediated effects. Such effects are said to be “hybrids of causal and constitutive relations, where the causal relations are exclusively intralevel” (Craver and Bechtel [2007], 547). In other words, according to Craver and Bechtel, causal relations occur at a level only, whereas constitutive relations occur between levels. Craver and Bechtel argue that this combination allows for mechanistic explanation without mysterious and problematic interlevel causation (Craver and Bechtel [2007], 551-555). By allowing for causation at a level, in combination with constitutive relations, Craver and Bechtel think that they can provide multilevel explanations; causation occurs at each level, and each level is related via constitution relations (Craver and Bechtel [2007], 561-562). Mechanistically meditated effects supposedly allow the new mechanists’ accounts to provide explanations of complex biological systems. Craver and Bechtel are aware that their account of mechanistically mediated effects does not always reflect how scientists characterize mechanisms, who often make use of descriptions of interlevel causation. To show that their account is nevertheless widely applicable, they suggest a general strategy for translating interlevel causes to mechanistically mediated effects. The basic idea is that putative top-down cases of causation can be cast in terms of causes at the higher-level and a constitution relation that takes parts at the lower-level `along for the ride’ (Craver and Bechtel [2007], 561). The examples that Craver and Bechtel use to demonstrate their strategy are relatively simple and intuitive. The real test for their suggestion, however, is its application to contemporary examples from complex biological sciences, which we turn to now. II.

Evolution of Optimal Levels of E.coli Fimbriation Fimbriae are hair-like structures that grow on bacteria like Escherichia coli (E.coli hereafter), which allows bacteria to cling on to host cells. When E.coli are expressing genes for fimbriation, they have a virulence factor. It turns out, however, that not all genetic variants of E.coli are virulent. In fact, it is well known that some non-disease causing strains (i.e., commensal) of E.coli colonize human gastro-intestinal tracts.3 Although these strains are commensal, they still have some virulence factors, such as fimbriae, but at a very low-level.4 What counts as having a ‘low-level’ in this context is that at any one time only a small proportion of the population is expressing fimbriae (Barnes and Chu [2010], 57). There are a variety of factors that determine whether a specific cell expresses fimbriae. The dominant determinant is the concentration of sialic acid in the host cell. This acid is released by host cells in response to low-levels of fimbriate bacteria. Interestingly, sialic acid is a potential nutrient for E.coli, but only up to a certain point. An increase in the number of bacterial fimbriation levels leads to a further increase in the amount of sialic acid production, but if 3 4

See Escherichia coli 2013 for specific, yet accessible information regarding E. coli. See Perna et al. 2001 and Vidovic et al. 2012 for more information on E.coli virulence.

Page 2 of 14

fimbriation levels go over a certain threshold, the host cells trigger an immune response and the bacterial colony is eliminated (Barnes and Chu [2010], 57). However, ingestion of E.coli bacteria is unavoidable, so if a colony gets eliminated, the host is inevitably re-colonized by new bacteria. Although not every individual bacteria will express fimbriae, all of them are free to take up the sialic acid, a bacterial nutrient, which is released by the host in the environment of the bacteria. However, the rate of switching between fimbriate and afimbriate states can vary between individual bacteria. So the main question biologists are asking is how these rates could have evolved so as to optimize the benefit for all while avoiding the deadly immune response by the host (Shaikh et al. [2007]). It seems that, somehow, E.coli cells have managed to coordinate their expression of fimbriae to get close to optimal levels at the population level. But there is no evidence to suggest that they achieve this by some means of communicating with each other. Nor does it seem possible that their ‘coordination’ could have evolved by means of individual selection, since the fate of a specific bacteria is tied to the fate of the group.5 So, biologists have suggested another mechanism: group selection.6 The features thought to be relevant to such a mechanism are as follows. First, a population of E.coli within a host is relatively genetically homogenous compared to the genetic heterogeneity across different hosts (Lukjancenko et al. [2010]). Second, although populations of E.coli remain relatively stable within a host, the ingestion of these bacteria is unavoidable. Hence, there is some weak interaction between populations of bacteria across different hosts. Third, when a host triggers an immune response that eliminates a population of bacteria, the host is eventually re-colonized by new bacteria. And, since migration between hosts is relatively weak, there is a strong founder-effect, or loss of genetic variation when a new population is established by a very small number of individuals from a larger population. Although variation will enter partly through occasional mutations, and partly through migration between hosts, newly established colonies will tend to be genetically homogeneous (Barnes and Chu [2010], 55-56). The proposed mechanism for how E.coli fimbriation evolved to reach optimal levels is then as follows. Re-colonization of hosts turns out to be the population-level equivalent of reproduction, where newly founded populations will be similar to their “parent” population. And populationlevel mutations occur through migration and mutations within the population. So, the longer it takes for a population to be eliminated, and the larger that population gets in a specific host, the more bacteria from that population will end up, on average, leaving by migration. This means that bigger and longer lasting populations are more likely to be the source for new colonies. This creates competition between groups: one group is fitter if it is less likely to become extinct and has a larger population, which translates into closer-to optimal levels of fimbriation (Barnes and Chu [2010], 72). This verbal explanation inevitably glosses over many details. Although it seems plausible, biologists want to test it by making the reasoning more precise by modeling this complex phenomenon. In particular, given that group selection is somewhat controversial, it would be advantageous to model this without having to talk about groups at all. The difficulty of this task 5

See Feldgarden et al 2003 for an informative discussion regarding selection and E.coli. We recognize that there is some controversy regarding group selection. We are not taking a stand on this debate. Moreover, we don’t think that an account of mechanisms should either. It should be general enough to accommodate whichever way the debate goes. 6

Page 3 of 14

lies in the many irreducibly complex features involved. For one, there are several places where randomness needs to be accounted for, including mutations between parents and offspring, the founder effects mentioned earlier, and migrations between groups. Another complexity is the heterogeneity of switching rates between cells. Since the proposed selection mechanism crucially depends on this, a model cannot simply take the “mean” switching rate. Lastly, there are several interactions that are irreducible, including interactions between the bacteria and the host. For these reasons, some biologists are opting to use agent-based models to develop and simulate the proposed selection mechanism (Barnes and Chu [2010], 56-77). In the interest of space, we will only summarize the details of the agent-based model that are relevant to our discussion. Our main task is to show that, contrary to the suggestion of Craver and Bechtel, the model that biologists have developed makes explicit use of interlevel causes, and that without these the model would not work. To argue that this is a counterexample to Craver and Bechtel, we will show that the interlevel causes of the model resist being translated into mechanistically mediated effects. That is, what is being modeled as an interlevel cause - a top-down cause in particular - by the biologist is not a mere coupling of higher-level causes and constitution relations between the higher-level and lower-level. i.

A Counterexample to Craver and Bechtel

One advantage of agent-based models is that they allow us to explore how one might construct a mechanism without bottom-up or top-down causes. This is done by explicitly specifying three features: agents, environment, and interaction rules both between agents and agents, and agents and the environment. Notice that a population is not explicitly represented, but is constituted by a collection of agents. So if we want to model (causal) patterns that occur at the level of populations, we need to figure out how to do that through interactions that occur between agents and the environment. When designing agent-based models, the process is similar to constructing mechanistic models. Consider, for example, how an agent-based model of the evolution of E.coli fimbriation can be construed as a mechanistic model. Each E.coli cell is regarded as an entity, an agent in the model. Each cell also engages in the activity of fimbriation, which is modeled as a feature of agents that can trigger certain interactions between agents and the environment. Cells may differ in their rate of fimbriation, and this is accounted for by heterogeneity across agents in their fimbriation rates. As for organization, in order for the cells to have the group-level effect that they do, they need to be in the same host – obviously, E.coli cells in John’s gastrointestinal tracts don’t cause Suzie’s gastrointestinal tracts to secrete sialic acid, or produce the immune effects. In the model, hosts are represented as certain regions in the environment and only interact with agents that are in that region. The most interesting part of designing the model comes with the organization of the activities that produce the relevant results at the level of the group. If fimbriation is synced in such a way that all E.coli in a host express it simultaneously, then that could have very different effects than if the cells are coordinated so that roughly the same number of cells are expressing fimbriae at any given time. If the population were large enough, the former would trigger an immune response, while the latter would not. These population level features are not modeled explicitly, but rather are calculated or derived from information about agents in a region of the environment, or host. When the model is simulated, the number of agents with fimbriation turned Page 4 of 14

‘on’ at a given time is counted. If that number turns out to be above a specified threshold, that region of the environment then expels all of the agents. If the number is below the threshold, then the environment produces a certain amount of food in proportion to how close the number is to the threshold. It is important to note that both of these operations are explicitly applied to agents. A population is exterminated because the agents that constitute it are eliminated. So far, the model seems to stay true to the Craver-Bechtel account. However, whether a host produces sialic acid or an immune response is an effect caused by properties of the population of E.coli cells within a given host. Since the population is an organized entity of cells and their activities, it sits at a higher mechanistic level than the individual cells. Hence, acid production and immune responses are effects that occur at the same level as the population level. Herein lies the challenge on the modeling front. Interactions between agents and agents, and agents and the environment are presumably representing causal processes that occur at the lower mechanistic level. Notice however that effects in the environment are caused by a feature of the group – the amount of fimbriation occurring at the population level. As a result, it seems that there is a causal relation between the levels that is mediated by the environment.7 In sum, the chain of constitution and causal relations looks like this: 1. 2. 3. 4. 5. 6.

Each agent is either expressing fimbriae or not. Agents constitute the population. F is the amount of fimbriation occurring at the population level, which is derived from the constitution relation in 2 and information from 1. The value of F causes the environment to produce some effect E. Depending on what E is, each agent either gets some amount of nutrients or is eliminated. Non-eliminated agents then update their state of fimbriation expression (i.e., turns on, stays on, turns off, or stays off).

The problem should be clear. The model needs some way of feeding information from the population level, the higher mechanistic level, to the lower mechanistic level at which the agents belong to. To do this, the model requires the environment to do something at the individual level according to information at the population level. This is a causal interaction, not something that occurs in virtue of a constitution relation. Given how the model was designed, it turns out that the environment is a causal intermediary between the population level and the individual, which violates Craver and Bechtel’s account. Moreover, there is no obvious way to change the model so that the top-down cause is replaced with a relation that is entirely constitutive. We are not saying that this would be impossible to do. Our point is that the model should be understood as a mechanism despite its failing to meet the Craver-Bechtel constraint. So, even if there were some way to recast the top-down cause with some complicated collection of constitution relations, we are claiming that biologists are making good epistemic headway without these further complications.

7

The environment plays various roles in biological modeling, which we will discuss later in the paper.

Page 5 of 14

III.

The Dilemma: Levels Gone Awry or Interlevel Interactions

We now turn to the broader dilemma that we think Craver and Bechtel are put into given the above example and others like it (mentioned below). The dilemma they face is as follows. Either their notion of a level forces us to interpret the models of scientists in a way that confuses both their own notion of level as well as those of the scientist, or they have to take scientific practice more seriously and allow for mechanisms to have interlevel interactions. We discuss each of these in turn. i.

Environment, Levels, and Causal Leakage

Craver and Bechtel’s approach yields at least two unacceptable outcomes. Their account does not allow for the role the environment frequently plays in mechanistic models, which is often a causal intermediary role between levels. Moreover, their account ‘flattens’ the notion of level, which is not reflected in contemporary scientific practice. One of the lessons we learned from the E.coli example is that the environment can play an important role in the characterization of a mechanism. In that example, the environment played the role of an intermediary between levels. But there are also other examples where scientists develop models that make explicit use of interactions between levels. For instance, behavioral scientists interested in mechanisms of learning often require a complex interplay between an organism and its environment in order to explain learned behaviors.8 Models used by systems biologists to depict the interplay between levels of biological systems purposefully set up models that reflect very specific levels and very specific relations between those levels.9 And, cases of circular causation throughout the sciences all demonstrate scientists using models that emphasize the intermediary role the environment plays within the model and that make explicit use of interactions between levels. 10 The majority of the examples Craver and Bechtel discuss don’t seem to emphasize the role the environment plays in scientific modeling, such as serving as a direct intermediary between levels. But the problem we are drawing attention to is just the same when the environment interacts with things at different levels. Craver and Bechtel use just such an example, but apparently fail to recognize its problematic nature. The example comes from the very same paper that we are directing our argument towards. We quote: The eye is a familiar and unproblematic example of a multilevel mechanism. At the highest level, the eye transduces light into a pattern of neural activities in the optic nerve. This process can be decomposed into lower-level components and their activities. The light enters the eye, it is inverted and focused by a lens, and it is projected onto the retina, where the information in the light is converted into a pattern of neural activity in 8

See Harms 1997 for a detailed discussion regarding foraging behavior of bumblebees and how scientists are modeling learning behaviors. 9 See Strand and Oftedal 2009 for information regarding systems biology and the importance of environment as an interlevel cause. 10 See Perovic and Miquel 2011 where they explicitly discuss the importance of interlevel causal reciprocity to for some kinds of explanations.

Page 6 of 14

the optic nerve. The conversion of light into patterns of neural activity by the retina can itself be decomposed into different components: in particular, the rods and cones that change their electrical state depending on the specific features of the light stimulus (such as wavelengths and intensity). Another level down, rod cell activation is also sustained by a mechanism. Light is absorbed by and activates rhodopsin, which then stimulates Gproteins. These G-proteins activate cyclic GMP phosphodiesterase, which catalyzes the conversion of cyclic GMP to 5’-GMP. Lowering the concentration of cyclic GMP causes sodium channels to close, reducing the inward sodium current and thereby hyperpolarizing the cell (see Kandel et al. [1991]). Each new decomposition of a mechanism into its component parts reveals another lower-level mechanism until the mechanism bottoms out in items for which mechanistic decomposition is no longer possible. (Craver and Bechtel [2007], 549) Contra Craver and Bechtel, this is far from an unproblematic example of a multilevel mechanism. Let’s pay a little closer attention. The eye is a mechanism that can be decomposed into parts: the lens, retina, optic nerve, etc. These parts can themselves be decomposed into parts: rods, cones, etc. These can in turn be decomposed into rhodopsin and G-proteins. And so on. What Craver and Bechtel seemingly failed to notice is how light gets put in the picture. By their very own description, they say that light interacts with the eye, that light interacts with the lens, that light interacts with rods and cones, and that light interacts with rhodopsin. So prima facie, it seems that light interacts with things that are at different levels; an environmental feature is an (indirect) intermediary between levels. To defend Craver and Bechtel, one might respond by saying that light interacts with the eye, but that light itself can be decomposed11 into, e.g., wavelengths or photons, and that these decomposed parts of light are what interact with the retina, or rods, or rhodopsin. We think this response can’t be sufficient. The eye has been decomposed into at least four levels, but light has only been decomposed into two: there aren’t enough levels of decomposition of light to go around! So this isn’t just a prima facie problem. It looks like it’s a very serious one. At the very least, it’s serious enough that it can’t do the job Craver and Bechtel want it to do, which is to give us a grasp of what they mean by levels and multilevel mechanisms free of interlevel causation. Craver and Bechtel may be right that the relationship between the higher-level of the retina and the lower-level of the rods is constitutive and not causal, but if something interacts with both of them, you have causal leakage between the levels.12 In other words, we are arguing that Craver and Bechtel have over looked a significant feature of mechanism, the mechanistic environment. Once the importance of the environment is taken into account, we see that often environmental features play the role of interlevel causal intermediary, the very thing that Craver and Bechtel hope to explain away. At this point, Craver and Bechtel either have to deny the importance of environmental features within certain mechanistic models, 11

Please note that here we are being charitable. We do not think that light can be decomposed into wavelengths primarily because light just is at a wavelength. Instead, we utilize this example to show how Craver and Bechtel’s strategy ignores the importance of environmental features and how those features interact within and between levels in a model. Moreover, we utilize their very example to show that deficiency. 12 The new mechanists do allow an entity to be at different levels, but not within the same mechanism. That is, an entity might be at a higher level than another entity for one mechanism, but then be at the same level for a different mechanism. However, the point we are making doesn't require shifts between mechanisms; the mechanism of the eye is fixed, yet light interacts with more than one level.

Page 7 of 14

thereby greatly diminishing the role mechanistic models play in contemporary scientific explanatory practices, or, as we are about to argue, they have to hold that such mechanistic models are grossly ill-conceived by the scientist. Systems biologists and scientists interested in behavioral mechanisms all seem to have well defined levels within their models. For example, when studying complex group hunting behaviors, scientists will often model two distinct levels, namely the group and the individual levels. However, according to the Craver-Bechtel account, the models that all these scientists are working with wouldn’t really be mechanistic. The diagnosis would be something like this: things that interact with one another are at a level, so scientists are wrong about what they think count as different levels (given that their models allow for things at the lower-level to interact with things at the higher-level). Consider the E.coli example. On Craver and Bechtel’s approach, causation is intralevel, so the group and the individual bacteria would be at the same level since there is a causal connection between them. So, if Craver and Bechtel are right, then it is now unclear what work levels are doing for the scientist. The individuals are not at the same constitutive level as the group. Not only do individuals make up the group (which is presumably a constitution relation, and so by Craver and Bechtel’s own lights should render them at different levels), but the whole motivation behind the modeling choices of the scientists is that selection at the individual level won’t work as an explanation. In other words, the Craver-Bechtel account would have us flatten the different posited levels into a single one. But this engenders a peculiar and unpalatable result: the population of the E.coli is at the same level as an individual cell, which in turn is at the same level as the environment (i.e., the host). Similarly and just as surprisingly, systems biologists would really just be interested in one level, if Craver and Bechtel are right. Surely that’s an unacceptable result, even by the new mechanists’ lights. Simply put, if Craver and Bechtel insist that their way is the right way to conceive of levels, then they must claim that lots of models within the biological sciences are either not mechanistic or that they are ill-conceived and ill-constructed. We don’t think this is the route the new mechanists would want to take. In short, something’s going amiss in Craver and Bechtel’s conception of a mechanistic level given the constraint that causation is strictly intralevel. Craver and Bechtel require the flattening of levels against the wishes and apparent practices of scientists. In other words, it doesn’t seem that the notion of ‘levels’ that Craver and Bechtel end up with is anything that scientists utilize or model in practice. Also, Craver and Bechtel seem to have overlooked the importance of the mechanistic environment, in that it very often serves as an intermediary between mechanism levels; the environment is often interacting at multiple levels. We believe that these problems can be remedied by allowing for interlevel causation. ii.

Interlevel Interactions and Epistemic Practice

The other horn of the dilemma is to take the epistemic practices of scientists seriously and thus consider making room for interlevel causation in our conception of mechanisms. This would go against Craver and Bechtel’s initial characterization of mechanism levels, but let us nevertheless suppose that such room was made in a way that maintains the spirit of their suggestion.

Page 8 of 14

Here is how Craver and Bechtel might make room for inter-level causation. One of the advantages of mechanistic accounts of explanation over their traditional rivals13 is that they readily capture the idea that explanations are sometimes partial. Consider the epistemic processes involved in scientific practice. Scientists often use models of mechanisms as epistemic stepping-stones to get to a (more) complete or fuller explanation of the phenomenon they are interested in. That means that, in addition to scientists aiming at target mechanisms (the “true” or “ideal” mechanism), there are also intermediate targets that help scientists acquire knowledge about the phenomenon that lets them continue to make progress14. For example, in the case of E.coli fimbriation, the agent-based model could be improved by figuring out a way to separate the individual and population level so that they don’t interact with one another. So, what Craver and Bechtel might say here is that epistemic progress is made because the model (the epistemically intermediate mechanism with interlevel causation) is becoming more like the target mechanism (which doesn’t have interlevel causation). And, because the model assumes interlevel causation, it can only be a partial explanation. We should say that we are not clear on whether the Craver-Bechtel account is supposed to characterize the target mechanisms that scientists progress towards, or whether it is (also) intended to apply to the mechanisms that are used as epistemic scaffolds. Either way, we think the trade-offs Craver and Bechtel have to make are too costly. We see the cost as follows. If their goal is to describe the intermediate mechanisms that scientists use or construct on their epistemic journey, then they are grossly mischaracterizing this practice. As we have already argued, many scientists seem to be constructing models that utilize and indeed require interlevel causation. We have covered one example in-depth where a model makes use of interlevel causation, the modeling of E.coli fimbriation. We need not look too far into the scientific literature to quickly recognize that large parts of behavioral science, evolutionary science, and systems biology require the explicit use of interlevel causation within their explanatory models. If, on the other hand, their goal is to give an account of mechanisms as targets at the end of the explanatory process, then they are giving up what makes the new mechanistic approach so attractive – a way to make sense of explanations that are partial. Moreover, their account is then also not one of scientific practice and model building, but rather of idealizations and ideal explanations. We do not believe that this fits with the spirit of the new mechanists’ project. We find that these costs aren’t worth Craver and Bechtel’s insistence that mechanistic levels be strictly intralevel. In the final section we discuss one more cost of the Craver-Bechtel account, which ties in with the direction we think the new mechanists should turn to.

IV. Interlevel Interactions: reduction, environment, and epistemic practices

13

We understand scientific laws as being the traditional rivals to mechanistic explanation. Historically, mechanistic explanations were supposed to have several benefits over laws. For instance, mechanistic explanations were supposed to be partial, non-universal, and accurately capture important features of scientific modeling practices. See Bechtel and Abrahamsen 2005. 14 Often, the new mechanists use the notion of mechanism sketch or schema to mean just this (MDC 2000 and Darden 2002).

Page 9 of 14

We find the most appealing way out of the dilemma sketched above is to make room for interlevel interactions. The supposed cost of this, given Craver and Bechtel’s motivation for mechanistically mediated effects, is that we do not retain “a univocal conception of causation as intralevel” (Craver and Bechtel [2007], 554). But we take ourselves to have shown that they were wrong to think that mechanistically mediated effects readily explain “the symmetry of the interlevel relationship and the techniques employed to investigate it” (Craver and Bechtel [2007], 554). So the cost we take on by extending the word ‘causation’ to include interlevel interactions is, if not negligible, worth what we get in return. One important feature we retain is the epistemic practice of scientists like those above. We think that interlevel causation better reflects the modeling practices of scientists. Moreover, it seems to us that as scientists attempt to understand more complex mechanisms, the role of the environment becomes increasingly important. Unfortunately, one of the features we feel has been underappreciated by the new mechanists is the importance of the environment within scientists’ models. This is unfortunate, primarily because often the environment plays either a direct or indirect intermediary role between levels. These are two significant reasons for allowing interlevel causation. We also think that there is an additional independent and important reason for considering interlevel causation. Notice, the new mechanists provide a primarily reductive explanatory strategy; explanations of mechanisms are achieved by decomposing mechanisms into their component parts, where higher-level mechanisms are explained by lower-level components (Machamer, Darden, and Craver [2000], Bechtel [2008]). According to the new mechanists, scientists can provide an explanation for a phenomenon of interest by first creating a model of the mechanism that produces that phenomenon. Then, scientists must either conceptually or physically decompose the mechanism into component parts and activities/interactions/operations that are organized in a certain way as to produce the phenomenon the investigator is interested in (Craver [2007]). It seems as though all of the new mechanists’ views fit within this reductionistic framework. It should be noted, most, with the exception of Bechtel, of the new mechanists deny being reductionists.15 However, there is still reason to believe that the new mechanists are fundamentally reductionistic. We are not claiming that the new mechanists’ approaches are strongly reductionistic, or they assume that there is a single most fundamental level at which entities are “real”, where causation happens, on which explanations rest, and where methodologies should focus. Instead, we think there is reason to believe that the mechanistic approaches are weakly reductionistic. Weak reduction occurs when a higher-level phenomenon is explained in terms of an underlying mechanism (Craver [2007], 161). So, weak reductionism assumes that causation, explanation, and methodology should focus on the parts of wholes but does not insist that these parts be the simplest or smallest entities possible. The MDC, Craver and Bechtel and Abrahamsen accounts can all be considered weakly reductionistic in this sense.16

15 16

See Darden 1991 and 2005, Glennan 2010 and Craver 2007. See Millstein 2013 for a more detailed argument regarding the new mechanists’ reductionism.

Page 10 of 14

Philosophical concerns and arguments against reductive strategies are not new.17 Broadly speaking, antireductionists argue that some lower-level phenomena are not explanatory unless they are placed within context, or within higher-levels (Schaffner [1993]). Here, explanations derive from higher-level phenomena. 18 Ample examples are pulled from areas such as developmental biology, evolutionary biology, and molecular biology. We argue that accounts of scientific explanation, especially ones that aim to capture complex multilevel biological explanation, ought not be merely reductive. As a result, one of the goals of the new mechanists’ project should be to say what it is about mechanistic explanations that make them non-reductive and how multiple levels of explanation combine to form cohesive explanations. The options available to Craver and Bechtel that we outlined above do not help them in this regard. The alternative we are suggesting does. Allowing for interlevel causation in conceptions of mechanisms can seed an answer for why mechanistic explanation is non-reductive. Mechanisms are non-reductive in that complex phenomena require interplay between levels that is not merely constitutive, and in cases like our E.coli example, the higher level plays an important explanatory role in virtue of having a causal connection to the lower level. So, if you have a multi-level mechanism with interlevel interactions, you have a prima facie way to defend how it is non-reductive: the higher-level cannot be completely understood in terms of mechanistically mediated effects of the lower level because interlevel causes are making use of parts at both levels. Interlevel causation can be the very wedge that the new mechanists need to defend the claim that mechanistic explanations are non-reductive. In sum, we believe that the benefits for allowing interlevel causation, as sketched above, far outweigh the benefits of maintaining “a univocal conception of causation as intralevel” (Craver and Bechtel [2007], 554).

17 18

See Fehr 2004 for a great overview of some of the problems with reductionistic strategies, broadly construed. See Brandon 1985 and Sterelny and Griffiths 1999.

Page 11 of 14

Bibliography Barnes, David J. and Dominique Chu. 2010. Introduction to Modeling for Biosciences. London: Springer. Bechtel, William and Adele Abrahamsen. 2005. “Explanation: A Mechanist Alternative”. Studies in History and Philosophy of Biological and Biomedical Sciences 36: 421-441. Bechtel, William. 2008. Mental Mechanisms: Philosophical Perspectives on Cognitive Neuroscience. New York: Routledge. Brandon, Robert. 1985. “Grene on Mechanism and Reduction: More than Just a Side Issue”. Philosophy of Science 2: 345-353. Craver, Carl F. 2007. Explaining the Brain: Mechanisms and the Mosaic Unity of Neuroscience. New York: Oxford University Press. Craver, Carl F. and William Bechtel. 2007. “Top-Down Causation without Top-Down Causes.” Biology and Philosophy 22: 547–563. Darden, Lindley. 1991. Theory Change in Science: Strategies from Mendelian Genetics. Oxford: Oxford University Press. Darden, Lindley. 2002. "Strategies for Discovering Mechanisms: Schema Instantiation, Modular Subassembly, Forward/Backward Chaining." Philosophy of Science (Supplement ) 69: S354-S365. Darden, Lindley. 2005. "Relations among Fields: Mendelian, Cytological and Molecular Mechanisms." Studies in History and Philosophy of Biological and Biomedical Sciences 36: 357-371. "Escherichia coli". 2013. CDC National Center for Emerging and Zoonotic Infectious Diseases. Retrieved 2013-09-18. Fehr, Carla. 2004. “Feminism and Science: Mechanism without Reductionism”. Feminist Formations 16 (1): 136-156. Feldgarden, M., D.E Dykhuizen, and E.V. Sokurenko. 2003. “The Evidence for Selection on fimI gene in E. coli Isolates”. American Society for Microbiology 103: R-016.

Glennan, Stuart S. 1996. "Mechanisms and the Nature of Causation." Erkenntnis 44: 4971.

Page 12 of 14

Glennan, Stuart S. 2002. "Rethinking Mechanistic Explanation." Philosophy of Science 69: S342--S353. Glennan, Stuart S. 2010. "Mechanisms, Causes, and the Layered Model of the World." Philosophy and Phenomenological Research 81 (2): 362-381. Harms, William. 1997. “Reliability and Novelty: Information Gain in Multi-Level Selection Systems”. Erkenntnis 46: 335-363. Lukjancenko, Oksana, Trudy M. Wassenaar, and David W. Ussery. 2010. "Comparison of 61 Sequenced Escherichia coli Genomes". Microbial Ecology 60(4): 708–20. Machamer, Peter, Lindley Darden, and Carl Carver. 2000. "Thinking About Mechanisms." Philosophy of Science 67: 1-25. Millstein, Roberta L. 2013. “Natural Selection and Causal Productivity.” Mechanism and Causality in Biology and Economics, eds. Hsiang-Ke Chao, Szu-Ting Chen, and Roberta L. Millstein. Boston: Springer Press.

Perna, Nicole T., Guy Plunkett, III, Valerie Burland, Bob Mau, Jeremy D. Glasner, Debra J. Rose, George F. Mayhew, Peter S. Evans, Jason Gregor, Heather A. Kirkpatrick, György Pósfai, Jeremiah Hackett, Sara Klink, Adam Boutin, Ying Shao, Leslie Miller, Erik J. Grotbeck, N. Wayne Davis, Alex Lim, Eileen T. Dimalanta, Konstantinos D. Potamousis, Jennifer Apodaca, Thomas S. Anantharaman, Jieyi Lin, Galex Yen, David C. Schwartz, Rodney A. Welch, and Frederick R. Blattner. 2001. “Genome Sequence of Enterohaemorrhagic Escherichia coliO157:H7”. Nature (409): 529-533. Perovic, Slobodan and Paul-Antoine Miquel. 2011. “On Gene’s Action and Reciprocal Causation”. Foundations of Science 16:31-46. Schaffner, K.F. 1993. Discovery and Explanation in Biology and Medicine. Chicago: University of Chicago Press. Shaikh, Nurmohammad, Nicholas J. Holt, James R. Johnson, and Phillip I. Tarr. 2007. “Fim operon variation in the emergence of Enterohemorrhagic Escherichia coli: an evolutionary and functional analysis”. Federation of European Microbiological Societies Microbiology Letters 273(1):58-63. Sterelny, Kim and Paul Griffiths. 1999. Sex and Death: An Introduction to Philosophy of Biology. Chicago: University of Chicago Press. Strand, Anders and Gry Oftedal. 2009. “Functional Stability and Systems Level Causation”. Philosophy of Science 76(5): 809-820. Page 13 of 14

Vidovic, Sinisa, Anil K. Mangalappalli-Illathu, Huiling Xiong, and Daren R. Korber. 2012. “Heat Acclimation and the Role of RpoS in Prolonged Heat Shock of Escherichia coliO157”. Food Microbiology 30(2): 457-464.

Page 14 of 14

POBAM Submission.pdf

Page 1 of 14. Page 1 of 14. Limiting Cases for the New Mechanists: Intralevel Causation is Insufficient for Modeling. Complex Biological Systems. Abstract. Mechanistic accounts of explanation pervade the philosophy of science. Typically, however,. such accounts struggle to explain complex biological systems. Craver and ...

335KB Sizes 0 Downloads 159 Views

Recommend Documents

POBAM paper (02-1.05.14).pdf
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. POBAM paper ...

POBAM 2014 typos fixed.pdf
tools of discovery. 1. Introduction. In recent decades, philosophers of science have incorporated insight from cognitive. science into their analysis of the ...

POBAM Barker & Velasco, Fixing Evolutionary Groups
Fixing Evolutionary Groups. Matthew J. Barker ([email protected]) and Joel D. Velasco ([email protected]). Abstract. We argue for a new conventionalism about many kinds of evolutionary groups, including clades, cohesive units, and popula

Nicholson & Gawne - WPhilBioW (POBAM draft) (1).pdf
... discussed under the heading 'Pre-History of Philosophy of Biology'. (Griffiths 2008). Page 3 of 20. Nicholson & Gawne - WPhilBioW (POBAM draft) (1).pdf.