Inferential explanations in biology

Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

Contents lists available at SciVerse ScienceDirect

Studies in History and Philosophy of Biological and Biomedical Sciences journal homepage: www.elsevier.com/locate/shpsc

Inferential explanations in biology Raoul Gervais, Erik Weber Centre for Logic and Philosophy of Science, Ghent University, Blandijnberg 2, 9000 Ghent, Belgium

a r t i c l e

i n f o

Article history: Received 24 October 2012 Received in revised form 14 June 2013 Available online 15 July 2013 Keywords: Explanation Mechanism Capacity Photoperiodism Pigeon navigation

a b s t r a c t Among philosophers of science, there is now a widespread agreement that the DN model of explanation is poorly equipped to account for explanations in biology. Rather than identifying laws, so the consensus goes, researchers explain biological capacities by constructing a model of the underlying mechanism. We think that the dichotomy between DN explanations and mechanistic explanations is misleading. In this article, we argue that there are cases in which biological capacities are explained without constructing a model of the underlying mechanism. Although these explanations do not conform to Hempel’s DN model (they do not deduce the explanandum from laws of nature), they do invoke more or less stable generalisations. Because they invoke generalisations and have the form of an argument, we call them inferential explanations. We support this claim by considering two examples of explanations of biological capacities: pigeon navigation and photoperiodism. Next, we will argue that these non-mechanistic explanations are crucial to biology in three ways: (i) sometimes, they are the only thing we have (there is no alternative available), (ii) they are heuristically useful, and (iii) they provide genuine understanding and so are interesting in their own right. In the last sections we discuss the relation between types of explanations and types of experiments and situate our views within some relevant debates on explanatory power and explanatory virtues. Ó 2013 Elsevier Ltd. All rights reserved.

When citing this paper, please use the full journal title Studies in History and Philosophy of Biological and Biomedical Sciences

1. Introduction At the starting point of this paper are two uncontroversial characteristics of the explanatory practices of biologist. First, capacities constitute a major type of explanandum in biology. Indeed, the biological capacities that stand in need of explanation are many. How do cells reproduce? How do plants and bacteria convert carbon dioxide into organic compounds? How are genetic traits preserved through generations? How do humans see depth? Second, biological capacities are often explained by means of a mechanistic explanation. A mechanistic explanation of a capacity is an explanation that explains the capacity by means of a description or model of a mechanism, where mechanisms, to use Machamer, Darden and Craver’s often quoted definition, are ‘‘. . . entities and activities organized such that they are productive of regular changes from start or set-up to finish or termination conditions’’ (2000, p. 3).

Because of these characteristics, many philosophers (from now on, we shall collectively refer to them as ‘mechanists’) have rightly stressed the importance of mechanisms and mechanistic explanations in biology (e.g. Baker, 2005; Bechtel, 2006; Bechtel & Abrahamsen, 2005; Darden, 2005, 2006; Glennan, 2002, 2005; Woodward, 2001, 2002). Starting with Machamer, Darden and Craver’s seminal paper ‘Thinking about Mechanisms’ (2000), the literature on mechanistic explanations has seen a considerable growth, not just in the philosophy of biology, but also in other domains that are usually covered with the general term ‘life sciences’.1 Ontologically speaking, we may safely assume that every biological capacity has a mechanism responsible for it. Epistemologically speaking however, our knowledge of these mechanisms is likely to be incomplete to a lesser or greater degree. This situation has led to the idea that explanatory progress in biology consists of filling in more and more details of our model as they become

E-mail addresses: [email protected] (R. Gervais), [email protected] (E. Weber). These other domains include (but are not limited to): neuroscience (Bechtel, 2008; Craver, 2007), cell biology (Bechtel, 2006) and genetics (Darden, 2006). Although the focus of this article is exclusively on biology, it is conceivable that our conclusion holds for (some of) these other domains as well, though this would of course require independent argumentation. 1

1369-8486/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.shpsc.2013.06.003

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

known to us. On this account, it is possible to rank mechanistic explanations on a continuum, ranging from how-possibly models, where the model is still abstract and highly speculative, via howplausibly models, to how-actually models, which constitute a complete and accurate description of the mechanism responsible for a given biological capacity (Craver, 2006). Of course, how-actually models are more a regulative ideal than they are realistically attainable, but the general idea is that as one adds flesh to a skeletal model, its explanatory value increases. However, what about those situations in which biologists know next to nothing about the mechanism responsible for a given capacity? In such a case, in keeping with the ontological assumption mentioned above, all we know is that there must be a mechanism, although its entities, activities and organization may be unknown. As we will show, this is by no means an uncommon situation. The continuum-position described above, plausible as it is, seems to suggest that in such cases, all we can do is wait until enough of these details become known—only then will we have any explanatory hold over the capacity in question. As we will argue below, this does not accurately reflect what is going on in biology. In fact, capacities whose underlying mechanisms are unknown can and do figure in genuine biological explanations. Since these explanations make reference to neither the entities nor the activities of the mechanism, they are non-mechanistic. Hence, it seems that here the conceptual apparatus of mechanistic explanations is not suitable to understand the explanatory practices of biologists. To understand what is going on in these situations, we must focus on the role of generalisations. As we will show below, these generalisations allow us to make certain explanatory inferences. We realize that this is somewhat going against the current of present day philosophy of biology, and we immediately concede that these explanations should not be understood in the strict, Hempelian sense of rigidly deducing the explanandum from a universal law of nature in combination with boundary conditions: they are no DN explanations. As the title of our paper indicates, we label them inferential explanations. We use this label because the explanations we are concerned with use generalisations and have the form of an argument. This is what they have in common with DN explanations. However, there are crucial differences, e.g. the fact that the generalisations can have exceptions.2 Other differences will become clear in the course of this paper. We want to make it clear from the start that we are not against mechanistic explanations. Rather, we defend the complimentary thesis that besides mechanistic explanation, a specific type of nonmechanistic inferential explanations is crucial to biology. In Section 2 we clarify some of the terminology that we will use. Then our argument for the complementarity thesis starts. In Sections 3 and 4 we present two examples of situations in which biological capacities (pigeon navigation and photoperiodism respectively) are explained without constructing a model of the underlying mechanism. We clearly show why these explanations are not mechanistic and also give a positive characterization (i.e. we put forward a substantial constraint on the kind of information the explanations have to include). This positive characterisation is necessary because, as we already said, we want to identify a specific subtype of inferential explanations. In Section 5 we investigate the value of the type of explanations identified in Sections 3 and 4. We will argue that they

357

are sometimes the only thing we have, that they are heuristically useful, and that they provide understanding of contrasts. In Section 6 we discuss the relation between the types of explanations we distinguish and the types of experiments biologists often perform in order to gather evidence for and against explanations. This section connects what we are saying about explanation in biology to experimental practice in biology. In Section 7 we compare our results to Jim Woodward’s views on explanatory power and also clarify how our results relate to the debates on the virtues of abstraction and idealisation. 2. Explaining biological capacities: some terminology By a capacity we mean the ability of a biological system to exhibit some kind of behaviour. The term capacity is chosen for obvious reasons: it might be that for some reason a system does not exhibit this behaviour. Thus, a failure of a system to perform a capacity at a specific time does not count against it having that capacity as such. Some capacities are very common among biological systems (e.g. metabolism, the capacity to reproduce etc.), while some are extremely rare (e.g. the capacity to generate electric fields, found only in a relatively small number of fish). In the next sections of this paper, we shall consider one example of a common capacity (the capacity of plants to flower) and one more rare capacity (the capacity of pigeons to home). Given the centrality of capacities, philosophers of biology have looked around for tools to provide an analytic account of the notion of capacity—one particularly influential account they found was Robert Cummins’ so-called ‘causal role’ (henceforth CR) account of functions.3 Consider Cummins’ definition: CR. X functions as a / in S (or the function of X in S is to /) relative to an analytic account A of S’s capacity to w just in case X is capable of /-ing in S and A appropriately and adequately accounts for S’s capacity to w by, in part, appealing to the capacity of X to / in S (Cummins, 1975, p. 762).4 This definition has played an important role in the debate on the meaning and explanatory role of function ascriptions in biology and engineering sciences (see e.g. Houkes & Vermaas, 2010). The core idea of Cummins is this: a capacity is analysed in terms of two or more distinct and simpler sub-capacities, which in turn can be analysed in terms of yet even more simple sub-capacities. Thus we explain a capacity by detailing how certain sub-capacities contribute to the overall capacity exhibited by the system, that is, to the original explanandum. This explanatory strategy was greeted with considerable enthusiasm by the philosophical community, who saw in it a naturalistic way to account for higher-level capacities by letting the process of division bottom out at a level at which the sub-capacities are so simple they are no longer considered problematic (Cummins, 1980; Dennett, 1987; Machamer, Darden, & Craver, 2000). According to the mechanists however, such an analysis is not enough to understand the explanatory practices in biology and other life-sciences. Here, to explain a capacity, researchers specify the mechanism responsible for it. A Cummins-style analysis limits itself to specifying operations, but, as Machamer et al.’s definition quoted in the introduction already makes plain, according to the mecha-

2 One of the main concerns with the DN model in the context of biology is the status of biological laws. Laws, in the sense of strict, exceptionless, non-contingent regularities, seem (largely) absent from biology (Beatty, 1995, 1997; Brandon, 1997; Sober, 1997). We think of biological laws in a revised sense, along the lines of Mitchell (1997, 2000), according to which they are pragmatic generalisations that allow for prediction, explanation and manipulation, whether they succeed or fail to meet the traditional criteria for lawhood. 3 Others, which we will not consider here, are the dispositional account (Bigelow & Pargetter, 1987) and the etiological account (Mitchell, 2003). 4 Thus, the CR account invokes the notion of ‘function’, which is of course a hazardous one in the context of biology. We should therefore be absolutely clear that in employing the CR account, functions are equivalent to ahistorical capacities, and do not refer to goals or intentions. It is best however to avoid the term altogether and stick to capacities; a policy we shall endorse throughout this paper.

358

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

nists, one must also include information about the parts that perform these operations. They speak of entities and activities, or parts and operations, that have to be specified. Thus, although the mechanists explicitly build upon Cummins’ CR account (Craver, 2001) they supplement it with additional requirements for a model to be of explanatory value: for a model to be of explanatory value, it has to describe the mechanism in terms of its entities, its activities and the way these activities are organized. Bechtel & Abrahamsen formulate this as follows: A mechanism is a structure performing a function in virtue of its component parts, component operations, and their organization. The orchestrated functioning of the mechanism is responsible for one or more phenomena. (2005, p. 423) Craver puts it as follows: [M]echanisms are entities and activities organized such that they exhibit the explanandum phenomenon. (2007, p. 6, italics removed). Let us now briefly summarize our terminology. First we have the following convention: A capacity is a biological system’s ability to exhibit some kind of behaviour. We assume (this is an ontological assumption, not a terminological convention) that in biological systems, such capacities are realized by some underlying mechanism. A mechanism is defined as follows: A mechanism is a collection of entities and activities that are organized such that they realize the capacity. For the use of mechanisms in explanations (the ‘analytic account A’ in Cummins’ definition), we have the following convention: A mechanistic explanation of a biological capacity is a description of the underlying mechanism. All this is very plausible and, again, we do not take issue with the claim that constructing models of the underlying mechanism is a common and successful strategy to explain biological capacities. However, in the next two sections we shall consider two examples of capacities the mechanism of which is largely unknown, yet are still explained by biologists—as we will see, these explanations make use of generalisations.

Two clarifications must be made before we go on. First, the explananda we are dealing with are not particular facts (about individual pigeons) but generalisations (about pigeons as a species). Second, the term ‘‘usually’’ expresses the fact that the generalisations have exceptions: they describe a strong trend in the set of pigeons, but they are not universal laws. For instance, if a breeder feeds alcohol to his/her birds before releasing them, they probably will not return home (at least not immediately).5 With respect to the first explanandum, it was shown that pigeon navigation depends on the position of the sun as a reference point. Thus, the following explanation can be formulated: L1 L2

Pigeons usually have a solar compass. Animals with a solar compass usually have the capacity to find their way back home on sunny days. Therefore: E1 Pigeons usually have the capacity to find their way back home on sunny days.

Like E1, L1 and L2 describe are strong trends in the set of pigeons. With respect to L1, the exceptions are pigeons that did not develop normally or have certain kinds of brain damage. With respect to L2, the exceptions are temporary disturbances, such as the administration of alcohol. The term ‘‘Therefore’’ expresses that E1 can be inferred form L1 and L2, without being too strict. In general, ‘‘therefore’’ can be used to denote a deductive or inductive relation. In this case (and in all examples that follow) it refers to an inductive inference relation. As already mentioned above, pigeons that are released on a clouded day have the capacity to find the way back home, hence E2. In a set of experiments, W. T. Keeton demonstrated that pigeons also have a magnetic compass. Keeton released birds that had magnets and birds with brass rods (which function as placebos) attached to them, both on sunny and on clouded days. The results were clear: on sunny days, the birds were unaffected, but on clouded days the birds carrying magnets became disoriented. In this way, Keeton arrived at the following explanation for E2: L3 L4

Pigeons usually have a magnetic compass. Animals with a magnetic compass usually have the capacity to find the way back home on clouded days. Therefore: E2 Pigeons usually have the capacity to find the way back home on clouded days.

Furthermore, the experiments of Keeton suggest conditions under which the two systems become operative:

3. Pigeon navigation L5 3.1. Keeton’s experiments The first set of examples of inferential explanations in biology we shall consider relates to the capacity of homing pigeons (Columba livia) to navigate. The source for this example is Keeton and Gould (1986, pp. 575–585). A trained pigeon can be taken from home, transported over very long distances (hundreds of miles are not uncommon) and still find the way back to its home after being released. Moreover, they are able to exercise this capacity both in sunny weather and on cloudy days. This gives rise to two explananda: E1 E2

5

Pigeons usually have the capacity to find their way back home on sunny days. Pigeons usually have the capacity to find their way back home on clouded days.

This example was suggested by one of the referees.

Pigeons usually have a solar compass and a backup magnetic compass that works only on cloudy days. L6 Animals with a solar compass and a magnetic compass that works only on cloudy days, usually have the capacity to find the way back to their house on sunny days even if they carry a magnet around their neck. Therefore: E3 Pigeons usually have the capacity to find the way back to their house on sunny days, even if the carry a magnet around their neck.

3.2. INT explanations In 3.1 we presented three examples of explanations of biological capacities. Let us now analyse their properties. First, they are

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

inferential explanations, in that they infer the explanandum generalisation from a set of other generalisations. This means that inferential explanations are at least possible in biology. Second, they posit the existence of a mechanism without describing it. Let us clarify what we mean by this. The claim that pigeons have a solar compass is, in our view, identical in meaning to the following claim: In the body of pigeons there usually are entities (of which we don’t know where they are and what they look like) that have certain unknown activities and are organized in an unknown way. These entities, activities and organization ensure that pigeons usually have the capacity (on sunny days) to determine the angle they have to maintain relative to sun. The claim that pigeons have a magnetic compass is in our view identical in meaning to the following:

359

3.3. A preliminary comparison Our view is that, compared to mechanistic explanations, INT explanations are a different but complementary way of ‘‘telling something a mechanism’’. A mechanistic explanation gives a model of the mechanism in terms of entities, activities and organisation; and INT explanation provides information about the types of inputs the mechanism processes. A mechanist can agree with this but maintain that the information in INT explanations is not interesting. In other words, a mechanist can acknowledge that INT explanations are possible, but deny that they are valuable. In Section 5 we will provide arguments to show that INT explanations are interesting. First, let us look at a second set of examples. 4. Photoperiodism 4.1. Garner and Allard’s experiments

In the body of pigeons there usually are entities (of which we don’t know where they are and what the look like) that have certain unknown activities and are organized in an unknown way. These entities, activities and organization ensure that pigeons usually have the capacity (on clouded days) to determine the angle they have to maintain relative to the magnetic field of the earth. If one agrees that this is the meaning of the generalisations L1 and L3, the explanations are non-mechanistic (because no information is given about the entities, activities or organization). However, from an ontological point of view they presuppose a mechanism: the generalisations cannot be true unless there is a mechanism. The second property gives us a negative characterisation of what is going on (the explanations are not mechanistic), while the first gives us a vague positive characterisations (inferential). The third property is gives us a more precise positive characterisation. L1, L3 and L5 do tell us something about the kind of information which pigeons process, and about the conditions under which they do it: information about their angle relative to the sun and about the magnetic field of the earth. In mechanistic terms: they do not describe the entities, activities or organisation of the mechanism, but they do tell us what types of inputs the mechanism processes. In this way, they do more than just saying that there is a mechanism6: they give information about what the mechanism as a whole does (viz. processing a specific type of input) without telling us anything about what the mechanism looks like. In order to see the importance of this third property, it is useful to consider the following example7: L7: L8:

Pigeons usually have wings. Animals that have wings usually have the capacity to regulate their blood glucose level. Therefore: E4: Pigeons usually have the capacity to regulate their blood glucose level.

L7 does not tell us anything about the kind of inputs the mechanism which regulates blood glucose level processes. Therefore, it differs crucially from the pigeon navigation examples. We propose to call explanations that have the three properties discussed here INT explanations (I for inferential; N for non-mechanistic; T for types of input). In this terminology, the example above is not an INT explanation.

For our second set of examples, we turn to W. W. Garner and H. A. Allard’s explanation of the capacity of plants to flower. Our sources are Keeton and Gould (1986, pp. 395–402) and Murneek (1948). Although the capacity to flower is a crucial part of the reproduction cycle of heterosporous plants, the details of the process vary widely across different species. Some flower in the spring, others in the summer or autumn. Garner and Allard noticed that a new variety of tobacco plant, the Maryland mammoth variety (Nicotiana tabacum), grew to excessive heights in the summer without blooming. However, if they took cuttings of the plant and grew them in the greenhouse during winter, the plant would bloom. Thus, the following explanandum presented itself: E5

Tobacco plants of the Maryland mammoth variety have usually flower in the greenhouse in the winter.

In addressing this explanandum, Garner and Allard conducted a series of experiments, eliminating as many variables as possible, and found that the apparent key determining factor was the number of hours of daylight the plants received. With this information, they could induce flowering in plants during the summer if they shielded the plants from sunlight for part of the day, while in the greenhouse they could inhibit blooming in the winter by exposing them to artificial lights during the night. Based on further research, they distinguished between three groups of plants: short-day plants, which flower if there is a relatively short daily exposure to light (usually less than 12 to 14 hours); long-day plants, which flower if there is a relatively long daily exposure to light (usually more than 12 to 14 hours), and day-neutral plants (plants in which flowering is not influenced by length of daily exposure to light). They dubbed this link between the length of exposure to light and flowering photoperiodism. The first results were published in the Journal of Agricultural Research in 1920, but they published several papers on this topic in the next 25 years. Subsequent research revealed that it is actually not the length of the day, but the length of the night which determines whether plants flower or not. Studies showed that it is not possible to prevent a long-day plant from flowering at the proper season by shielding it from light for an hour during the middle of the day. However, it is possible to prevent a short-day plant to flower in season by exposing it to light for a short interval during the night (minutes or even seconds). Similarly, long-day plants can be

6 The claim that there is a mechanisms is trivial, given our ontological assumption (see Section 1) that every biological capacity has a—known or unknown—mechanism responsible for it. 7 This example was suggested to us by one of the referees.

360

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

caused to flower in the wrong season (i.e. during winter) by means of a light flash that interrupts the dark period. In short, the key factor in the flowering of Maryland mammoth plants is the amount of time the surface of the leaves are not exposed to sunlight. However, the terminology (long-day, short-day, day-neutral) has been preserved, with a different meaning. A short day plant is a species that flowers after dark periods of a minimum duration. A long day plant is a species that requires short dark periods in order to flower. This gives us the ingredients to construct a possible INT explanation for E5: L9 Tobacco plants of the Maryland mammoth are short-day plants with a minimal dark-period-length of 12 hours.8 A1 In the greenhouse during winter the amount of time the plants are not exposed to light exceeds 12 hours. Therefore: E5 Tobacco plants of the Maryland mammoth variety usually flower in the greenhouse in the winter. In this explanation, L9 is a generalisation which tells us something about the information the ‘‘flowering mechanism’’ processes: the length of uninterrupted dark periods. It does not tell us anything about what the ‘‘flowering mechanism’’ looks like or where it is located. A1 is an auxiliary hypothesis. It is easy to see that similar explanations can be given for the behaviour of other short-day plants. This is exactly what Garner and Allard did in the case of the Biloxi soybeans (Glycine max), which they planted throughout the months of May, June and July. Even though these intervals meant that there were considerable differences in the growing periods of the soybeans, they nevertheless began to flower in September en masse. This gives rise to the following INT explanation: L10 A2 Therefore: E6

Biloxi soybeans are short-day plants with a minimal dark-period-length of 10 hours. From September on the amount of time the plants are not exposed to light exceeds 10 hours. Biloxi Soybeans usually flower in September, regardless of the month in which they are planted.

the length of day influenced flower buds only indirectly, through causing the leaves to produce a hormone that induced the buds to flower. But this hormone, which he called florigen, was hypothetical, and as such the mechanism he proposed was at best speculative. Indeed, more than seventy years of subsequent research failed to isolate this hypothesized hormone and describe its chemical structure. This has led some biologists to the conclusion that there is no florigen, or that the term at best refers to a functionally defined concept, the actual filler of which still needs to be identified. To give an example of this latter type of undertaking, one proposal was that flowering would be controlled by the ratio of two or more other hormones. We will briefly come back to this issue in Section 5.2. Whatever the exact mechanism is, the experiments of Chailakhian show that light does not stimulate flowering by acting directly on the buds: there is a stimulus (minimal dark-period-length or minimal light-period-length) that is passed on from the leaves of the plant to the buds. This insight allows gives us some spatial constraints on where the mechanisms is to be found. The relevance of this will be discussed in Section 5. For now it is important that the experiments allow us to give more informative INT explanation than the ones in 4.1. For E5 we now have: L11

Tobacco plants of the Maryland mammoth are short-day plants which start to produce and transport the flower stimulus if its leaves are exposed to a minimal dark period of 12 hours. A3 In the greenhouse during winter the amount of time the leaves of the plants are not exposed to light exceeds 12 hours. Therefore: E5 Tobacco plants of the Maryland mammoth variety usually have the capacity to flower in the greenhouse in the winter.

The crucial difference is ‘‘leaves’’: the input that is processed is minimal dark-period-length for the leaves, not for e.g. the buds or the root or for the plant as a whole. This is more specific than in 4.1. For E6 we have: Biloxi soybeans are short-day plants which start to produce and transport the flower stimulus if its leaves are exposed to a minimal dark period of 10 hours. A4 From September on the amount of time the leaves of the plants are not exposed to light exceeds 10 hours. Therefore: E6 Biloxi Soybeans usually flower in September, regardless of the month in which they are planted.

L10 With the division between the three plant groups in place, one can make similar explanations of the behaviour of other plants, such as chrysanthemum, dahlia and cocklebur (short-day) and beet, clover and larkspur (long-day). 4.2. INT explanations for photoperiodism Let us now take a further step. In 1936 the Russian researcher M. H. Chailakhian removed the leaves from the upper half of chrysanthemums and put paper between the upper and lower half so that one could be exposed to daylight while the other remained shielded. Then he exposed this upper half to long days, and the untreated lower half to short days, which resulted in the plant flowering. The reversed procedure, exposing the lower half to long days and the defoliated upper half to short days, resulted in the plant not flowering. From these experiments, Chailakhian concluded that

5. The value of INT explanations in biology 5.1. Taking stock Let us take stock. Our two case studies support the following claims:

8 It should be kept in mind that the numbers cited in this section are approximate values. In their 1920 article, Garner and Allard themselves drew the somewhat rough distinction between short- and long-day plants that we mentioned above, but in their summary they did note that: ‘‘In a number of species studied it has been found that normally the plant can attain the flowering and fruiting stages only when the length of day falls within certain limits . . .’’ (1920, p. 603). Subsequent research must of course be carried out to give more precise values for each species. Despite the fact that the Garner and Allard’s experiments are well documented, the particular values for the Maryland mammoth and the Biloxi soy bean are not easily obtainable. For the Maryland mammoth, we rely on Foster & Kreitzman (2009); in the case of the Biloxi soy bean, on Garner (1933, p. 349). This gives us minimal dark-period-lengths of 12 and 10 hours respectively. Although these values are approximate, and subsequent research may yield more different/more precise ones, this will make no difference to the philosophical points developed in this section.

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

(1) It is possible to construct INT explanations of biological capacities. (2) Biologists do construct such explanations. Mechanists can agree with both statements yet maintain that these explanations are uninteresting. In this section we present three arguments against such a view: we argue that (i) INT explanations are sometimes the only ones we have, even after many decades of research (5.2), (ii) that they are heuristically useful (5.3) and (iii) that they are interesting in their own right (5.4).

5.3. Heuristic value The heuristic value of inferential explanations that posit but do not describe mechanisms lies in the fact that they suggest new explananda. In the pigeon case, the explanations lead to the following new explananda E6

E7

5.2. Ongoing debate about the mechanisms With respect to photoperiodism, we already mentioned that the mechanism is still subject to debate. Currently, the flowering is thought to result from the combination of photoreceptor proteins like phytochrome and cryptochrome and the circadian clock. Yet the circadian clock as such is no more than a biochemical mechanism, which receives environmental cues as input and has certain behaviour as output. In this highly abstract form, it is postulated to explain the circadian rhythms of plants and fungi as well as animals. The precise details of this clock will vary from species to species, even within the realm of plants. To arrive at a full mechanistic explanation for photoperiodism in the Maryland mammoth and Biloxi soybean then, more work is needed. The same applies to the homing pigeons. Though the experiments of Keeton date from the 1970s (they were published in Scientific American in 1974) it is still impossible to give mechanistic explanations of the capacities. To illustrate this, let us take a closer look at the magnetic compass. A relatively recent proposal is that iron particles (superparamagnetic magnetitie or SPM particles) in the nerve terminals of sensory nerves in the upper beak of the homing pigeons might play a role in the mechanism underlying their capacity to find the way home (Fleissner et al., 2003). The hypothesis is that these particles react to the magnetic field of the earth, so that the nerve cells in the upper beak act as magnetoreceptors, passing on information to the brain, allowing the pigeon to determine its direction, height and location. However, the researchers stress that these SPM particle-clusters in the beak are only a candidate for the responsible magnetoreceptor (Fleissner et al., 2003, p. 360). Moreover, their conclusions have been disputed recently by a group of researchers who argue that the cells in which the SPM articles are found, are in fact not nerve cells at all, but rather specialized white blood cells (macrophages), whose function is to recycle iron particles of red blood cells (Treiber et al., 2012). If that is true, then it is implausible that they play a role as magnetoreceptors, as white blood cells do not possess the ability to convey information to the brain. Moreover, the number of SPM cells varies widely among individual pigeon beaks, which seems at odds with their supposed roles as magnetoreceptors. Treiber et al. conclude that ‘‘. . . our work reveals that the sensory cells that are responsible for trigeminally mediated magnetic sensation in birds remain undiscovered. These enigmatic cells may reside in the olfactory epithelium, a sensory structure that has been implicated in magnetoreception in the rainbow trout’’ (Treiber et al., 2012, p. 369). In short, the message of Treiber et al. is: ‘we don’t know yet where the magnetic compass is located, and since we have failed with the beak, let’s now look at the nasal cavity, because that is where rainbow trout have it’. Interestingly, the next hypothesis that will be considered by this research group apparently results from a comparatively simple instance of analogy reasoning. Borrowing some terminology from Machamer et al., this does not even amount to a ‘mechanism sketch’ (2000, p. 18), let alone a full blown mechanistic explanation.

361

Why do pigeons have the capacity (on sunny days) to determine the angle they have to maintain relative to the sun? Why do pigeons have the capacity (on clouded days) to determine the angle they have to maintain relative to the magnetic field of the earth?

If these questions are answered, the results can be used to build a mechanistic explanation for the original explananda (E1 and E2). In other words: the explanations we have considered here are useful steps towards mechanistic explanations. In effect then, they share this virtue (of suggesting new explananda) with mechanistic explanations. This is also the case in the photoperiodism example, where the explanations in 4.2 lead to the following questions: E8 E9

Why do plants have the capacity to produce a flowering stimulus in their leaves which depends on night length? Why do plants have the capacity to transmit this flowering stimulus from the leaves to the bud?

5.4. Understanding contrasts Finally, we think that the type of inferential explanations we have been considering (i.e. the ones that posit mechanisms without describing them) are intrinsically valuable because they provide understanding of contrasts between species. Let us go back to the pigeons one more time. Consider the following question: E10

Why do pigeons have the capacity to find their way back home while other sedentary birds do not have this capacity?

The explanations discussed in Section 3 suffice to understand this contrast: the other sedentary birds don’t have a solar compass, nor a magnetic one. We do not need the details about how the capacity is implemented in pigeons in order to understand what makes pigeons special compared to other species of resident birds. We know that pigeons are able to process specific types of information, and that is what sets them apart. Another example is this: E11

Why do woodcocks migrate during the night, while pigeons cover long distances during the day?

Here the answer is that woodcocks, like other nocturnal migrants, use star constellations as navigation cue. Again, we do not need to know the details in order to understand the difference between the two species. We know that they process different types of input, and that is sufficient. 6. Types of explanations and types of experiments 6.1. Craver on constitutive relevance and types of experiments The main challenge for scientists who want to give a mechanistic explanation of a capacity is to establish the constitutive rele-

362

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

vance of the entities and activities that are mentioned in the explanans. Craver formulates this problem as follows: Not all parts are components. Consider again the difference between mechanisms and machines. Machines contain many parts that are not in any mechanism. The hubcaps, mud-flaps, and the windshield are all parts of the automobile, but they are not part of the mechanism that makes it run. They are not relevant parts of that mechanism. Good mechanistic explanatory texts describe all of the relevant components and their interactions, and they include none of the irrelevant components and interactions. (2007, p. 140) The crucial question now is: how can we show that an entity X and its activity / indeed are components of the mechanism of the w-ing of an S? Craver’s answer is labelled the mutual manipulability account9: [A] component is relevant to the behavior of a mechanism as a whole when one can wiggle the behavior of the whole by wiggling the behavior of the component and one can wiggle the behavior of the component by wiggling the behavior as a whole. The two are related as part to whole and they are mutually manipulable. More formally:(i) X is part of S; (ii) in the conditions relevant to the request for explanation there is some change to X’s /-ing that changes S’s w-ing; and (iii) in the conditions relevant to the request for explanation there is some change to S’s w-ing that changes X’s /-ing. (2007, p. 153, italics in original) Whether condition (ii) is satisfied can be tested by means of bottom-up experiments. Craver distinguishes two types: interference experiments (inhibitory bottom-up experiments) and stimulation experiments (excitatory bottom up experiments). Here is Craver’s characterization and example of the first subtype: In interference experiments, one intervenes to diminish, disable, or destroy some putative component in a lower-level mechanism and then detects the results of this intervention for the explanandum phenomenon. The assumption is that if X’s /-ing is a component in S’s w-ing, then removing X or preventing it from /-ing should have some effect on S’s ability to w. ... Lesion experiments, for example, are interference experiments in which something intervenes to remove a portion of the brain and one then detects the effects of the lesion on task performance. (2007, p. 147) The second subtype is characterized and illustrated follows: In stimulation experiments, one intervenes to excite or intensify some component in a mechanism and then detects the effects of that intervention on the explanandum phenomenon. The assumption is that if X’s /-ing is a component in S’s w-ing, then one should be able to change or produce S’s w-ing by stimulating X. ... Fritsch and Hitzig performed a series of experiments on dogs in which they delivered low-grade electrical stimuli to a cortical area now known as the motor strip (see Bechtel forthcoming). Localized stimuli along this area produce regular and repeatable movements in specific muscles, including the legs, the tail, and the facial muscles. (2007, p. 149)

9

For an up to date discussion of this account, see Leuridan (2012).

Whether condition (iii) is satisfied can be tested by activation experiments, which are excitatory top-down experiments. Here are Craver’s characterization and example: In activation experiments, one intervenes to activate, trigger, or augment the explanandum phenomenon and then detects the properties or activities of one or more putative components of its mechanism. . . . The basic assumption behind activation experiments is that if X is a component in S’s w-ing, then there should be some difference in X depending on whether S is w-ing or not. ... There are several common varieties of activation experiment at all levels in neuroscience. In PET and fMRI studies, one activates a cognitive system by engaging the experimental subject in some task while monitoring the brain for markers of activity, such as blood flow or changes in oxygenation. (2007, p. 151) The link between the experiments that Craver describes (and for which he uses the general label ‘‘interlevel experiments’’) and mechanistic explanations of capacities can be formulated in two ways: (1) Biologist who want to give mechanistic explanations of capacities must do bottom-up and top-down experiments. (2) The fact that biologists want to give mechanistic explanations of capacities explains why they do perform bottomup and top-down experiments. These are two sides of the same coin. 6.2. Hypothetico-deductive experiments Bottom-up experiments and top-down experiments require that one intervenes on parts that may be components of the mechanism. This is impossible in cases where biologists give INT explanations, because the parts cannot be localized. The information these experiments would yield is also superfluous: we don’t make any claims about parts being components of the relevant mechanism or not; constitutive relevance is not an issue in these explanations. How are inferential explanations that posit but do not describe mechanisms backed up? The experiments we have briefly described in Sections 3 and 4 when we discussed the explanations, are hypothetico-deductive experiments. What they do fits very well the description of such experiments which Carl Hempel gave in chap. 2 of Philosophy of Natural Science (Hempel, 1966) which contains his famous analysis of the Semmelweis case. From the hypothesis that a mechanism of a certain type is present, the scientists derive that a specific causal relation is to be expected. From the alternative hypothesis (the absence of the mechanism) they derive that the causal relation is expected to be absent. These hypothetical derivations have the following general format: If mechanism M is present, then one expects a causal relation between variable C and variable E. If mechanism M is absent, then one expects no causal relation between variable C and variable E. Note that the derivations have causal relations in their consequent. The experiment is performed in order to find out whether or not the causal relation obtains. Let us illustrate this. In the experiments with the magnets, the hypothetical derivations are the following:

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

If pigeons have a magnetic backup compass then on clouded days one expects a causal relation between carrying a magnet around the neck or not (C) and average flight direction (E). If pigeons do not have a magnetic backup compass then on clouded days one expects no causal relation between carrying a magnet around the neck or not (C) and average flight direction (E). The experiments of Keeton are randomized trials which test exactly the causal relation that is at stake here: there is an experimental group with magnets around their neck and a control group with copper rods (which function as placebos). The difference between the average flight direction of the experimental and control group was statistically significant, so his experiments support claim that there is a causal relation between C and E. By means of modus tollens, the hypothesis that pigeons do not have magnetic backup compass is rejected. Like with Craver’s interlevel experiments, the link between the hypothetico-deductive experiments and inferential explanations that posit but do not describe mechanisms can be formulated in two ways: (1) Biologist who want to give INT explanations of capacities must do hypothetico-deductive experiments. (2) The fact that biologists want to give INT explanations of capacities explains why they do perform hypotheticodeductive experiments. Again, these are two sides of the same coin.

7. INT explanations, explanatory power and explanatory virtues We think that what we say here about explanation of capacities can be seen as an extension of Woodward’s view about the explanatory power of singular causal explanations. We go into this issue in Section 7.1. In 7.2 we clarify that what we have said here is of a different nature and goes further than what some people (including we ourselves) have said about abstraction versus richness as explanatory virtue.

363

Woodward confines his theory to explanations of singular events. The ideas we have developed in Section 3 till 5 entail that his view can be extended to explanations of capacities of groups (which are a specific type of generalisations). INT explanations are inferential and have the form of an argument. But their explanatory power does not reside in the fact that the explanandum could be (inductively or deductively) expected, but in the fact that they ‘‘identify a condition such that changes in that condition would lead to some alternative to its explanandum’’ (2003, p. 193). For instance, the explanation which involves the magnetic backup compass tells us that pigeons will lose this capacity if the information about the magnetic field of the earth is distorted (e.g. by putting magnets around their neck as in Keeton’s experiment). 7.2. Abstraction versus richness in models of mechanisms Mechanistic explanations can vary in richness, i.e. in the level of detail the model provides about the mechanisms entities, activities and organisation. In a previous paper (Gervais & Weber, 2013) we have argued that in the cognitive sciences richness (though in itself a desirable property) must be balanced against another explanatory virtue, viz. plausibility. A richer explanation is not always better than a less rich one. Michael Strevens also makes this point in chap. 8 of his 2008 book, where he presents several explanations of Boyle’s law in order to discuss the advantages of abstraction in the explanation of regularities. Several other philosophers have made similar points about weighing virtues of different mechanistic explanations of the same capacity. In order to situate the points we have raised in this paper in the philosophical literature on scientific explanation, it is important to see that we are not talking about more or less rich or complete models. Rather, the idea is that, even if you decide (voluntarily or forced due to lack of knowledge) not to construct a model of a mechanism (and thus automatically go for ‘‘zero richness’’) it is possible to answer some explanation-seeking questions. In this paper, we presented an alternative to mechanistic explanations, rather than discussing desirable properties of mechanistic explanations (although that is an important issue). 8. Conclusion

7.1. Woodward on singular causal explanations In his book Making Things Happen, Jim Woodward discusses a classic example (2003, p. 187): All ravens are black. a is a raven. ––––––––––––––– a is black. He claims that this is not a satisfactory explanation because it . . . . . . doesn’t tell us about the conditions under which raven a would be some other color than black. (p. 193; italics in original) A satisfactory explanation for this case is described as follows: According to the manipulationist account of explanation I am proposing, it is only when one has identified conditions relevant to the manipulation of a raven’s color that one has provided an explanation of why it is black. (p. 193) This condition implies that merely showing that the explanandum was to be expected (as a Hempelian covering law explanation does) is not sufficient for having any explanatory power.

In this paper we have presented two clusters of situations in which biological capacities are explained without constructing a model of the underlying mechanism (pigeon navigation and photeriodism). We showed why these explanations are not mechanistic and also gave a positive characterization: INT explanations. INT explanations are a specific subtype of inferential explanations. They are non-mechanistic (they do not describe the entities, activities or organisation of the mechanism) but they do tell us what types of inputs the mechanism processes. We have argued that INT explanations are valuable for three reasons: they are sometimes the only thing we have, they are heuristically useful, and they provide understanding of contrasts. We realize that the foregoing will surprise some philosophers of biology. The philosophical trend is away from inferential models of explanation, not towards it. Among the mechanists, there is a sentiment that, given its central role in the explanatory practices of biology, the notion of mechanism has received too little attention in 20th century philosophy of science, and moreover that this neglect has something to do with the dominance of Hempel’s DN model of explanation, which in turn is thought to be the result of a focus on physics. Bechtel and Abrahamsen put it succinctly: Given the ubiquity of references to mechanism in biology, and sparseness of reference to laws, it is a curious fact that mechanistic explanation was mostly neglected in the literature of 20th

364

R. Gervais, E. Weber / Studies in History and Philosophy of Biological and Biomedical Sciences 44 (2013) 356–364

century philosophy of science. This was due both to the emphasis placed on physics and to the way in which explanation in physics was construed. (Bechtel & Abrahamsen, 2005, p. 423) In general, the view seems to be that explanations inferring the explanandum from some general statement are of little value. Here are a few quotes that we think are representative of the sort of view we have in mind: In this sense, the covering law model is inaccurate when it states that all science consists of a search for real ‘‘general laws.’’ One can read an entire article in Science on research findings in biology and not encounter anything a scientist would call a general law. (D’Andrade, 1986, p. 22) The received view of scientific explanation in philosophy (the deductive–nomological or D–N model) holds that to explain a phenomenon is to subsume it under a law. However, most actual explanations in the life sciences do not appeal to laws specified in the D–N model. (Bechtel & Abrahamsen, 2005, pp. 421–422) For these three reasons [accidental generalizations, explanatorily irrelevant premises, and the failure of nomic expectability], the CL model of explanation has generally faded from philosophical currency. Also for these three reasons, the CL model is not an especially useful starting place for thinking about the norms of explanation in neuroscience (Craver, 2007, p. 40) Although we concur with the mechanists that up until the turn of the millennium, the notion of mechanism received too little attention, we believe that this neglect has been made up for during the past decade. It is now more than ten years since Machamer, Darden and Craver’s famous paper, and the JSTOR website reveals that, over a 3 year period, it is the most cited paper published in Philosophy of Science (32 citations the last 3 years; http://www.jstor.org/action/ showMostCitedArticles?journalCode=philscie, retrieved 14-062013).10 This indicates that a lot of work has been done on mechanisms and mechanistic explanations. While we agree that mechanistic explanations are important for understanding capacities, we think that care should be taken not to make the same mistake Bechtel and Abrahamsen (justly) point out in the quote above: to focus on one type of explanation at the expense of others. Given the present situation in philosophy of biology, the warning should be that the focus on mechanistic explanations should not lead us to neglect another type of explanation of biological capacities: INT explanations that, although they do not conform to Hempel’s DN model, do infer the explanandum from other general statements. Acknowledgments The research for this paper was supported by the Research Fund Flanders (FWO) through project nr. G.0031.09. References Gervais, R., & Weber, E. (2013). Plausibility versus richness in mechanistic models. Philosophical Psychology, 26, 139–152. Baker, J. M. (2005). Adaptive speciation: The role of natural selection in mechanisms of geographic and non-geographic speciation. Philosophy of Science, 36, 303–326. Beatty, J. (1995). The evolutionary contingency thesis. In G. Wolters & J. Lennox (Eds.), Concepts, theories, and rationality in the biological sciences (pp. 45–81). Pittsburgh: University of Pittsburgh Press.

Beatty, J. (1997). Why do biologists argue like they do? Philosophy of Science, 64, S432–S443 (proceedings). Bechtel, W. (2006). Discovering cell mechanisms: The creation of modern cell biology. Cambridge: Cambridge University Press. Bechtel, W. (2008). Mental mechanisms: Philosophical perspectives on cognitive neuroscience. New York: Routledge. Bechtel, W., & Abrahamsen, A. (2005). Explanation: A mechanist alternative. Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences, 36, 421–441. Bigelow, J., & Pargetter, R. (1987). Functions. Journal of Philosophy, 84, 181–196. Brandon, R. N. (1997). Does biology have laws? The experimental evidence. Philosophy of Science, 64, S444–S457 (proceedings). Craver, C. F. (2001). Role functions, mechanisms, and hierarchy. Philosophy of Science, 68, 53–73. Craver, C. F. (2006). When mechanistic models explain. Synthese, 153, 355–376. Craver, C. F. (2007). Explaining the brain. Oxford: Clarendon Press. Cummins, R. (1975). Functional analysis. Journal of Philosophy, 72, 741–765. Cummins, R. (1980). Functional explanation. In N. Block (Ed.), Readings in the philosophy of psychology (Vol. 1, pp. 185–190). Cambridge, MA: Harvard University Press. D’Andrade, R. (1986). Three scientific world views and the covering law model. In D. W. Fiske & R. A. Shweder (Eds.), Metatheory in social science (pp. 19–41). Chicago: University of Chicago Press. Darden, L. (2005). Relations among fields: Mendelian, cytological and molecular mechanisms. Philosophy of Science, 36, 349–371. Darden, L. (2006). Reasoning in biological discoveries: Essays on mechanisms, interfield relations, and anomaly resolution. Cambridge: Cambridge University Press. Dennett, D. C. (1987). True believers. In The intentional stance (pp. 13–35). Cambridge, MA: MIT Press. Fleissner, G., Holtkamp-Rötzler, E., Hanzlik, M., Winklhofer, M., Fleissner, G., Petersen, N., et al. (2003). Ultrastructural analysis of a putative magneto receptor in the beak of homing pigeons. Journal of Comparative Neurology, 458, 350–360. Foster, R. G., & Kreitzman, L. (2009). Seasons of life: The biological rhythms that enable living things to thrive and survive. New Haven, CT: Yale University Press. Garner, W. W. (1933). Comparative responses of long-day and short-day plants to relative length of day and night. Plant Physiology, 8, 347–356. Garner, W. W., & Allard, H. A. (1920). Effect of the relative length of day and night and other factors of the environment of growth and reproduction in plants. Journal of Agricultural Research, 18, 553–606. Glennan, S. (2002). Rethinking mechanistic explanation. Philosophy of Science, 69(Suppl.), S342–S353. Glennan, S. (2005). Modeling mechanisms. Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences, 36, 443–464. Hempel, C. F. (1966). Philosophy of natural science. Englewood Cliffs: Prentice-Hall. Houkes, W., & Vermaas, P. E. (2010). Technical functions. On the use and design of artefacts. Dordrecht: Springer. Keeton, W., & Gould, J. (1986). Biological science (4th ed.). New York and London: W. W. Norton & Co.. Leuridan, B. (2012). Three problems for the mutual manipulability account of constitutive relevance in mechanisms. British for the Journal Philosophy of Science, 63, 399–427. Machamer, P. K., Darden, L., & Craver, C. F. (2000). Thinking about mechanisms. Philosophy of Science, 67, 1–25. Mitchell, S. D. (1997). Pragmatic laws. Philosophy of Science, 64, S468–S479 (proceedings). Mitchell, S. D. (2000). Dimensions of scientific law. Philosophy of Science, 67, 242–256. Mitchell, S. D. (2003). Biological complexity and integrative pluralism. Cambridge: Cambridge University Press. Murneek, A. E. (1948). History of research in photoperiodism. In A. E. Murneek & R. O. Whyte (Eds.), Vernilazation and photoperiodism. A symposium (pp. 39–61). Waltham, MA: Chronica Botanica Company. Sober, E. (1997). Two outbreaks of lawlessness in recent philosophy of biology. Philosophy of Science, 64, S458–S467 (proceedings). Strevens, M. (2008). Depth. An account of scientific explanation. Cambridge: Harvard University Press. Treiber, C. D., Salzer, M. C., Riegler, J., Edelman, N., Sugar, C., Breuss, M., et al. (2012). Clusters of iron-rich cells in the upper beak of pigeons are macrophages not magnetosensitive neurons. Nature, 484, 367–370. Woodward, J. (2001). Law and explanation in biology: Invariance is the kind of stability that matters. Philosophy of Science, 68, 1–20. Woodward, J. (2002). What is a mechanism? A counterfactual account. Philosophy of Science, 69(Suppl.), S366–S377. Woodward, J. (2003). Making things happen. New York: Oxford University Press.

10 The second best, Larry Laudan’s ‘A Confutation of Convergent Realism’ has only 20 citations; Kitcher’s ‘Explanatory Unification’ has 13, Hempel & Oppenheim’s ‘Studies in the Logic of Explanation’ 8.