Representing and explaining: the eikonic conception of explanation Boston University OpenBU http://open.bu.edu Philosophy BU Open Access Articles 2018 Representing and explaining: the eikonic conception of explanation This work was made openly accessible by BU Faculty. Please share how this access benefits you. Your story matters. Version Citation (published version): A Bokulich. 2018. "Representing and Explaining: The Eikonic Conception of Explanation." Philosophy of Science, Volume 85, Issue 5, pp. 793 - 805. https://doi.org/10.1086/699693 https://hdl.handle.net/2144/34297 Boston University http://www.bu.edu/disc/share-your-open-access-story/ 1 Forthcoming in Philosophy of Science Representing and Explaining: The Eikonic Conception of Scientific Explanation Alisa Bokulich Philosophy Department Boston University abokulic@bu.edu Abstract: The ontic conception of explanation, according to which explanations are "full-bodied things in the world," is fundamentally misguided. I argue instead for what I call the eikonic conception, according to which explanations are the product of an epistemic activity involving representations of the phenomena to be explained. What is explained in the first instance is a particular conceptualization of the explanandum phenomenon, contextualized within a given research program or explanatory project. I conclude that this eikonic conception has a number of benefits, including making better sense of scientific practice and allowing for the full range of normative constraints on explanation. 1. Introduction: The vast majority of philosophical work on scientific explanation to date has been focused on accounts of explanation. An account of explanation is a claim about how explanations work: that is, do they work by deducing the explanandum statement from the relevant laws and initial conditions (such as in the covering-law account), or by tracing the relevant causes (as in a causal account), or by identifying the mechanistic components and their activities (as in the mechanistic account)? This paper is not about accounts of explanation; rather, it is about what might be called conceptions of explanation. A conception of explanation is a view about what explanations are. Currently, the orthodox conception of scientific explanation is the ontic conception, according to which explanations are "full-bodied things" (Craver 2014, 40); that is, explanations just are the concrete causes or mechanisms in the world themselves. Versions of the ontic conception have been defended by Wesley Salmon (1989), Carl Craver (2007, 2014), and Michael Strevens (2008). It is only in recent years that the ontic conception has come under scrutiny with significant criticisms being raised (e.g., Wright 2012, 2015; Illari 2013; Sheredos 2016; Bokulich 2016). However, the issue of conceptions of explanation remains so undertheorized that rival alternatives to the ontic conception have yet to be fully articulated and defended. My aim in this paper is to outline and defend an alternative conception of scientific explanation that I call the eikonic conception (from the Greek word 'eikon' meaning representation). The eikonic conception can be understood as falling under the general umbrella of "epistemic" conceptions1, a version of which has been defended 1 As we will see, however, it is a different sense of "epistemic conception" than Salmon (1984, 1989). 2 (though typically only within the context of mechanistic explanations) by what might be called the "San Diego School" of explanation (e.g., Bechtel and Abrahamsen (2005), Bechtel (2008), Wright (2012, 2015), Sheredos (2016)). Like William Bechtel et al.'s version of the epistemic conception, I argue that "[scientific] explanation is fundamentally an epistemic activity performed by scientists" (Bechtel 2008, 18), with explanations themselves being an outcome or product of such an epistemic activity. The eikonic conception emphasizes the importance of representations and representational choices in constructing an explanation. As I will argue, representations are not just involved at the level of the explanans, nor just at the level of the 'explanatory text' (e.g., the particular diagram or equation); rather representations play a prior and more fundamental role in our very conceptualization of the explanandum phenomenon itself. I will motivate the need for this new conception of scientific explanation in three ways: first, by (briefly) reviewing the rival ontic conception and its shortcomings (in Section 2); second, by examining scientific practice, with a couple of case studies (in Section 3); and third, by highlighting the philosophical benefits that come from moving to the eikonic conception of scientific explanation (Section 5). The eikonic conception is outlined in Section 4, paying particular attention to the different ways that an explanandum phenomenon can be represented, depending on the context and aim of the explanatory project. Although I will be rejecting the ontic conception, it is important to note that ontic constraints still play a central role on the eikonic conception. What form those ontic constraints take, however, depends on the particular context of the inquiry (see, e.g., Bokulich 2016). In what follows, I grant that explanation and understanding are "success terms", in that they require getting something right about the way the world is, and more generally, I take the eikonic conception of explanation to be compatible with a broadly realist approach to science. As will become clear, my primary concern here is with scientific practice, that is, with developing a conception of scientific explanation that can best help us understand what scientists are actually doing when they offer scientific explanations. 2. The Ontic Conception An early version of the ontic conception of explanation was proposed by José Alberto Coffa and was elaborated by Wesley Salmon most clearly2 in his book Four Decades of Scientific Explanation where he writes, Coffa is a staunch defender of the ontic conception of scientific explanation. . . . For Coffa, what explains an event is whatever produced it or brought it about. . . . Explanations, in his view, are fully objective and . . . exist whether or not anyone ever discovers or describes them. Explanations are not epistemically relativized, nor . . . do they have psychological components, nor do they have pragmatic dimensions" (Salmon 1989, 133). 2 Salmon also discusses the ontic vs. epistemic conceptions in his earlier (1984) work, but there the discussion is even more confusingly entangled with discussions of determinism vs. indeterminism. 3 Salmon similarly endorses the ontic conception, asserting that "[e]xplanations exist in the world" (Salmon 1989, 86). Salmon contrasts the ontic conception with what he calls the epistemic conception.3 However, Salmon is not using the notion of a conception of explanation in quite the same sense as is being used here. For Salmon, the ontic-epistemic distinction provides a taxonomy for different accounts: causal and mechanistic accounts fall under the ontic conception, while Hempel's covering-law (nomethetic-inferential) account and van Fraassen's erotetic ("why questions") account fall under the epistemic conception. As I am using the term, however, a conception is a claim about what explanations are, independent of the choice of any particular account. So one could adopt either an ontic or epistemic conception of the causal account, or similarly adopt an ontic or epistemic conception of a nomothetic account. This early literature, which fails to clearly distinguish between accounts of explanation and conceptions of explanation is thus not particularly helpful for our present project. More recently the ontic conception has been vocally defended by Carl Craver (2007, 2014), who offers a nice, clear statement of how the ontic conception of explanation should be understood: Conceived ontically . . . the term explanation refers to an objective portion of the causal structure of the world, to the set of factors that produce, underlie, or are otherwise responsible for a phenomenon. Ontic explanations are not texts; they are full-bodied things. They are not true or false. They are not more or less abstract. They are not more or less complete. They consist in all and only the relevant features of the mechanisms in question. There is no question of ontic explanations being "right" or "wrong," or "good" or "bad." They just are. (Craver 2014, 40) The ontic conception has been endorsed by several other scholars working on scientific explanation, such as Michael Strevens (2008). Indeed one could say that the ontic conception has become the orthodox view in the philosophy of scientific explanation today. To summarize, on the ontic conception, explanations just are (in the sense of 'are identical to') the particular tree branch (that broke the window) or the particular electron (that ionized a water molecule). The claim is not just that these things are causes or causally relevant (which they undoubtedly are), but that they are further scientific explanations. On the ontic conception, scientists and scientific theorizing are not actually needed for there to be scientific explanations. Surprisingly, it is only in recent years that the ontic conception has come under scrutiny. A striking problem for ontic theorists is that they have trouble talking consistently about explanation in a way that is true to the ontic conception.4 For example, in Craver's paper "The Ontic Account of Scientific Explanation" (which, on the terminology being urged here, should have been titled "The Ontic Conception of 3 He also contrasts both with the modal conception, which will not be discussed here. 4 For this criticism of Salmon's ontic conception see Wright (2015) and for a criticism of Craver's version see Illari (2013) and Bokulich (2016). 4 Scientific Explanation") he notes that two goals of an adequate theory of explanation are explanatory demarcation and explanatory normativity. If, however, one substitutes in "the causes and mechanisms in the world themselves" for the occurrences of "explanation" in his discussions, as the ontic conception requires, his statements become incoherent: The theory should distinguish explanations [causes] from other forms of scientific achievement. Explanation [Cause] is one among many kinds of scientific success; others include control, description, measurement, and taxonomy. . . . The second goal is explanatory normativity. The theory should illuminate the criteria that distinguish good explanations [causes] from bad”. (Craver 2014, 28). Such statements and projects only make sense if one rejects the ontic conception of explanation and adopts an epistemic or representational conception instead. As Wright (2015) notes it is by slipping back and forth between representational and non- representational notions in unacknowledged ways that the ontic conception has arguably enjoyed an unwarranted success. Phyllis Illari (2013) has similarly urged, In so far as Craver is interested in normative constraints on explanation, presumably he is not interested in constraints on mechanisms—the mechanisms themselves simple are—he is presumably interested in how ontic features . . . constrain explanatory descriptions of mechanisms. (Illari 2013, 243) If Illari is right, then it becomes all the more important to start fleshing out what such an epistemic or representational conception would look like, so that we can begin to address the further question of what form those ontic constraints on our descriptions or representations should take. Some progress on this latter question has recently been made by Sheredos (2016) and Bokulich (2016). As shown next, the motivation for moving to an eikonic conception derives not just from the shortcomings of the ontic conception, but also from the need to better account for scientific practice. 3. Motivating the Eikonic Conception: Scientific Practice When scientists investigate and seek to explain a complex entity or phenomenon in the world, they typically begin with a simplified representation. Different subfields of science will often represent a phenomenon differently, reflecting not only the different interests they may have in that phenomenon, but also the different theoretical resources and methodological tools available within that subfield. This point can be seen particularly clearly in Helen Longino's (2013) discussion of investigations into human aggression, for example. Longino examines five different subfields of science: quantitative behavioral genetics, molecular behavioral genetics, developmental psychology, and social-environmental studies. Each of these subfields conceptualizes and operationalizes human aggression differently. She writes, [I have] surveyed five major approaches or families of approach to studying human behavior. . . . But how behavior is understood as an object of 5 investigation differs: it can be treated as disposition or as episode, as a dimension of variation in a population among populations, or as an individual characteristic. (Longino 2013, 103) She notes that they not only conceptualize the phenomenon of human aggression differently, but also carve up the space of possible causal explanatory factors differently. So for molecular behavioral genetics, what she calls genotype 1 (allele pairs) and genotype 2 (whole genome) are part of the measured causal space for possible explanatory factors, while physiology (e.g., hormones) and anatomy (e.g., brain structure), intrafamily environment, and socio-economic status are unmeasured space. On the other hand, for social-environmental approaches, intrafamily environment and socioeconomic status are part of the measured causal space, while genotype 1 and 2, anatomy, and physiology are unmeasured, and hence not part of the causal space of resources for explanations. The other subfields (e.g., quantitative behavioral genetics and developmental psychology) carve up the measured causal space differently yet again.5 Longino concludes that this plurality of approaches is not in fact in competition, each one offering insights into different aspects of the phenomenon. She writes, [E]ach approach can provide partial understanding of human behavior broadly understood, and . . . each can provide answers to the questions specific to it, but . . . certain features of the investigative space preclude their full integration or the elimination of one or more in favor of a single encompassing and unified approach. (Longino 2013, 125) The plurality of representations of the explanandum phenomenon then is not a weakness of the investigation, but rather a source of strength. Confining ourselves to just one of these approaches would result in understanding less about the nature of human aggression, than we learn through this plurality of representations. One can also find a plurality of representations of an entity or phenomenon within a single field of science. Here the plurality is not due to the different conceptual and methodological resources of different subfields, but rather due to the fact that certain representations are more useful for answering certain kinds of questions about an entity or phenomenon (while other representations lend themselves to answering other sorts of questions). An example of an entity of this sort is water. Given the ubiquity and familiarity of water it may come as a surprise that explaining its properties—many of which are quite anomalous—is an active, indeed fraught area of scientific investigation: Water is the most extensively studied molecule of unique importance to life. Yet our understanding of how this deceptively simple compound of just three atoms gives rise to the many extraordinary properties of its liquid phase is far from complete. The complexity of the water properties combined with multiple possible levels of approximation (e.g., quantum vs. classical, flexible vs. rigid) 5See Longino 2013, pp. 128-129, Figures 2, 3, and 4. 6 has led to the proposal of literally hundreds of theoretical and computational models for water. (Izadi et al. 2014, 3863) No one model of water is able to account for all the experimentally well-known liquid- phase properties, and hence different models, involving different representations of water, have to be used in different contexts (Guillot 2002). Depending on which properties of water scientists want to explain, they will choose a different representation. At the broadest level, there are three main classes of representations of water. First, there are continuum representations, where water is conceived of as a continuous substance with macroscopic properties taken to be well defined down to infinitesimal volume elements. Second, there are what are known as classical atomistic representations of water, which are the familiar “ball and stick” (or ball and spring) representations of water. And third, there are quantum representations of water. A challenge for quantum representations of water, however, is that one cannot exactly solve the Schrödinger equation for even a single water molecule. Within each of these broad classes there are several possible further ways that water can be represented. Take, for example, classical atomistic representations of water, which can be further divided into three classes of models. There are, first, fixed-charge, rigid models, with fixed atom positions (i.e., fixed |OH| bond length and