1 Scientific Ontology from the Ground Up

It is both an honor to receive the welcome critique of this volume and, having digested its contents, a somewhat intimidating task to reply. Much if not all of the commentary to which I will respond in this paper arose during the V Workshop del Círculo de Buenos Aires and in subsequent discussions, maturing into fully-fledged analyses of a broad sweep of thinking across the epistemology and metaphysics of science. Needless to say, I am grateful to have this opportunity to engage with the resulting challenges and to defend the ideas that gave rise to them here.

To set the stage before digging in more systematically, let me note that I have always thought of the philosophical clash between proponents of scientific realism and antirealism—a dispute about how best to interpret what the sciences are telling us about world—as unfolding against the backdrop of a metaphilosophical clash regarding certain (epistemological and other) presuppositions of these views. These two spheres of interest are clearly intertwoven, but they are most commonly discussed separately, and this policy is followed by all of my interlocutors in what follows. In keeping with the spirit of their arguments, I will divide my responses in two: in Sect. 2, I will focus primarily on concerns about what might be described as ‘ground-level’ debates about scientific realism and ontology and, more specifically, my own proposal for how best to formulate scientific realism, which I call ‘semirealism’. In Sect. 3, I will focus primarily on what might be described as ‘metalevel’ debates about realism, antirealism and ontology more generally. Having said this, given the interwovenness of these domains and an irresistible tendency I have to think about both simultaneously, I will reserve the right to import the occasional thought about one domain into discussions of the other where this seems helpful to the possibility of dissolving a pressing concern—at least, I will not resist the attempt!

Following this recipe, in Sect. 2, after briefly reviewing the basic motivation for and the central tenets of semirealism, I will consider a number of concerns about whether either is compelling. This encompasses both epistemological and metaphysical issues: the former concerning the notion that semirealism integrates the most convincing aspects of entity realism and structural realism, yielding a regulative principle for identifying what precisely is worthy of realist commitment; and the latter concerning how this unifies concepts common to many articulations of scientific realism, such as causation, laws of nature and natural kinds. In Sect. 3, after briefly reviewing my contention that all interpretations of scientific knowledge have metaphysical dimensions, given a plausible conception of what this could mean, I consider several objections. This too encompasses issues of both epistemology and metaphysics, not least because it is a central part of my view that these issues are, in fact, inextricable. Here I will be concerned with whether all forms of antirealism in particular have plausibly metaphysical dimensions, and whether the ‘Pyrrhonian’ conclusions I draw from this analysis—regarding the shared, rational acceptability of different stances underlying at least some forms of realism and antirealism—is tenable.

2 Semirealism and Dispositional Realism

To begin, let us consider semirealism as a proposal for scientific realism. Roughly speaking, realism in this arena entails a positive epistemic attitude toward the content of our best theories and models, interpreted as furnishing true or approximately true descriptions of all manner of mind-independent features of the world. In part in response to the seriousness of natural (and sometimes explicitly antirealist) concerns about the fulsome extent of this understanding of science, it seemed to me that the most sophisticated accounts of realism appearing in recent decades had become increasingly measured and selective about the proper subjects of realism, in the sense that they endorsed only specific components of what might otherwise be identified as the broader content of our best science taken at face value. Most persuasive along these lines, I believed, were entity realist approaches, which advocate for belief in the existence of entities under certain conditions of causal knowledge and intervention, and structural realist approaches, which advocate for belief in the existence of certain structural relations as opposed to putative entities thereby related. Furthermore, it seemed to me that the most persuasive insights of these different approaches were, surface appearances to the contrary, compatible with one another.

Thus, the task of spelling out this nascent semirealist thought essentially became one of explaining the nature of this imagined compatibility. The key, I held, is a particular understanding of the properties of things described by our best science. For reasons similar to those animating entity realism, the most convincingly warranted aspects of theories and models, as revealed by scientific practice, correspond to the existence of well-detected properties. At the same time, an independently motivated understanding of the nature of such properties, as being intimately connected to relational aspects of the world prized by structural realists, served as a bridge. On this view it is natural to think of these properties as having a modal nature—I adopted the term ‘dispositional’—which then builds the bridge. And with this much in place, a fuller elaboration of all of what seem the most heavily utilized concepts in explications of realist approaches to the sciences (as mentioned above: laws, kinds, etc.) followed, or so I would contend, in a highly unified manner. Semirealism thus took shape as a proposal for integrating the epistemological dimensions of scientific realism, regarding what we learn from science, with the metaphysical dimensions of a plausible account of this, stemming from what we may learn on philosophical reflection.Footnote 1

In light of all the skeptical worries that confront the idea of scientific realism in the many, perennial debates surrounding it, I believe the position just sketched or something close by amounts to a maximally defensible form of it. Quite apart from whether this is true, however, I hope the preceding snapshot is sufficient to suggest a number of avenues of possible concern, to which I will turn now.

2.1 The Unification of Entity Realism and Structural Realism

In their paper ‘On the Explanatory Power of Dispositional Realism’, Gentile and Lucero (2024) argue that the ‘synthesis’ of entity realism and structural realism offered by semirealism is problematic. If this charge were to stick, given what I have just said about the union of insights associated with these prior views forming the kernel of the semirealist proposal, from which all else follows, that would be very bad news indeed (for the semirealist). Clearly, on their face, entity realism and structural realism seem contradictory: the former is a realism about certain entities and the latter is a denial of this (at least, as the term ‘entities’ is often conceived, as denoting individuals of some sort); the latter is a realism about certain structures and the former is a denial of this (at least, as the term ‘structure’ is often conceived, as denoting aspects of reality of a different sort than the existence of entities per se). Gentile and Lucero appreciate that the semirealist claim here is not about the compatibility of entity realism and structural realism as such, but rather the compatibility of the ‘best’ or ‘core’ insights of these positions. Thus clarified, is the union proposed still problematic? The answer, of course, depends on which (putative) insights of these positions are counted as best, or as belonging to their respective cores.

Let me return to the notion of what’s ‘best’ momentarily, as this may seem a tendentious starting point: one might worry that what I identify as the best insights of entity realism and structural realism are so identified merely because they are compatible with semirealism. On the contrary, I believe there is independent motivation for certain aspects of these views, but will set this contention to one side until we have considered their ‘core’ commitments. Does entity realism, properly construed, entail a denial of claims that go beyond mere entity-existence claims, including claims about at least some relational structures of the sort advocated by structural realism? It is difficult to see how it could, since a knowledge of certain relations of entities (e.g., relations to our means of detection) is generally required in order to establish their existence. Thus, while falling well short of endorsing realism with respect to theories and models in toto, entity realists must be amenable to realism regarding at least some structural aspects: specifically, certain causal structures involving entities.Footnote 2 This is instructive, I think, in illustrating the complexities of formulating versions of selective realism. ‘Entities, not theories’ is fine as a slogan, but defensibly elaborated, entities and certain highly selective aspects of theory must come together for the sake of entity realism.

The question of whether structural realism excises a knowledge of entities is more vexed. Granted, I do say at one point (in passages discussed by Gentile and Lucero) that “[a]t the core of this picture is the idea that objects […] are ontologically otiose” (Chakravartty 2007, 71). But note: this is a description of one very particular sort of structuralism, ontic structural realism, and even this requires qualification, for it is a description of one very particular sort of ontic structural realism (eliminative OSR; e.g., French 2014)! ‘Structural realism’ is a label for a genus of views, not all of which preclude a knowledge of entities. Eliminative positions rule out a knowledge of entities in virtue of eliminating entities as an ontological category entirely. Others, however, are non-eliminative, permitting a knowledge of entities construed in an appropriately structural way; for example, in terms of objects whose identities are wholly determined by their places in sets of structural relations, or which have no irreducibly intrinsic properties or natures. Again, slogans may deceive. ‘Structures, not entities’ is striking, but more generically, (ontic) structural realism may be elaborated as ‘any form of structural realism […] that inflates the ontological priority of structure and relations’ (Ladyman 2023). Since many forms of structural realism allow for some form of knowledge of entities, a denial of the latter cannot be a core commitment of structural realism simpliciter.

Having seen, then, that the ‘core’ of entity realism does not preclude at least some knowledge of structural aspects of the world, and that the ‘core’ of structural realism does not preclude at least some knowledge of entities, the door now seems open to some rapprochement, however partial. Semirealism is an attempt to embody this idea in a conception of scientific realism as, in the first instance, a realism about well-detected properties. Groups of properties (or property instances) that tend to cohere are commonly described as entities, and the properties themselves are arguably, plausibly conceived as having a structural dimension: what makes them the properties they are is a matter of what relations entities having them enter into, or are thereby capable of entering into. This exemplifies what I take to be the most compelling insight of entity realism, namely, that our abilities to interact with and manipulate entities produce the strongest empirical evidence we have for their existence. It also incorporates what I take to be the most compelling insight of structural realism, namely, that the relational structures involved in such investigations are central to them. There is much more to say and defend here, but perhaps enough has already been said to show that there is nothing incoherent in seeking to unify the best insights of entity realism and structural realism.

2.2 The Minimal Interpretation of Structure: a Priori?

One motivation for many versions of selective realism, including my own, is to answer skepticism stemming from the historical fact of theory change over time. Scientific realism proposes a positive assessment of the general content of established science, however qualified in various ways, and thus it is only natural to consider whether (or to what extent) this assessment may be undermined by changes in scientific theorizing over time—hence the now immense literature on the so-called ‘pessimistic induction’ on the history of science. Selectivity yields a response if it identifies aspects of theories and models that have comparatively greater epistemic warrant, and are thus more likely to survive theory change, while other aspects, for all we know, may be more susceptible to more substantial change. A further skeptical challenge arises, however, for any such proposal: how can one be sure that whatever aspects one identifies in past theories as having had sufficient warrant for belief are not merely so identified on the basis of perceived similarities to our own best theories today? The worry is one of rationalization post hoc: the selective realist’s case is cloaked in a mantle of differential epistemic warrant, but underneath, narratives of historical continuities may simply be constructed so as to suggest a progression toward the sciences of today; and aspects that appear to have survived into the present are simply represented as having had greater epistemic warrant, supporting realism, in the past.

Semirealism, I contend, is not susceptible to this worry, because it furnishes an explicit formula for identifying aspects of theories that have greater epistemic warrant in the present. I call this ‘the minimal interpretation of structure’. The basic idea, in keeping with the previous section, is that those properties and relations that are minimally required to interpret the formalism of a theory as it pertains to detection, intervention, and manipulation are those for whose existence we have greatest evidential support. Of course, these commitments can be read into the past—we detect negative charge today, just as we detected it a century ago—but the important point is that the epistemic principle of grounding belief in robust causal knowledge is something that can be defensibly projected just as well into the future. According to semirealism, it is these aspects of scientific theories and models that are minimally required for successful scientific practice and, consequently, most likely to be retained.

Gentile and Lucero see something worryingly a priori in this forward-looking feature of semirealism: ‘the a priori criterion makes a precarious distinction, since we run the risk of leaving aside some properties that really have epistemic credentials, a fact that could be revealed in the future, when the theory evolves towards its successor’. Now, surely there is a sense in which the distinction between aspects of theories that have superior epistemic warrant (denoting what I call ‘detection properties’) and those that do not (denoting ‘auxiliary properties’) is precarious: in the development of science, some of what is now auxiliary may become detectable. This is nicely illustrated in Gentile and Lucero’s example of Jean Perrin’s stunning experimental work before and after the turn of the last century, which convinced many of the existence of atoms. It is difficult, though, to see any concern for realism in this, let alone semirealism. As evidence grows stronger, so does warrant for belief. From a semirealist perspective, it is precisely because Perrin illuminated certain detection properties that realism about these properties and their relations to our detectors became more amenable to realism. It is thus difficult to understand Gentile and Lucero’s worry that ‘only a retrospective view allows us to know what properties, which were initially classified as auxiliary, became detection properties, due to refined instruments of detection and new interventions’. Perrin’s experiments lent support to realism at the time of Perrin, not merely in retrospect.

I suspect there is some confusion here about what the concern about post hoc rationalization is, exactly. Its most serious implication is that one is never in a position to interpret the content of science as realists do, because such beliefs are always hostage to what one will learn later, in the fullness of time, about what has survived in some successor science—at which point realism in connection with earlier science may be rationalized post hoc, not earlier—but this would be to make realism tenable only in the ideal limit of inquiry and not before. It is precisely this worry that is defused on a semirealist view of epistemic warrant. Once scientific inquiry allows for the identification of detection properties, realism is justified in the present, and we have good reason to think that this justification will carry into the future. It is thus not the case, as per the concern, that realism is always a matter of rationalization post hoc, held hostage to a future that is always in front of us.

2.3 The Dispensability of Natural Kinds to Scientific Realism

Having discussed certain epistemological dimensions of semirealism, let us turn now to some metaphysical issues, beginning with what seems an appealing unification of concepts commonly employed by scientific realists in articulating their positions. Earlier I mentioned the idea that semirealism, which is first and foremost a realism about well-detected properties, successfully unifies concepts of causation, laws of nature and natural kinds. While sympathetic to this regarding our concepts of properties and laws, Gentile and Lucero hold that to extend the analysis to kinds is to go too far: (1) ‘natural kinds […] have rather a derivative reality, if any’; and (2) ‘we do not need to postulate the existence of natural kinds, since the laws are sufficient to facilitate inductive and predictive practices’. (1) and (2) appear to represent rather different concerns—the first metaphysical, and the second epistemological—so let me consider each in turn.

What sort of ‘reality’ do natural kinds have, if any? The question of whether nature comes pre-packaged, as it were, into categories of things awaiting discovery, has a pedigree dating back thousands of years. In the passages targeted here (Chakravartty 2007, Ch. 6), I take pains to suggest that the very idea is not only dubious but perhaps even inextricably mired in ancient and medieval theorizing that is difficult to square with modern scientific practices of classification. Ultimately, given all of this, my intended focus is not so much natural kinds as ‘scientific kinds’, and the resulting view is deflationary about kinds as an ontological category as such.Footnote 3 Nothing about this, however, is in tension with the idea that ‘kindhood’ is a concept that appears in many influential articulations of scientific realism, and that semirealism is well equipped to make sense of this talk of kinds. To put the point another way: whether or not kinds have an ontologically derivative status is strictly irrelevant to the fact that the notion of kindhood is central to much realist discourse. I believe that we can make sense of this discourse in terms of the spatiotemporal coherence of certain groups of properties and their relations (the latter often described as laws). This is the basis of the proposed, semirealist unification of these concepts.

This leaves us free to consider the further worry expressed in (2), above, that kinds are dispensable not merely ontologically, but epistemologically. I use the metaphor of ‘sociability’ to describe the fact that certain groups of properties cohere, sometimes very strictly (underwriting historical conceptions of kinds as having essences), and sometimes more loosely (underwriting more recent conceptions of kinds associated with looser clusters of properties). While it is arguably the case that any kinds thus conceived may be ontologically decomposed into groups of properties (which, one might argue further, are demarcated conventionally), the concept of kinds is not so easily dismissed from the perspective of inferential practice. Not all law-like generalizations, for instance, are described in terms of relations between properties (e.g., Coulomb’s law, F = k (q1⋅q2)/r2, with variables representing properties); some are described in terms of relations between kinds of entities (e.g., ‘planets in solar systems have elliptical orbits’). In the latter case, while the relevant kinds may be ontologically derivative of certain properties, they are also derivative of sociable groupings of properties, and the latter are epistemologically significant, for they reflect not merely the relations of individual properties, but facts about relations between cohering distributions of properties in spacetime.

It is often facts about property coherence, not merely properties themselves, that fuel inductive success in the sciences. Consider cases in which we successfully extrapolate predictions regarding the behavior of one member of a kind based on observations of the behaviors of other members. This sort of inferential practice extends well beyond cases in basic physics and chemistry, where the underlying relations of properties responsible for such behaviors are well understood, to cases involving more complex systems, where specific arrangements of property instances are crucial to producing particular behaviors and we have little or no knowledge of how these underlying properties and their arrangements give rise to the behaviors at issue (let alone in law-like ways), as in studies of phenomena such as animal behavior. Contra Gentile and Lucero, then, the notion of sociability in the analysis of kind concepts is not merely a postulation; nor is it empty or dispensable. It is a metaphor for the empirical discovery that some sets of properties tend to cohere, which is undeniably significant from the point of view of scientific epistemology.

2.4 Dispositions, Models and Contrastive What-Questions

Arguably, the modal notion of properties I adopt (see the introduction to Sect. 2) in unifying commonly cited features of scientific realist conceptions of knowledge affords epistemic goods beyond unification. For example, as Spehrs (2024) notes in ‘Dispositional Realism, Conflicting Models and Contrastive Explanation’, a dispositional view of properties provides the realist with a promising way to understand the common scientific practice whereby different and incompatible models are used to predict and explain phenomena pertaining to one and the same target system in the world. The mutual inconsistency of such models poses a prima facie challenge to realist interpretations of them, but a dispositional understanding of scientific properties reveals such inconsistences to be merely apparent, or so I suggest (Chakravartty 2017, Ch. 6).Footnote 4 The basic idea is that when different models appear to characterize a target system incompatibly—for instance, treating fluids as continuous media in studies of flow, and treating them as collections of discrete particles in studies of diffusion—the descriptive content of these models need not be interpreted as attributing contradictory properties (e.g., continuity; discreteness), but rather dispositions to behave in different ways (e.g., like a continuous medium; like a collection of discrete particles) in connection with different phenomena.

This way of thinking about targets of inquiry amounts to what I describe as a sort of pluralism about behavior—scientific properties are sometimes associated with dispositions to behave in rather different ways—and to elaborate this, I draw an analogy to established thinking about contrastive explanations. Often, there are different but nonetheless correct answers to a question of the form ‘Why x?’. Correct answers may include, for example, descriptions of different factors that are causally relevant to the production of x. An appropriate answer in any given context of inquiry is one that furnishes the information sought, which may be illuminated by posing the question contrastively: ‘Why x, rather than y?’. ‘Why did Spehrs critique Chakravartty’s view of explanation rather than his view of kinds?’ and ‘Why did Spehrs critique Chakravartty’s view of explanation rather than have a nap?’ suggest the appropriateness of different sorts of answers. By analogy, I propose that one may think of the use of incompatible models in terms of (ontological) questions of the form ‘What is x?’. Different contexts of inquiry specified by studies of different behaviors of target systems may call for different but nonetheless correct answers in terms of the dispositional properties of those systems.

Spehrs worries that identifying a contrast class of potential explananda pointing to appropriate dispositional answers to what-questions courts some sort of circularity. What this circularity is and why it is worrisome is unclear to me, but it appears to stem from the thought that modeling the behavior of a target system entails that one already knows that the system is disposed to behave that way. This may indicate, I suggest, not a worry about circularity so much as one about explanatory triviality or emptiness in the manner of Molière’s physician, who explained how opium causes drowsiness by attributing to it a ‘dormitive virtue’. But this is not a fitting template for ontological what-questions and dispositional answers. Imagine that one is interested in flow-type behaviors of fluids and engages in scientific inquiry into these phenomena. This effort is not, of course, successfully concluded by noting that fluids are disposed to flow. On the contrary, one learns through a process of integrated theorizing and experiment that certain flow-type behaviors can be successfully modeled as processes in a continuous medium. This is neither trivial nor empty, nor does it suggest, I believe, that ‘the way in which the relevant contrast space of contexts could be identified […] is rather puzzling’, since the relevant contexts are simply comprised of scientific investigations into how best to model the various system behaviors we are concerned to study.

This confusion regarding what, precisely, would count as a serious answer to a contrastive what-question also emerges, I think, in Spehrs’ (separate but connectable) concerns that (1) all of this seems well suited to a pragmatist, not a realist understanding of mutually incompatible models, and that (2) ‘an index of behaviors’ is not sufficient to individuate target systems, given the possibility of multiply realized behaviors. On the first point, no doubt, a pragmatic attitude toward incompatible models is possible; but in that case, there would be no need to defuse their incompatibility, merely apparent or otherwise! The realism of my view is exemplified in taking attributions of dispositions—even behavioral dispositions described analogically, as in the example of fluids, above—as correctly describing (in part) the nature of something in the world. It is true for both pragmatists and realists that earth-satellite systems are Newtonian systems to certain degrees of approximation, and that we commonly employ models that meet contextual needs for accuracy or tolerances for error (e.g., Newtonian models may suffice for purposes of placing satellites into orbit), but their interpretations of the content of these models differ substantially. Both will agree that finer-grained and more accurate descriptions are possible, but only realists (of the sort I have described) will explain this in terms of more detailed and accurate descriptions of underlying dispositions.

This brings us to the second point (2) above, about whether behavioral descriptions suffice, and here I am inclined to resist Spehrs’ concern that ‘a dispositional consolidation of incompatible descriptions does not always tell us about the nature of a target system, but only how it behaves or is disposed to behave’. After all, ‘not always’ is compatible with ‘sometimes’ and even ‘often’, and distinguishing the nature of something from its behaviors too strongly threatens to introduce a false dichotomy. The way something behaves may be part of its nature, even if more detailed descriptions of properties underlying behavior are revealed by further inquiry. And the fact that the sciences are often driven to reveal such details, and that realists interpret this (where warranted by evidence) as giving fuller descriptions of natures, further distinguishes their position from pragmatism. Furthermore, the fact that some behaviors are multiply realized is not a compelling basis for a general skeptical argument for doubting our ability to distinguish target systems. If our epistemic situation were such that we are often confronted with qualitatively different objects of inquiry that behave in all the same ways—and, worse still, share all further, underlying properties that science may reveal—then yes, skepticism would rule. In that case, however, an account of how realists may understand what incompatible models are telling us about the world would be the least of our concerns.

2.5 Semirealism With or Without Realism About Dispositions

Let me conclude this first half of my reply, about semirealism as a proposal for scientific realism, with an issue that gives pause to both Gentile and Lucero, and Spehrs: my appeal to dispositions in describing the modal nature of properties of scientific interest. Gentile and Lucero note, entirely correctly, that I conceive the project of unifying and integrating our concepts of properties, causation, laws, and kinds in such a way as to produce a defensible, metaphysical underpinning for realism, as being open, ultimately, to different sensibilities regarding metaphysics, including at least some varieties of empiricism. How could this be, given the antipathy of many empiricists for de re modality and metaphysically substantial accounts of causation and laws, let alone dispositions? Let me intensify this incredulity before dissolving it. Spehrs notes, also correctly, that the dispositional realism I adopt is not ultimately essential to my argument—that one might instead adopt a metaphysically deflationary account of dispositional ascription, for example, by appeal to a semantic reduction of dispositional terms to conditionals that forgo all dispositional vocabulary. But surely all of this generates a dilemma. Either I am proposing a form of scientific realism, in which case it is not acceptable to empiricists; or I am proposing something acceptable to empiricists, in which case it is not a form of scientific realism.

Before facing this dilemma head-on, let me affirm for the record that I do not myself endorse conditional analyses of dispositional ascription (indeed, I do not believe that any are successful), or attempts to recast talk of causation, laws, etc. in wholly deflationary ways—but I do say that anyone who believes that such analyses and deflations are cogent may help themselves to semirealism (cf. Chakravartty 2007, 42, 112 and especially 124–125). A clarification of this reveals the underspecified and thereby misleading nature of both horns of the supposed dilemma above. Regarding the first horn: semirealism, as a form of scientific realism, clearly cannot be acceptable to empiricists who are scientific antirealists, harboring no pretense to knowledge of (for instance) strictly unobservable entities and processes. It is nonetheless open to empiricists who are not scientific antirealists, but who endorse empiricist explications of certain concepts typically employed in articulations of scientific realism (causation, laws, etc.). Regarding the second horn of the dilemma: since not all empiricists are scientific antirealists even if they prefer to render certain concepts I discuss in other ways, once again, I leave the door open. It is at least possible that semirealism, which is at heart a realism about well-detected properties, may be enticing to at least some of them, mutatis mutandis.

Perhaps this will invite further protest. How can a philosophical position be open to such variability in the interpretation of its central concepts? To answer this question well requires a discussion of what I call ‘epistemic stances’, more than one of which, I believe, is compatible with scientific realism, and whose consideration I defer to the following sections. For now, let me simply note that there is a reason I called the book on which the commentators in this section have focused primarily A Metaphysics for Scientific Realism—‘A’, not ‘The’. Here I can do no better than to recall (from Chakravartty 2013, 57) that

when it comes to elaborating the core concepts underpinning scientific realism, I believe that one must and should admit a degree of voluntarism. The challenge for the realist, in the face of no small amount of skepticism, is to show that the relevant concepts can be given any sort of consistent, coherent, and unified explication. I believe I have done so, and invite others to do the same in accordance with other possible explanatory priorities, and thus make semirealism their own.

3 Epistemic Stances Underlying Ontology

Debates about scientific realism and antirealism are embedded in a more expansive philosophical context of disputes about realism and antirealism more generally. One way of characterizing the latter very generally would be in terms of debates about the very nature of epistemic warrant, which would include disputes about scientific realism as special cases. My own interest is in this more specific domain of warrant for belief in relation to the sciences, but even this has a broader scope than debates about scientific realism as they are commonly construed, as pertaining to ‘explicit’ subject matters of scientific inquiry—explicitly targeted phenomena—such as chemical processes and economic consequences of high inflation. My domain of interest extends further (as intimated in Sect. 2.5) to what one might call ‘implicit’ subject matters of science, such as the nature of properties, causal relations, and laws, which are mentioned abundantly in scientific discourse in passing, but whose natures are not targeted in scientific practice as subject matters of explicit focus. Disputes about this implicit content have recently fallen under a number of headings, whose boundaries and degrees of co-extensiveness are unclear: naturalized metaphysics; scientific metaphysics; inductive metaphysics; the metaphysics of science. In part due to this unclarity, it is not especially obvious where debates about scientific realism leave off and debates about naturalized metaphysics (etc.) begin.

My own take, on which the commentators in this section focus, begins with the contention that interpreting the outputs of scientific practice in such a way as to yield warranted beliefs about the world requires the application of some or other philosophical understanding of how empirical evidence is properly assessed; some aspects of this assessment may be invisible in scientific practice but illuminated by the philosophy of science. There are, I argue, different, conflicting, and nonetheless rational approaches one might take toward such assessment (note the reference to voluntarism at the end of Sect. 2.5), and these different approaches, commonly identified with epistemologies of science that are inclined toward deflationary, or empiricist, or more metaphysical interpretations of the outputs of science, differ regarding how and where they identify what counts as genuinely scientific ontology. I associate these approaches with underlying epistemic stances: collections of attitudes, values, aims, and policies concerning the assessment of evidence, promoting belief or agnosticism with respect to the content of science. Perhaps most controversial here is the idea that since various agents come to different but nonetheless rational conclusions about the justified scope of scientific ontology, debates between them are ultimately irresolvable in principle.Footnote 5

My use of the term ‘scientific ontology’ in relation to the foregoing is intended to specify, as mentioned above, a more general, in some ways metaphilosophical arena of concern in which more specific debates about scientific realism and antirealism are embedded. This broader sphere is home to further and no less contentious discussions; for example, about whether metaphysics is as pervasive as I contend it is, and whether different stances are in fact rational and ultimately indefeasible. Let us turn now to some challenges on these and related fronts.

3.1 Metaphysical Presuppositions of Empirical Science

‘Metaphysics’ means different things to different people. For some it is simply a branch of philosophy engaged with topics that were, for largely contingent, historical reasons, lumped together. For others it is a form of philosophical inquiry distinguishable from others by a distinctive methodology. One might add to this list, but to understand what is at stake in discussions of scientific ontology, something like the second conception just noted is most relevant, even if it is fanciful to hope for much detail or precision in specifying the relevant methodology. The most common elucidation of it is simply to note that metaphysical inquiry is largely a priori: metaphysical arguments proceed substantially via reflection on non-empirical considerations, placing significant weight on intuition and conceptual analysis; metaphysical claims are ones that follow from such reasoning, or that are otherwise difficult to assess in any other way. One of the key ideas for which I argue here is that all approaches to scientific ontology that hold that we have some knowledge of the world, however qualified—this excludes only thoroughgoing skepticism and disinterest—involves some metaphysics, whether in the form of metaphysical presuppositions or metaphysical inferences used in drawing ontological conclusions.

In his paper ‘Can Science Escape Metaphysics?’, Gaeta (2024) contests the assertion that metaphysics inevitably plays a role in scientific ontology, whether presuppositionally or inferentially. Let me examine his remarks about the former here and return to the latter momentarily (in Sect. 3.2, and more extensively later). The idea that a priori presuppositions are part of scientific practice has been widely proposed, from the ‘constitutive’ and ‘functional’ principles of logical empiricism (Friedman 2001; Stump 2003), to the metaphysical elements of paradigms (Kuhn 1970), to much more besides (see Chakravartty 2017, 72). Gaeta, however, insists that even in cases where commitments playing crucial roles in scientific practice are not especially sensitive to empirical evaluation initially, they are nevertheless subject to telling empirical inquiry in the fullness of time. He illustrates this in connection with a case I consider: the early 20th-century conflict between Robert Millikan and Felix Ehrenhaft regarding their experiments to investigate the question of whether there is a fundamental unit of electric charge. Glossing over the nuances of disputed historical narratives, a plausible case can be made that Millikan’s assumption that charge is quantized, and Ehrenhaft’s that it is continuous, significantly influenced their interpretations of their data, leading to their positive and negative answers, respectively, to the overarching question.

Of course, this was just the beginning for electric charge. In the wake of these and other experiments and over time, the notion that charge is quantized and admits of a fundamental unit was broadly accepted on the basis of considerable, additional evidence. Gaeta quotes Richard Feynman’s discussion of how ‘biases’ influencing earlier work were later hashed out, noting that ‘in contrast to Chakravartty’s reading of those episodes, [Feynman] welcomes the fact that such prejudices can be overcome by performing more careful experiments’. Now, it is an alarming thing to be charged with hostility toward overcoming prejudice or, for that matter, careful experiments (as it happens, I am a huge fan of both), but all of this is beside the point. As I observe in the passages discussed, in some cases it is reasonable to wonder whether ‘competing hypotheses are properly characterized as metaphysical assumptions, or whether they are in fact empirical hypotheses that are being tested in the course of […] research’ (op. cit., pp. 44–45). In other cases, certain commitments seem much less sensitive to anything resembling ‘direct’ empirical testing and stubbornly so. Consider, for instance, the assumption in Einstein’s theory of Special Relativity that the speed of light is the same in all directions. Furthermore, it is often hard to know if and when cases of the latter type may one day evolve into cases of the former.

Quite independently of which sort of case is at issue, though, consider our epistemic predicament at any given time. Leaving aside the issue of how the quality or quantities of empirical evidence may grow in future (something we may not be able to guess very well in the present), in the meantime, what many would identify as a priori presuppositions of scientific work do include descriptions of, and play roles in inferences to, aspects of scientific ontology. Neither Millikan nor Ehrenhaft, it seems, thought it best to suspend judgment until the possibility of some future verdict of history. Perhaps Gaeta would be tempted to say ‘so much the worse for them’ and ‘all’s well that ends well’—whatever is correctly judged to be insulated from compelling empirical assessment today will be cashed out empirically one day. In that case perhaps it would, in fact, be best simply to defer judgments about scientific ontology until that day has come. But is there any reason to expect agreement on the matter of when that day has arrived, if ever? The very question of whether a hypothesis is metaphysical, or sufficiently vulnerable to empirical testing to warrant belief, is contested by different epistemologies of science. Indeed, according to some (to which I turn next), when it comes to the existence of electrons and charge, that day of assent will never come.

3.2 Metaphysical Dimensions of Empiricist Views of Science

As mentioned earlier, one of the central claims I make about scientific ontology is that all approaches to it that interpret science as yielding some sort of knowledge of the world involve metaphysics in some way. Included in the scope of this claim are approaches that take this knowledge to be highly circumscribed, including some forms of empiricism and instrumentalism. A major focus of Gaeta’s appraisal of this is a defense of empiricism. As I will now attempt to show, however, this defense is both question-begging in ways that take it off course, and unnecessary as a defense per se, since nothing I affirm amounts to an objection to empiricism so much as an exploration of it in light of my understanding of scientific ontology. On my view, many empiricist approaches surely embody rational epistemic stances.Footnote 6

Let us consider first my charge of begging the question. Gaeta notes that ‘if following the empiricists, we understand metaphysics […] as beliefs which transcend experience […] and ontology in the […] deflationary sense (for instance, as Carnap and sometimes Quine understood it), then it is perfectly conceivable […] that scientific theories are free of metaphysical ontological commitments’. Indeed; but none of this is at issue in my claim that empiricism typically involves at least some metaphysics. The term ‘metaphysics’ is sometimes used merely to designate content that goes beyond what a given epistemic agent holds to be warrantable (or perhaps even worthy of consideration, literally construed). For the logical empiricist, talk of electric charge is metaphysics, but for the entity realist, it is not. Gaeta’s appeal to empiricist uses of ‘metaphysics’ reflects only this sort of idiosyncratic usage and begs the question in favor of it. It has no bearing on my claims about empiricism stemming from the more general characterization of metaphysics given above (in Sect. 3.1), in terms of the nature and function of the a priori in ontology. My contention is that even beliefs about the observable world require some recourse to this (more on which below). Thus, Gaeta’s appeal to deflationary approaches to ontology beyond the observable (‘transcending experience’) also misses the mark, for it is only the nature of experience that is at issue in my claims regarding empiricism.

With these clarifications in hand, let us now focus on the question of whether empiricism needs defending. Not on my account: while some empiricists—but only some, I believe—may disagree with the claim that there are a priori dimensions of experience, nothing in this claim serves to undermine possible empiricist demarcations of the epistemically warranted content of theories and models (claims about the observable) from further content that fails to satisfy their strictures for belief (everything else). Gaeta’s principal argument targets a particular rendering of the notion of theory-laden observation attributed to Norwood Russel Hanson and Thomas Kuhn, but these are specific exemplifications of the more general idea I intend, and nothing I intend is married to any one such view (op. cit., p. 57). The general idea is simply that there is something plausibly regarded as a priori in our characterizations of experience: the ways we experience the observable and the associated categories of things we thereby recognize and describe. Different hypotheses concerning how we do this, elaborated in terms of hardwiring, acculturation, and other forms of training or conditioning, are precisely what is theorized and explored in specific accounts (and in cognitive science, as Gaeta notes). What all of these views share, though, is just the idea that we ourselves bring something to the table in order to formulate experiences and descriptions of scientific (and other) observables, and to draw inferences about them.

What we bring to the table is not learned from experience itself, though we may, of course, through experience (whether consciously or unconsciously), gravitate toward categories and descriptions that serve our scientific and other purposes well. Is any of this controversial? I think not, but in any case, it is endorsed by empiricists themselves, as Gaeta concedes (e.g., fn. 8). Granted, there are ways of conceiving empiricism on which this does not translate into a role for a priori dimensions of human apprehensions of the observable world. Idealists may take themselves to describe only experience, not anything ‘external’ to which experiences may or may not correspond; quietists may refuse to be drawn on questions regarding the nature of experience and its (possible) objects at all. By this same token, neither of these approaches to thinking about the outputs of science offers an account of scientific ontology: a description of what science reveals about what things and kinds of things there are in the world, and what those things are like. Once we take the sciences to reveal such facts, presuppositions employed in apprehending and describing observables, and inferences employed in reasoning about them—for example, to infer that a theory is empirically adequate (that it yields truths about observable phenomena)—seem indispensable. That is to say, we make recourse to a priori dimensions of experience and reasoning about experience in giving accounts of scientific ontology.

No doubt an empiricist might, as Gaeta points out, go a different route and opt for a wholly deflationary attitude toward all questions of ‘external’ existence, even those concerning observable phenomena. This possibility, however, is subject to the same dichotomy of empiricisms just rehearsed. If the intention of a full-blown deflationist is to furnish an account of the output of theories and models exclusively in terms of impressions and ideas and nothing more, or to remain unwaveringly silent on questions of what exists, theirs is not an account of scientific ontology as such. If the intention is rather, in a Kantian or neo-Kantian spirit, to describe the subject matters of the sciences in terms of some kind of fusion of the transcendentally ideal and the empirically real, they are hardly avoiding metaphysics, even if this particular proposal would be for a Kantian or neo-Kantian metaphysic as opposed to any straightforwardly realist one. This brings us back, once again, to my original claim that any account of empiricism that is plausibly interpreted as furnishing an account of scientific ontology will make recourse to at least some metaphysics, in line with how I have explicated each of the pivotal notions on which this claim relies.Footnote 7

3.3 Varieties of Disagreement About Scientific Ontology

Earlier I identified a domain of interest for scientific ontology extending over both the explicit targets of inquiry in the sciences, as well as a number of implicit subject matters. The latter, while mentioned in scientific discourse, seem to require theorizing of a more metaphysical nature than is typically associated with the sciences if the characteristics or natures of these things—causation, laws of nature, etc.—are to be understood. Here, disagreements (of which debates about scientific realism and antirealism form a subset) abound: which areas of inquiry produce warranted beliefs about the world, and of what sort? My contention that different epistemic stances underlie judgments concerning which parts of this domain yield answers to questions of scientific ontology and which parts are better suited to suspension of belief, and arguments to the effect that even (some) conflicting stances can be rational, are intended to give answers to these questions.

In his paper ‘Disagreement About Scientific Ontology’, Borge (2024) is interested in overlapping questions, framed in terms of concepts and terminology now familiar from recent discussions in the epistemology of disagreement. This offers several points of contact and contrast with my own thinking about disagreement, perhaps most importantly relating to Borge’s central, analytical device corresponding to epistemic stances, which he calls ‘epistemic perspectives’ (Borge 2024, in this issue):

An epistemic perspective includes a set of epistemic policies oriented towards the achievement of some preferred epistemic goals, guided by specific values. Given a proposition, they determine the weight of the evidence in its favor and the rational doxastic attitude to be taken towards it […]. [T]he only noticeable difference between a perspective and a stance is that the latter, unlike the former, seems to be related to the production of knowledge that is (at least in a weak sense) vulnerable to empirical information, e.g. scientific knowledge. Epistemic perspectives[…]can be adopted in broader epistemic contexts, including the ones of (pure) metaphysics, ethics and other domains of pure a priori knowledge (domains in which items like intuitions can be regarded as evidence).

Thus described, how different are stances and perspectives? Both incorporate values and policies relevant to evaluating evidence, but what of the contrast noted above regarding domains of application? I suspect that some initially plausible differences may vanish upon reflection, but others may survive. My own discussion of stances arose in the context of thinking about scientific ontology, but this all by itself need not rule out the possibility of applying the concept more broadly, even in domains where empirical vulnerability is not merely weak, but (arguably) very weak. On my view, assessments of how robust empirical vulnerability must be, exactly, in order to facilitate warranted belief, will depend in part on how much evidential weight an agent places on explanatory power, and how powerful she takes any given explanation to be. It is for this reason that some believe one or another interpretation of quantum mechanics, or of quantum gravity, where empirical vulnerability is hard to fathom. To whatever extent empirical phenomena have purchase in other domains of inquiry such as ethics, the same may be true there (though I have not been so brave as to stray from the borders of scientific ontology to argue this). How further projects in metaphysics stack up here will likely be controversial, but in uncontroversially pure domains of abstraction such as mathematics, perspectives may be more applicable than stances after all.

The most forceful case for distinguishing epistemic stances and epistemic perspectives would seem to rest on their more precise roles in determining doxastic attitudes with respect to one and the same narrowly defined topic. Is there sufficient warrant for belief in quasiparticles, or individual organisms made up of colonies, or dark matter? The primary function of a stance is to determine, for any such area of inquiry, whether belief is warranted or whether one should be agnostic. Thus, conflicting epistemic stances do not support different beliefs of the form P and ~ P, respectively, but rather judgments to the effect that an area of inquiry has the resources—combinations of empirical vulnerability and explanatory power—to yield epistemic warrant, or that it does not, in which case one should suspend belief. In contrast, Borge suggests that different epistemic perspectives do in fact produce beliefs of the form P and ~ P in connection with debates on particular questions. His flagship example is of rival accounts of the metaphysics of laws of nature. Whether one finds Humean (best systems) views or governing (relations between properties) views or dispositional views of laws correct, he argues, reflects different assessments of and tradeoffs between the ‘weight’ of the metaphysical posits involved and the putative explanatory power that positing them would afford.

This contrast between what epistemic stances and epistemic perspectives license in terms of doxastic attitudes points to a substantive difference, whose significance is enhanced by something that Borge and I both endorse in our own ways: the rational indefeasibility of well-formed stances, in my case, and perspectives, in his. The idea is that in cases where conflicting stances (perspectives) meet certain minimal constraints of rationality, no non-question-begging means of ruling them out remain; the viability of an epistemic stance (perspective) is ultimately exhausted by its internal coherence, both logical and pragmatic.Footnote 8 The consequences of this, however, are profoundly different in the two cases. Given that conflicting judgments afforded by different but rational stances regarding a particular question of ontology concern only whether inquiry is capable of producing warranted belief, such differences in stance merely reflect how far agents are willing to go, along a spectrum of increasingly metaphysical debates about ontology, in forming beliefs. In contrast, conflicting but nonetheless rational perspectives (or diverging applications of policies within perspectives) allow contrary beliefs about matters of fact to be rational, and indefeasibly so. It is rational to believe that laws are relations between properties; it is also rational to believe that they are not. I suspect that some may deem this a bridge too far in the direction of thoroughgoing epistemic relativism.

3.4 Pyrrhonism and Realism, Scientific and More Generally

The primary function of epistemic stances, on my view, is to determine where lines are drawn between domains of inquiry in which belief is apt and ones in which it is better to suspend judgment, but this too has pluralistic and relativistic implications.The pluralism follows from the idea that more than one stance may be rational, and the relativism from the idea that rational stances may conflict with one another and are thus, in that case, not simultaneously adoptable by any one agent on pain of irrationality. Henschen (2024) worries about the conclusion I draw from this regarding the very nature of debates about scientific ontology in his paper ‘Subatomic Particles, Epistemic Stances, and Kantian Antinomies’. As he notes, I appeal to Sextus Empiricus’ rendering of Pyrrhonian skepticism to illustrate a sense in which these debates are, it turns out, irresolvable. His worry stems from what he sees as selective attention on my part to disputes between those holding versions of the empiricist stance, on one side, and those holding versions of the metaphysical stance, on the other. This has the unfortunate consequence of neglecting a further possibility that may serve to undermine my Pyrrhonian diagnosis: a Kantian approach.

Let me spell out the content of Henschen’s worry, which I take to comprise two substantial claims. The first is that a Kantian approach to scientific ontology can be viewed as dissolving conflicts between empiricism and various sorts of realism—regarding both the explicit and implicit content of scientific theories and models, and perhaps more besides. The second claim is a consequence of the first, to the effect that the Pyrrhonian conclusion I reach about stances is thus unmotivated: in virtue of the Kantian dissolution of debates between those with conflicting empiricist and metaphysical stances, there is no basis for skepticism about a principled resolution of these debates after all. Each of these claims poses a challenge, which I will attempt to answer in the following Sect. (3.5). In order to make my case, however, it will be helpful first to clarify more precisely what my Pyrrhonian conclusion is, exactly. With this in hand, the hope of wriggling out of a skeptical diagnosis of the fate of disputes about scientific ontology is dashed, or so I will argue.

Henschen (op. cit.) describes my appropriation of Pyrrhonism in terms of ‘the ability to appreciate the equal strength […] of the arguments that empiricists and metaphysicians develop to support their claims’. A minor point here, with major repercussions, concerns the implied restriction to interlocutors adopting the empiricist and metaphysical stances. This would exclude those with deflationary stances, in whose company Henschen (correctly, I believe) locates his Kantians, but this is not my intention. I will return to this point below, but for now let me focus on what seems an ambiguity in this description of my view in terms of the equal strength of ‘arguments’ used to ‘support’ conflicting claims. How strict or loose are these uses of the terms ‘argument’ and ‘support’? Sextus is strict on both counts: the Pyrrhonist, he says, is able to appreciate the equal strength of arguments for and against factual propositions; an argument is an assessment of reasons and evidence culminating in a putative demonstration of the truth of a proposition and, thereby, offers direct support for belief. Epistemic stances, however, comprising attitudes, values, aims, and policies, are not strictly propositional. Their adoption may at best involve (and not always explicitly) a looser sense of ‘argument’, and ‘support’ for whatever beliefs may eventuate is indirect, through demarcations of domains of inquiry where belief is apt.

Thus, I do not assert, as one might on a more straightforward application of Sextus, that Pyrrhonian skepticism is the inevitable outcome of the cut and thrust of arguments about specific ontological claims. Rather, the thought is that something like Sextus’ notion of the equal strength of considerations on both sides of a question applies to disagreements about which stance, among rationally permissible options, one should adopt. Here, ‘equal strength’ pertains to the only truly neutral feature of assessment a stance may exhibit, namely, epistemic rationality. Assessments of other comparators would seem to presuppose the very attitudes, aims, and policies that constitute one or another stance, and to insist on the preferability of one in particular on the grounds that other rational options differ in these respects would be simply to beg the question. It is at this juncture that I find inspiration in Pyrrhonism. Noting the ‘equal strength’ of certain stances qua rationality, and lacking further, stance-neutral criteria with which to adjudicate them, we should accept with equanimity the fact that there is no uniquely correct way to understand the scope and limits of scientific ontology. Different agents may, without violating plausible canons of rationality, apply different values to this determination, reflecting how much epistemic risk they are willing to accept. With this clarification, let us turn now to face Henschen’s worry head-on.

3.5 The Kantian Synthesis of Metaphysics and Empiricism

Recall that there are two questions at issue here. The first is whether a Kantian approach to scientific ontology resolves conflicts between those drawn toward (relevant versions of) the empiricist stance and the metaphysical stance, respectively. My answer to this first question is ‘yes, and no’. The second question is whether, if the answer to the first question is ‘yes’, my Pyrrhonian diagnosis of the nature of scientific ontology is undermined. My answer to this second question is ‘no’. In less enigmatic terms and in conclusion, let me now flesh out the reasoning behind these answers in hopes of providing some final illumination of the project of scientific ontology.

Henschen prefaces his advocacy of a Kantian approach to ontology with the observation that, in the book he engages (Chakravartty 2017), I seem preoccupied with clashes between those holding empiricist and metaphysical stances. As a result, the Kantian approach is ‘underrated’. While this charge does not reflect my own view of a number of Kantian approaches, I do appreciate that my attention to others may give this impression, which I regret. As a first step toward rectifying this, let us recall that, as noted above, Kantian conceptions of scientific ontology operate with stances that are deflationary in the sense that they recast a literal or face-value understanding of ontology, conceived as an investigation into what exists mind-independently (commonly associated with forms of realism), in other terms: in this case, as an investigation into a mind-dependent reality, a phenomenal world, representing a fusion of human ways of understanding with an otherwise ineffable, mind-independent, noumenal world. The fact that there are many ways of expounding this general idea—of explicating transcendental conditions of experience—accounts for a variety of Kantian and neo-Kantian positions, but all are deflationary in my sense. Often, deflationary stances embody considerable unease with or a distaste for what may be regarded as overreaching ambitions to know things-in-themselves, and the construction of naively optimistic theories of truth and reference to match.

This understanding of Kantian conceptions of scientific ontology, fueled by deflationary stances, yields an answer to the first question arising from Henschen’s discussion, of whether a Kantian approach resolves the opposition of empiricist and metaphysical stances. Much of his discussion is devoted to a case: a conflict between the (in some ways) empiricist-leaning metaphysics of David Lewis and the metaphysics of ontic structural realism regarding issues of identity and individuality.Footnote 9 Henschen suggests that the arguments for these conflicting ontologies ‘have equal strength’, producing contrary propositions resembling a Kantian antimony (very roughly and more generically than Kant intends, a contradiction in which each of the contradictory elements seems reasonable). Kant himself argued that resolving his antinomies requires a Kantian understanding of our knowledge of the world, and Henschen argues that a similar resolution is fitting in his own case. Without going into the details of these cases, however, we are already in a position to see why my answer to the question of whether a Kantian position resolves such conflicts is ‘yes, and no’. The answer may well be ‘yes’ if operating with an epistemic stance giving rise to the Kantian position; it may well be ‘no’ if not. For reasons given in the previous section, so long as the relevant, alternative stances are rational, it is difficult to see how the answer could be otherwise.

My parsing of Henschen’s worry is complicated, I think, by his presentation of his illustrative case of dueling points of view as a conflict between different claims about identity and individuality. For reasons canvassed earlier in connection with Borge’s discussion of disagreement (Sect. 3.3), if the conflict here were of this sort only, Borge’s epistemic perspectives would be a more appropriate target of Henschen’s challenge than epistemic stances. On reflection, though, I believe that Henschen’s intervention should be read differently. His resolution of the case he examines invokes an approach to ontology that goes well beyond mere disputes about individuality. If, as seems clear, his recommendation is for the adoption of a deflationary stance incorporating Kantian epistemic policies, to view his case simply as one concerning conflicting propositions would be to see only the tip of the iceberg and not the underlying stance he intends. Once we appreciate that the Kantian approach he recommends is, in fact, a particular realization of the deflationary stance—one that I would grant is rational, and not merely for the sake of argument—a vindication of my Pyrrhonian conclusion regarding scientific ontology is complete. Where Henschen glimpses a resolution of debate via Kant, a broader panorama reveals the coherence of one stance among others, each with its own conception of what a fitting resolution would be.