Author Archives: Kareem Khalifa

About Kareem Khalifa

Department of Philosophy, Middlebury College

Ontology with Lessons from Pyrrho and Sextus

Chakravartty concludes his defense of stance voluntarism by drawing analogies with the ancient skeptics Pyrrho and Sextus Empiricus.

  • The Agrippan Trilemma
  • Background

This was an ancient skeptical argument. Here’s the setup.The trilemma’s key assumption is what might be called the No Free Lunch Principle:

If q is S’s justification for believing that p, then S must be justified in believing that q.

This assumption seems plausible to many. Yet, it leads to one of the toughest skeptical arguments in the history of Western philosophy. Here’s the gist:

Suppose that you claim to know that p (e.g., that frogs are green.)

How do you know that p?

Because q (e.g., you’ve seen frogs in good lighting conditions, and they appear green)

But how do you know that q?

Because r

How does this end? Three options:

  • It doesn’t end (infinite regress).
  • r is justified by p or by q (circular reasoning)
  • r is unjustified (arbitrary assumption)

Infinite regresses, circular reasoning, and arbitrary assumptions fail to yield justification. Hence, it appears that no belief is justified.[1]

Nevertheless, each of the three bullet points can be given a positive spin. The most exotic are infinitists who hold that infinite justification isn’t as utopian as it may initially seem. Coherentists distinguish between “vicious” and “virtuous” circles, and then argue that circular reasoning of the latter sort is justificatory. Finally, foundationalists hold that beliefs can be self-justifying without being arbitrary.

  • Should We Be Skeptical About Stances?

Recall that stances are justified via coherence. Chakravartty also describes them as “foundational,” albeit only in a qualified sense. However, he also asserts the following:

A stance is not something for which one gives justifications as such, but rather something that one adopts because it reflects what one values, epistemically. (242)

One might not think that Chakravartty is letting himself off the hook too quickly here. Although one does not give justifications for stances, considerations of coherence point to at least one way in which stances can be justified or not. Moreover, one may wonder whether one’s choice of core epistemic values is arbitrary. If so, then stances fall on the third horn of Agrippa’s trilemma.

  • Isotheneia and Ataraxia

Chakravartty suggests that we can learn other lessons from Pyrrhonian skepticism. First, we can align competing arguments to appreciate their equal strength or isotheneia. Chakravartty suggests our situation with competing stances is similarly in a state of isotheneia. He takes this as ground for collaborative epistemology, which he defines thusly:

elaborating stances and the values that favor them, entertaining them dispassionately, seeking to understand their purchase on ourselves and our interlocutors— this must not be conceived as a surreptitious means of arguing for the superiority of any given rational stance, for as we have seen, the relevant mark of strength here is rationality, and all rational stances are rational. Regarding the question of superiority there can be only aphasia, speechlessness, and suspension of judgment. (244)

In ancient skepticism, recognition of isotheneia brings about ataraxia or peace of mind. In short, we see that trying to hash out the superiority of one stance over another is futile, and so we stop engaging in such endeavors. Here we get one of Chakravartty’s clearest pronouncements on the upshot of his book:

The disagreements were ill formed. The attempts to resolve them were wrongheaded. With a clearer understanding one is now free to focus attention elsewhere, on issues worthy of philosophical agonizing. One stands relieved. (245)

[1] Here’s a slightly fancier version of it:

A1. For all S, p, and q, if q is S’s justification for believing that p, then S must be justified in believing that q.

A2. If A1 is true, then all justification results in an infinite regress, a vicious circle, or stopping at an arbitrary assumption.

A3. We cannot possess an infinite chain of justification.

A4. Viciously circular reasoning does not provide justification.

A5. Stopping at an arbitrary assumption does not provide justification.

A6.\ No belief is justified. (A1-A5)

The Nature and Provenance of Epistemic Stances

  1. An Indefeasible Persistence of Ontological Disagreement

Recall that Chakravartty distinguished between two kinds of “ontological uncertainty” in the previous chapter:

  • Given inconsistent descriptions of a given object, event, process, etc. within a domain, which description are we justified in believing? For example, are fluids continuous media or collections of discrete particles? This kind of uncertainty was Chakravartty’s focus in the previous chapter.  
  • In which ontological domains is belief justified and which is it not? For example, should we be realists about DNA and antirealists about processes? This is Chakrvartty’s focus in this chapter.

One might think that debates about scientific ontology are “philosophical business as usual.” Few perennial debates in philosophy ever admit of resolution or clear victors. There are two ways to look at this:

  • Given enough time, effort, and ingenuity, a clear victor will emerge by offering a decisive argument.
  • In principle, no such victory is possible, because the best arguments that can be offered will always presuppose some of the disputed grounds between opposing viewpoints.

Chakravartty adopts the latter view, but holds that the problematic presuppositions are best regarded as epistemic stances.

  • Stances Revisited
  • What is a stance? For Chakravartty’s purposes, stances are collections of “attitudes, commitments, and strategies” regarding “how purported knowledge of scientific ontology is generated.” (206)
  • Stances have non-propositional content. Propositions can be expressed as declarative sentences and can follow “that” in English. Consequently, they are not beliefs. For example, it would be incorrect to say that the empiricist stance involves belief that explanations that posit unobservable entities are bad.
  • They reflect different assessments of empirical vulnerability and explanatory power, and thereby dictate one’s tolerance for epistemic risk.

Question: One thing that has puzzled me is this claim about non-propositional content. Aside from the fact that it produces the voluntarist’s desired result, I don’t see the motivation for denying that (i) empiricists believe that certain explanations are bad, while affirming that (ii) empiricists take a negative stance toward those same explanations. The irresoluble nature of ontological debate might be that it is very hard to say what makes something an explanation, what makes it good or bad, etc. That could all be couched in terms of propositional commitments, such as beliefs, and the intractability of the debates might simply be due to the difficulty in coming to know these propositions, rather than some difference in stances. I don’t know how to adjudicate between Chakravartty’s position and this alternative. Any ideas?

Three stances are sketched:

  • The Deflationary Stance

D1.      Reject traditional philosophical (i.e., realist) understandings of scientific ontology.

D2.      A fortiori, reject the analyses of truth and reference with which they are typically explicated.

Chakravartty associates this stance with historicist, social constructivist, and pragmatist philosophers of science. He mostly seems to associate it deflationary views with “non-literal” interpretations of science, which is one way (but not the only way) in which to interpret D2. (I have voiced my misgivings about calling these views “deflationary” in an earlier post.) Chakravartty largely focuses on the other two stances.

  • The Empiricist Stance

E1.       Reject demands for explanation in terms of things underlying the observable.

E2.       A fortiori, reject attempts to answer these demands by theorizing about the unobservable.

Note that this stance puts greater weight on empirical vulnerability (or “empirical answerability,” as we prefer) and less weight on explanatory power.

  • The Metaphysical Stance

This is essentially the mirror opposite of the empiricist stance:

M1.      Accept demands for explanation in terms of things underlying the observable.

M2.      Attempt to answer these demands by theorizing about the unobservable.

Note that this stance puts less weight on empirical vulnerability (or “empirical answerability,” as we prefer) and greater weight on explanatory power.

  • Voluntarist Primer on Choosing Stances & Beliefs

Chakravartty endorses stance voluntarism: which stance one adopts is a matter of choice.

This entails a modest form of “indirect” doxastic voluntarism: what one believes is a matter of choice.

Essentially, one can make a direct choice as to what stance one adopts, and this will influence what one believes However, this is consistent with having no direct choice over what one believes; only an indirect choice. (Compare: one can make a direct choice as to how much evidence one seeks, and this will influence what one believes.)

How does one choose a stance then? Chakravartty appeals to epistemic values:

“The stances of agents reflect the things they value, epistemically, including certain kinds of information and explanation, certain kinds of evidence and argument, and intuitive judgments about what kinds of information, explanation, evidence, and argument support inferences to ontological claims and to what degrees, all of which then translates into certain epistemic policies.” (220)

However, he does not think that there is much to be said about how epistemic values are adopted. They seem to be the “bedrock” in his account.

  • It’s worth noting that there is quite a bit of debate about epistemic value in epistemology. It’s not clear how much of it bears on this discussion. Yet, as we’ll see below, Chakravartty must have some constraints on what counts as epistemic value if he is to defend his permissive conception of rationality.
  • Epistemic Stances in Conflict: Rationality & Robustness

Thus, stances are epistemic policies about how ontological knowledge should be produced from scientific claims, and they are chosen freely based on one’s core epistemic values. So, ontological commitments (beliefs about what exists) can differ as a result of one’s stance. But is every choice in stance a rational one? Chakravartty asserts the following:

If we regard the scope of ontological commitment as something that may vary, rationally, as a function of different stances, the conception of rationality at issue will have to be “permissive” in the sense that it allows (potentially) more than one stance and resulting set of beliefs, given evidential considerations, to count as rational. (223)

Following van Fraassen, he takes the rationality of adopting a stance to consist of three things:

  • The stance must not lead to logical incoherence, such that beliefs contradict each other.
  • The stance must not lead to probabilistic incoherence, such that degrees of belief violate the probability calculus.
  • The stance must not lead to pragmatic incoherence, such that the beliefs that are in tension with the attitudes and orientations of the stance itself and lead to “self-sabotage.” Later, and perhaps more clearly, he describes pragmatic incoherence as the requirement that “the stance is not self-defeating, in the sense that it leads one to believe propositions that are in tension with (thus ‘sabotaging’) the very application of the epistemic values that define the stance itself.” (232)

Chakravartty uses the blanket term “internal coherence” to cover the satisfaction of all three conditions.

Given the permissive conception of rationality, what debate can there be between those holding conflicting stances? Chakravartty makes the following points:

  • One can criticize another stance for its internal incoherence.
  • Showing that well-established stances—such as the three sketched above—are incoherent is extremely difficult. Why?
    • Stances are more resilient than the ontological beliefs they license or “facilitate.” One can revise the latter without revising the former, and often this is what is done in the face of recalcitrant evidence.
    • The only way to criticize a stance directly is by targeting the values that led to its adoption. But the deflationist, empiricist, and metaphysical stance’s core values appear internally coherent.

Consequently debates between partisans of different stances are deep and intractable. The positive lesson that Chakravartty draws is:

“the nature of the debate is transformed: we can articulate our stances, put our epistemic values on the table for examination by ourselves and by others, explain how and why they resonate with us, invite others to empathize, and encourage the same with respect to our interlocutors. This is the basis of a collaborative epistemology in the context of voluntarism.” (228)

  • In Defense of Permissive Norms of Rationality for Stances

Chakravartty concludes by addressing two final challenges to voluntarism.

  • The Relativist Argument

One objection to permissive rationality goes as follows:

R1.       If rationality is permissive, then the same body of evidence can justify the belief that P and also justify the belief that not-P.

R2.       The same body of evidence cannot justify the belief that P and also justify the belief that not-P.

So, rationality is not permissive.

For those who are interested, Chakravartty cites the following article, which is “burning up the charts” in epistemology:

  • Schoenfield, M. 2014: “Permission to Believe: Why Permissivism Is True and What It Tells Us about Irrelevant Influences on Belief,” Nous 48: 193– 218.

More generally, voluntarism in philosophy of science has many affinities with permissivism in epistemology. For a recent criticism of permissivism, see:

  • Schultheis, Ginger, 2018, “Living on the Edge: Against Epistemic Permissivism”, Mind, 127(507): 863–879. doi:10.1093/mind/fzw065

Chakravartty denies the first premise, and replaces it with the following:

R1*.     If stance voluntarism is true, then the same body of evidence can justify the belief that P and also justify the suspension of belief about P.

This is because of epistemic risk. Consider an analogy: confronted with the same data, some medical doctors might believe that a medical treatment is effective while others might suspend belief until more/different evidence comes in, owing to worries about harmful side-effects. If this strikes you as plausible, then Chakravartty is simply extending this to debates about scientific ontology. Some philosophers might believe an ontological claim, while others suspend judgment until there is more empirical evidence.

  • The Pathology Argument

The second objection to stance voluntarism is as follows:

P1.            If rationality is permissive, then pathological stances will be rational.

P2.            Pathological stances are irrational.

So, rationality is not permissive.

What is meant by a pathological stance? The paradigmatic example (232) is of a stance that disregards relevant evidence.

Chakravartty challenges the first premise, counterarguing as follows:

C1.           If rationality is permissive, then every rational epistemic stance is pragmatically coherent.

C2.           Any epistemic stance that permits the disregard of relevant evidence will be pragmatically incoherent with respect to acquiring truth.

C3.           Every epistemic stance has the acquisition of truth as one of its core values.

Not-P1.    Even if rationality is permissive, pathological stances will not be rational.  

Let’s grant C1. Is C2 true? Won’t quite a bit depend on which evidence is disregarded and which truths are valued? Perhaps Chakravartty would argue that the kinds of cases I have in mind don’t speak to the pathology objection. Truly pathological stances are ones in which all relevant evidence for all valuable truths is disregarded. Finally, consider C3. Might someone seek something that falls short of truth that is still epistemically valuable? Chakravartty might reply that such agents are not seeking something that falls short of truth; they are simply keying in on certain truths to the exclusion of others. For instance, empiricists only value truths about observables.

Knowledge Under Ontological Uncertainty

  • Inconsistent Ontologies and Incompatible Beliefs

Some quick reminders: ontology is the philosophical study of what exists. But ontological claims require an epistemology: when is one justified in claiming that something exists? Here, we have seen that ontological claims are justified via metaphysical inference, and metaphysical inference is influenced by judgments of epistemic risk. Epistemic risk, in turn, is finding an optimal balance between empirical answerability (“vulnerability”) and explanatory power. But what one considers an optimal balance is stance dependent. Consequently, there is a continuum of ontological claims that different philosophers and scientists will deem justified given their different stances.

If one accepts this picture of ontological justification, then two kinds of uncertainty are likely to arise.

  • In which ontological domains is belief justified and which is it not? For example, should we be realists about DNA and antirealists about processes?
  • Given inconsistent descriptions of a given object, event, process, etc. within a domain, which description are we justified in believing? For example, are fluids continuous media or collections of discrete particles? This kind of uncertainty is Chakravartty’s focus in this chapter.  

Focusing on the latter, note that stance voluntarism has ample room of “reasonable” or “faultless” disagreement: individuals with different stances may have inconsistent beliefs, yet both are justified in those beliefs.

Chakravartty contrasts the stance voluntarist’s approach to reasonable disagreement with a neighboring position called perspectival realism.

  • Belief and Ontological Pluralism/Perspectivism

Suppose that we have a pair of inconsistent descriptions D and not-D. Perspectivists propose to resolve some of these inconsistencies by thinking of these two descriptions as elliptical for:

            According to Perspective A, D.

            According to Perspective B, not-D.

Note that these two “according-to claims” are consistent. For instance, imagine that A and B are standing in a line with a third person, C, in between them. To use a non-scientific example, according to A’s perspective, C is on the right, but according to B’s perspective, C is on the left. There is no contradiction in this. The same thing can be said in science, but perspectives won’t simply be spatial positions: they will include models, instruments, and the like.

As Chakravartty (2017, 174) notes, “not all cases of inconsistent models give rise to the worry of incompatible beliefs.” First, an underlying theory can explain away the apparent inconsistency. Second, one of the inconsistent models may be on better footing than the other, such that the inconsistency clearly counsels rejection of the latter model. Third, where the inconsistent models appear to be on a par, then suspension of belief is appropriate. However, at least the first two cases are catch as catch can. As such, perspectivism might well be the preferred strategy when those alternatives are unavailable, and suspension of belief seems inappropriate.

Chakravartty considers two kinds of perspectivism:

P1. Knowledge of scientific ontology is bound within specific contexts because our epistemic abilities do not extend as far as perspective-transcendent knowledge.

P2. Knowledge of scientific ontology is bound within specific contexts because there are no perspective-transcendent ontological facts or states of affairs to be known.

Note that P2 implies P1, but not vice versa.

  • Perspectivism’s Trilemma

Chakravartty claims that perspectivism of this sort will end up with one or more of the following untoward consequences:

  • Irrelevant: While certain aspects of scientific practice are perspectival, they frequently have no ontological upshot. For instance, the following is a bad inference: Lavoisier proposed the theory of oxygen from a particular perspective. Therefore, our epistemic abilities do not extend as far as perspective-transcendent knowledge that oxygen combusts.
  • Unstable: A putatively perspectival ontology collapses into a non-perspectival or deflationary ontology upon closer scrutiny.
  • Incoherent: Some perspectival claims do not really resolve the problem of inconsistent models; a contradiction still lurks.

Chakravartty’s goal: to show that these problems arise for the key arguments for P1 and P2. Let’s consider three arguments for P1 and P2, and Chakravartty’s objections to them.

  • Argument from Partial Modeling
  • PM1. If scientific modeling only provide partial representations of their targets, then non-perspectival facts are unknowable.
  • PM2. Scientific models only provide partial representations of their targets.

P1.       Therefore, non-perspectival facts are unknowable.

  • Chakravartty’s Objection: PM1 is false. It’s possible to have knowledge of non-perspectival facts even if complete representations are impossible. Consequently, perspectivism is irrelevant or unstable.
  • Argument from Conditioned Detection

CD1.    If different detectors produce (seemingly) inconsistent but equally acceptable descriptions of the same target system, then non-perspectival facts are unknowable.

CD2.    Different detectors produce inconsistent but equally acceptable descriptions of the same target system.

P1.       Therefore, non-perspectival facts are unknowable.

Perspectivism and CD1: Suppose that Detector D1 indicates that a target system T has property x, while Detector D2 indicates that T has property y, and that nothing with x can also have y or vice versa. Then, to avoid a contradiction, we should adopt perspectivism: According to D1, T is x; According to D2, T is y.

Chakravartty and CD1: Scientists can frequently show that detectors are describing one and the same perspective-independent target. In this case, once again, perspectivism is either irrelevant or unstable.

  • Argument from Meaning & Reference

MR1.   If the structure of the world depends on the concepts we impose upon it, then there are no non-perspectival facts to be known.

MR2.   The structure of the world depends on the concepts we impose upon it.

P2.       There are no non-perspectival facts to be known.

  • Chakravartty’s Objection:
  • Suppose that one person’s concepts entail that T is x, and another’s, that T is not x.
  • If the structure of the world depends upon the concepts we impose upon it, then T is both x and not-x.
  • So, either (a) everyone’s concepts entail the same thing for every target system or (b) the structure of the world does not depend upon the concepts we impose upon it.

In the first case, (a), we seem to have a perspectivism that only allows one perspective, which isn’t really perspectivism at all (incoherence). In the second case, (b), MR2 is false, and so there’s no good reason to accept P2 (instability).

  • Context-Transcendent Pluralism About Ontology

The uncertainty with which we began still requires pluralism about ontology. However, Chakravartty takes the preceding arguments to indicate that this pluralism should not be perspectival. He offers two alternatives.

  • Pluralism about Packaging

First, there is pluralism about packaging, the position “that entities and kinds of entities may come in different sorts of packages, corresponding to different aggregations, even at one and the same level of analysis” (190). Which package is appropriate depends on the particular prediction, explanation, or other scientific aim is relevant.

Regardless of whether one thinks of the aggregate as merely conventional or having reality over and above its constituents (anti-reductionism), there is no difficulty with the problem of inconsistent models. For conventionalists, this collapses into a deflationary ontology, where only the lower-level entities exist and the higher-level talk is merely convenient. For anti-reductionists, there is no inconsistency, because different aggregates are different things, and thus no inconsistency arises.

  • Pluralism about Behavior

Second, there is pluralism about behavior. This hinges on dispositional realism. Suppose that a target T has a dispositional property, such that, if condition C1 is the case, then T behaves like an x; if C2, then T behaves like a y, where x and y are incompatible properties. (Note that objective conditions are playing the same role of dissolving the contradiction that perspectives played for perspectivists.)

  • Not Perspectivism

Neither pluralism about packaging nor pluralism about behavior are perspectivist, since on either view, one can transcend one’s context/perspective when making ontological claims. There are a number of perspective-independent facts out there, and context/perspective functions more like a spotlight for focusing on particular facts of interest. Just as facts do not fade out of existence when we turn our spotlights elsewhere, these pluralist positions do not claim that aggregates or dispositions depend on perspectives for their existence.

  • Ontological Explanation and Contrastive Why-Questions

Chakravartty concludes by showing that pluralism about ontologies lead to “anything goes” about ontology. For instance, pluralism about packaging requires any description of an aggregate to serve some scientific goal (explanation, prediction, etc.) Since dispositions are out there in the world, packaging about behaviors are objectively constrained.

Chakravartty proposes a theory of contrastive why-questions that is compatible with a harmless form of perspectivism: “Investigative contexts afford perspectives, which in turn facilitate knowledge of non-perspectival facts.” (197)

On such a view, questions of the form, “Why P?” are often elliptical for questions of the form, “Why P rather than Q?”, where context frequently indicates the value of Q (often called the foil of a why-question). Note that A may correctly answer this question, but not be a correct answer to the question, “Why P rather than R?” For instance, “Why did Khalifa (rather than Chakravartty) teach the seminar in Middlebury’s philosophy department?” only needs to point to the fact that Khalifa is a Middlebury faculty member. However, this fails as an answer to the question “Why did Khalifa (rather than Besser) teach the seminar in Middlebury’s philosophy department?,” which would have to point to certain teaching rotations in Middlebury’s philosophy department. Note that the two answers are compatible but their truth does not hinge on perspective. Regardless of your perspective, Khalifa is a member of the Middlebury Philosophy Department, and Chakrvartty is not, for example. So a kind of non-perspectival pluralism emerges.

Structures: science as a constraint on scientific ontology

  1. Fundamental physics and ontology

The Standard Model of particle physics posits entities you may have heard of such as quarks and gluons. Prior to its inception, the world was thought to consist of four fundamental forces:

  • strong forces (the force that holds between, e.g., a proton and a neutron in an atomic nucleus)
  • weak forces (the forces at play in radioactive decay)
  • electromagnetic forces
  • gravitational forces

The Standard Model unifies the first three.

The philosophical task is “to describe the ontology of fundamental physics in a way that is compatible with the notion that it has some causal or physical oomph.” (135)

Although Chakravartty is agnostic about nearly every aspect of this “oomph,” he notes one important commitment shared by all “traditional” theories of causation: that causation involves entities. Entities are broadly construed to include objects and events, construed in myriad ways.

This entity-based ontology stands in tension with structuralism, which frequently downplays the importance of entities in order to emphasize relations or structures.

2. “Particles”

Colloquially, the Standard Model is frequently thought to posit a variety of particles—protons, neutrons, electrons, bosons, mesons, etc.

However, these should not be thought of as point-like entities bouncing into each other. That would be to impose classical mechanics on the quantum world. In the Standard Model, “particles” are modes of excitation of a quantum field. 

Rather, “particle” is merely a placeholder for the properties associated with them (mass, charge, spin). This is an occasion for ontological theorizing, where various puzzles arise because quantum mechanics is weird. One of the biggest puzzles is permutation invariance:

  • Swapping one particle for another of the same type within a system does not constitute a new physical arrangement. But if it is impossible to distinguish particles in this way, are they really distinct? So, it’s not even clear what we are claiming is real when we say, e.g., that electrons are real.

  • 3. Structuralist interpretations

The standard (non-structuralist) picture holds that relations ontologically depend on their relata, the entities that stand in those relations.

  • For instance, take the relation “is a student of”. If you and your professors (the relata of this relation) did not exist, then this relation would be empty. If you did some add/drop gymnastics such that you took courses with a different professor, the relation would change. (Notice the difference with fundamental particles: on the non-structuralist/standard picture, “is a student of” is permutation variant.)

Non-eliminative structuralism inverts the standard picture: entities ontologically depend on the relations in which they stand.

  • Entities are just positions (locations, nodes) in patterns or structures. They have no intrinsic properties.
  • It is opposed to particles having “kind identities.” While everyone agrees that different kinds of particles are individuated by mass, charge, and spin (among other properties), non-eliminative structuralists are unique in taking these to be extrinsic properties.

Eliminative structuralism goes even further: everything is a relation; nothing is an entity.

  • Talk of entities is merely a convenient shorthand/useful fiction for referring to relations/structure.

Structuralism does a nice job with the Standard Model. If a particle can’t be individuated in a physical arrangement, then the most we can say is that it plays some role in that arrangement. But arrangements are just sets of relations, i.e., structures.

  • 4. The Dilemma

Chakravartty presents a dilemma to both kinds of structuralism:

General schemaEliminative structuralismNon-eliminative structuralism
Given a version of structuralism, determine the locus of causal efficacy.Concrete structures are the locus of causal efficacy.Entities are the locus of causal efficacy.
In evaluating this view, encounter a metaphysical challenge.How can structures be concrete if they have no relata? (cf. the standard picture.)For this to be structuralist, entities must be defined entirely in terms of their extrinsic properties. But what kind of entity is that?
In replying to the challenge, posit a contentious ontological primitive.Posit “relations-in-themselves” or concrete structures.Posit “internally extrinsic entities.”
Dilemma: accept the contentious posit, or reject this version of structuralism.Accept relations-in-themselves or reject eliminative structuralism.Accept internally extrinsic entities or reject non-eliminative structuralism.
  • 5. Dissolving the dilemma

Chakravartty uses the framework from Part I of the book to show how one can adopt different attitudes to different claims about subatomic particles. Most realists will endorse “that there are subatomic particles, conceived as entities subject to various detections, measurements, manipulations, and novel predictions” (159). There can be varying degrees of tentativeness in one’s attitude toward the more specific structuralist (eliminative and non-eliminative) proposals about particle ontology.

As you might have guessed, the varying degrees of tentativeness are a reflection of one’s epistemic risk, which in turn is determined by one’s stance. Chakravartty notes that suspension of belief is compatible with other kinds of epistemic attitudes. For instance, one might suspend belief about the structuralist ontology of particles, but still adopt “a pragmatic attitude toward theorizing about the finer-grained ontological natures of subatomic particles” that serves to “elaborate conceptual pictures of the ontology of fundamental physics in different ways, which may then be heuristically fruitful for scientific investigation down the road” (162).

Dispositions: Science as a Basis for Ontology

  1. Dispositions in the Philosophy of Science

Goals of this chapter: to show how the framework in Part I of Scientific Ontology (a sliding scale for metaphysical inferences indexed to differing degrees of epistemic risk) applies to actual work in scientific ontology (in this case, dispositions).

  1. What is a disposition?

A disposition is “a causal power” (100) that is only manifested under certain conditions.

  • For example, salt is soluble = salt, when immersed in water, dissolves.
  • So, salt has a disposition to dissolve in water.

Dispositional properties are contrasted with categorical properties, which are not manifested under specific conditions.

  • For example, salt’s property of being a sodium compound is categorical.
  1. Dispositional realism

Realists about dispositional properties justify their position via metaphysical inference: dispositions are alleged to best explain a number of things including:

  • Empirical generalizations,
  • The overlap between different realist positions,  
  • Scientific practices of explanation, extrapolation, and the use of inconsistent models.
  1. Dispositional antirealism

Some find dispositional properties too mysterious. They might even be realists about instances of dissolving salt while denying that there is a dispositional property, solubility, standing over and above these instances. Or they might hold that salt’s molecular structure (a categorical property) exists and that solubility is just a way of describing that structure, but once again, this doesn’t require an extra commitment to solubility.

  • Explanatory Power I: Unifying Scientific Realism

Chakravartty uses dispositional realism to provide a unified account of several theses and concepts in the metaphysics of science. As he notes, however, whether one is moved by this depends on one’s stance.

  • Unifying Entity and Structural RealismBackground.

Let’s begin with a brief history of scientific realism.

Scientific realism holds that our best scientific theories provide approximately true descriptions of both the observable and unobservable world.

Its chief challenge is the Pessimistic Induction, which roughly stated, runs as follows holds that today’s most successful theories are false because our most successful theories in the past were subsequently discovered to be false. Typical realist responses involve two moves:

  • To raise the standards of “most successful.” Two “high-grade successes” are:
    • Experimental intervention. Intervening to produce an effect in a well-designed experiment.
    • Novel prediction. Making a bold prediction, unique to the theory, that is borne out.
  • To narrow which parts of a theory are “approximately true.” Only those parts of the theory indispensable to the high-grade success warrant realist commitment.

This is called selective realism. The two most prominent kinds of selective realism are:

  • Entity realism. This holds that specific entities are real (typically the ones that figure in successful experimental interventions). It frequently denies that other parts of a theory—including many of the descriptions of entities—are accurate.
  • Structural realism. This holds that we should be realists about the relations between entities, but should be agnostic about the entities that stand in those relations. Furthermore, it is only the abstract (mathematical) structure of these relationships that gets preserved over time in science, and this is the only thing we should be realists about.
  • Unifying entity and structural realism.

On the face of it, entity realists should be pessimistic about structures, and structural realists should be skeptical about entities. But this needn’t be so. Chakravartty argues as follows:

U1.      If entity realism is true, then entities can be detected or manipulated.

U2.      If an entity is detected or manipulated, then scientists have knowledge of the detector or the manipulation’s causal powers.

U3.      Causal powers = dispositions.

U4.      If scientists have knowledge of dispositions, then dispositional realism is true.

U5.      So, if entity realism is true, then dispositional realism is true.

U6.      If dispositional realism is true, then there is a relation between that entity, the effect it produces when its causal power is manifested, and the conditions in which that power is manifested, and that relation is typically expressible in mathematics.

U7.      If there is a relation between that entity, the effect it produces when its causal power is manifested, and the conditions in which that power is manifested, and that relation is typically expressible in mathematics, then structural realism is true.

U8.      If dispositional realism is true, then structural realism is true.

Comment: This is a pretty unorthodox “unification.” Typically, A unifies B and C if B and C follow from A. (Think of how Newtonian mechanics gives a unified account of the tides and planetary orbits.) In this case, dispositional and structural realism follow from entity realism. So, entity realism should be the unifying framework!

Possible reply: The reasoning offered here doesn’t capture the fact in greatest need of explanation—namely how both entity and structural realism could simultaneously be true.

  • Unifying Properties, Causation, Laws of Nature, and Scientific Categories

C1.       Entities participate in causal processes.

C2.       That entities have disposition-conferring properties best explains why they participate in causal processes.

C3.       So, entities have disposition-conferring properties.

K1.      Science routinely succeeds in categorizing entities in ways that allow for reliable generalizations (laws) and predictions using those categories (natural kinds).

K2.      That entities have disposition-conferring properties best explains this success.

K3.      So, entities have disposition-conferring properties.

  • Stances Enter the Fray

These unifying arguments only have force if one subscribes to entity and structural realism (in §2.1) and causation, laws, and natural kinds.

Furthermore, one might be unimpressed by how empirically unconstrained (no “empirical answerability”) these explanations/unifications are.

This is a matter of one’s stance.

  • Explanatory Power II: Giving Scientific Explanations
    • Transcendental Argument for Dispositions: Big Picture

P1.       Science explains many phenomena.

P2.       That science explains many phenomena presupposes the existence of dispositional properties.

C.         Therefore, dispositional properties exist.

  • Dispositional Regress Argument
  • If C is a categorical property and D is a dispositional property, then for every explanation of D in terms of C, C causes D under some circumstances but not in others.
  • For all X and Y, if X causes Y under some conditions and not others, then X is disposed to cause Y.
  • So, every explanation of a dispositional property presupposes another dispositional property.
  • Dispositional Exercise Argument
  • It sometimes appears that no disposition is being exercised when there is good reason to believe that it is. (see electron example from Cartwright, p. 116)
  • The explanation of such phenomena presuppose counteracting dispositions.
  • Argument from Abstraction
  • Science abstracts away some parts of a system in order to explain how a system would behave if the remaining parts were the only operant causes.
  • So, abstract explanations presuppose dispositions.
  • Stances, Again

The dispositional antirealist will claim that these arguments only show the indispensability of dispositional language, not of dispositional properties.

Many of these arguments beg the question against the dispositional antirealist.

In many cases, disposition-talk can be swapped out for non-disposition-talk. For example, putative dispositions of an isolated system are not manifested in more complex systems. So, abstraction might merely be a useful tool for guiding research, but it need not be construed as describing dispositions.

However, are dispositions even necessary here? “Behaviors like [those of enzymes] are surely amenable to dispositional description, but they are also surely amenable, if one is that way inclined, to description simply in terms of entities with categorical properties.”

  • Explanatory Power III: Consolidating Scientific Knowledge
  • Science uses inconsistent models of the same phenomenon.
  • That these phenomena have dispositional properties (i.e., that they behave differently under different circumstances) best explains why science can use inconsistent models effectively.

This, too, runs into the typical stance-y treatments. Among other things, empiricists don’t think that theories need to accurately describe unobservables, so they won’t be bothered by models that are inconsistent about unobservables. (It will kind of be like using a Philips and flathead screwdriver under different circumstances.)

  • Property Identity

Chakravartty suggests that dispositional realism will only work with certain kinds of empiricism:

  1. Either dispositions determine a property’s quiddity, i.e., why that property is what it is, or quiddities are fundamentally unknowable.
  2. In science, quiddities are assumed to be knowable.
  3. So, if dispositional antirealism is true, then science regularly fails to produce knowledge of scientific properties.

Chakravartty suggests that dispositional antirealists who hold that scientists produce knowledge of scientific properties are guilty of self-sabotage. A more promising dispositional antirealist should deny that questions about quiddities should be answered.

Naturalism and the Grounding Metaphor

  • In Hopes of Demarcating Scientific Ontology

What distinguishes scientific ontology from non-scientific ontology? Chakravartty proposes the norm of naturalized metaphysics (NNM):

…the principle that scientific ontology is properly delimited by metaphysical inferences and propositions that are sufficiently informed by or sensitive to scientific-empirical investigation as to provide or constitute ontological knowledge relating to the sciences. (67)

However, consonant with his voluntarism, there can be substantial differences about what counts as “sufficiently informed by or sensitive to scientific-empirical investigation.” As he puts it, “no stance in, no ontology out” (65-66).

            At least one place where these differences play out is in delimiting the subject matter of the sciences that is germane to scientific ontology. There are at least two places where this affects how the NNM is applied:

  • The explicit subject matters of the sciences are things that can fall out of reading the sciences at face value. For instance, molecular biology’s explicit subject matter includes gene transcription, DNA, and RNA.
  • The implicit subject matters of the sciences, “things whose natures are not the face-value targets of scientific work, but which are rather mentioned in passing” (69). This includes properties, causal relations, laws of nature, possibilities, and necessities.

Here are some of the questions that arise which Chakrvartty appears to think are stance-dependent. (1) Should scientific ontology restrict itself to explicit subject matters? (2) If scientific ontology is not restricted to explicit subject matters, then which implicit subject matters are proper targets of analysis? Different scientific ontologists may diverge in their applications of the NNM in answering these questions.

  • On Conflating the A Priori with That Which is Prior

Some claim that science presupposes metaphysics; others, such as Chakravartty, speak of metaphysical inferences. What’s at stake in this distinction?

A scientific domain D presupposes a metaphysical claim M if scientific investigation in D would not be possible without M.

  • Examples: measurement presupposes that quantities exist. Chakravartty also mentions classical mechanics presupposing that physical spaces obeys the axioms of Euclidean geometry as another example.

Some who appear skeptical of metaphysical inference accept that science frequently presupposes metaphysics. (Who? Chakravartty does not say.) How is this a coherent position? Chakravartty three possibilities:

  • First Possibility: “Metaphysical” presuppositions aren’t really metaphysical

The first possibility runs as follows:

P1.       Metaphysical presuppositions are not a priori.

P2.       The conclusions and criteria of evaluation in metaphysical inferences are a priori.

P3.       Only a priori claims are “really” metaphysical.

C.         So, metaphysical inferences, but not metaphysical presuppositions, are “really” metaphysical.

Chakravartty criticizes P1, arguing that presuppositions cannot be directly observed. “it was not because the geometry of spacetime was somehow empirically detected to be non-Euclidean that Einstein ushered in a new way of thinking about spacetime with his theory of general relativity.” (74)

  • Note: Chakravartty isn’t very clear about this, but I think the implicit point seems to be that there is a metaphysical inference from the predictions of relativity theory to the metaphysical presupposition that space is non-Euclidean.
  • Second Possibility: Presupposed metaphysics is less problematic than inferred metaphysics, version 1

Here’s the second way of cleaving metaphysical presuppositions from metaphysical inferences:

P1.       Metaphysical presuppositions do not concern ontology.

P2.       Metaphysical inferences do concern ontology.

P3.       Only metaphysics that concerns ontology is problematic.

C.         So, metaphysical inference, but not metaphysical presupposition, is problematic.

Chakravartty argues that the most plausible way of making P1 true is to adopt a deflationary metaphysics, in which putatively metaphysical claims are really about something non-metaphysical (for instance, merely about social practices in the scientific community.) However, deflationists will also reject P2. So, deflationism cannot fund this argument.

  • Third Possibility: Presupposed metaphysics is less problematic than inferred metaphysics, version 2

P1.       Metaphysical presuppositions are often tacit.

P2.       Metaphysical inferences are often explicit and deliberate.

P3.       Only explicit and deliberate metaphysics is problematic.

C.         So, metaphysical inference, but not metaphysical presupposition, is problematic.

The problematic assumption here is P3. Why think that something’s being explicit is problematic and something’s being tacit is not?

  • How Not to Naturalize Metaphysical Inferences

What exactly is the relationship between science and scientific ontology, such that the latter can be distinguished from other “non-scientific” ontologies? Chakravartty considers two proposals.

  • The Heuristic Conception

This approach to naturalized metaphysics, associated with Quine, sees philosophy as doing important conceptual preparatory work before handing off a topic of research to the empirical sciences.

Chakravartty objects that the heuristic conception is always out of time, as it were. We would have no way of knowing whether we were doing scientific ontology now, since it could only be redeemed by future science.

  • The Continuity Conception

Philosophy should be continuous with science’s aims, methods, subject matters, and criteria of evaluation.

Chakravartty thinks that this is more or less correct, save for the continuity of external subject matters with internal ones that are more philosophical in nature. Roughly, this means that science’s claims about explicit subject matters “ground” scientific ontology’s claims about its implicit subject matters. As Chakravartty notes, this grounding metaphor is in need of unpacking.

For instance, there is not a tidy division of labor with science (in say, its explicit subject matter) providing a posteriori constraints on ontology’s a priori claims. As we’ve already seen, science itself is fraught with a priori claims.

  • Unpacking the Metaphors: “Grounding” and “Distance”

The “ground” of scientific ontology is empirical inquiry.

“Distance” from this ground can be construed in terms of epistemic risk—given the empirical inquiry in question, what is the probability that the conclusion drawn from it (via metaphysical inference) is false?

Epistemic risk is a function of two things:

  • Empirical vulnerability: “how susceptible a proposition is to empirical testing.” (85)[1]
  • Explanatory power: “a measure of how well a metaphysical inference or resulting proposition satisfies the criteria typically associated with good explanations of the data of observation and experience,” such as “simplicity, internal consistency, coherence with other knowledge, and the capacity to unify otherwise disparate phenomena.” (87)

The more empirically vulnerable a statement is, the less its explanatory power and its epistemic risk.

These pull us in opposite directions: greater empirical vulnerability is good but greater explanatory power is also good. So, they trade off each other.

  • Theorizing versus Speculating

Naturalized metaphysicians frequently pride themselves on doing something akin to high-level scientific theorizing. They contrast this activity with the speculation characteristic of more traditional metaphysics.

Chakravartty contends that “There is no objective distinction between theorizing and speculating in the context of scientific ontology.” (89)

He argues for this as follows:

P1. If an objective distinction between theorizing and speculating exists, then there is a fact of the matter about the appropriate level of epistemic risk, i.e., the balance between testability with explanatory power, when drawing inference from the empirical content of science.

P2. There is no fact of the matter about the appropriate level of epistemic risk when drawing inference from the empirical content of science.

C.   Therefore, no objective distinction between theorizing and speculating exists.

The contentious premise is P2, so he offers some examples to motivate it. The only one that he really discusses in any detail is novel prediction. A novel prediction is an unexpected (typically precise) prediction that turns out, upon subsequent investigation, to be true. When a theory makes a novel prediction, its empirical vulnerability increases, so the epistemic risk in accepting it decreases. Does this mean that the following is true?

All and only theories that make novel predictions exhibit an acceptable level of epistemic risk.

If so, there would be a fact of the matter about the appropriate level of epistemic risk (P2 in the previous argument would be false.) However, Chakravartty argues that many good theories do not make novel predictions but are explanatorily powerful (natural selection). So, it cannot be that the absence of novel predictions can be used to rule out some metaphysical approaches as non-scientific ontologies.

[1] A more standard word for this is “testability.” This more conventional word-choice seems preferable in my opinion, since, for Chakravartty, epistemic vulnerability is a good a thing, yet there are many cases in ordinary language in which vulnerability is a bad thing (vulnerability to attack, for example).

Science & Metaphysics, Then and Now

Note to students: As I mentioned in class, this is written more in the form of a handout for you. Your own summaries might benefit from being a bit more prose-y, as my first post and Farhan’s post were. Additionally, the discussion questions here are simple: there are a battery of arguments. As you read through them, ask yourself, “Do I accept these arguments? If not, which premise do I reject?”

Main Objective: Chakravartty seeks to refute the claim that some conceptions of scientific ontology require no metaphysical inferences. More positively: all conceptions of scientific ontology require metaphysical inferences.

  1. What are metaphysical inferences?

Metaphysical inferences are a species of Inference to the Best Explanation (IBE). This inference pattern is typically represented inductively:

            P is a fact in need of explanation.

            Q is the best explanation of P.[probably]


Roughly, an explanation best explains a phenomenon if and only if it optimizes the following kinds of criteria: simplicity, internal consistency, coherence with already-established science, scope, and unification.

Unlike other IBE’s, metaphysical inferences have pronounced a priori dimensions. Chakravartty provides three indications of what he means by this (35):

  • The explanatory hypothesis, Q, is the product of conceptual analysis.
  • The explanatory hypothesis, Q, describes a state of affairs that cannot be observed through the use of the unaided human senses.
  • The criteria that make one explanation better than another are defined and weighted by one’s intuitions.

Philosophers are likely to disagree about whether a particular IBE counts as a metaphysical inference, owing to the following sticking points: How strictly is “a priori” being defined?

  • For instance, if its antonyms, “a posteriori”or “empirical”, are restricted to claims that are knowable by the unaided human senses, then metaphysical inferences are legion.
  • However, there are more liberal notions of “a posteriori” that include any claims that are:
    • “Sensitive to” observation or
    • Knowable via the outputs of instruments such as microscopes and telescopes.

This more liberal notion of “a posteriori” will have a more restricted set of IBEs as metaphysical inferences.

2. Is Science Inherently Metaphysical?

Because science emerged from philosophy and theology, there is no doubt that earlier science is fraught with metaphysical inferences. However, what of contemporary science? Chakravartty looks at the arguments for and against.

2.1. Anti-Metaphysics: Empiricism

E1.       If some metaphysical inferences are sound, then we should accept some claims that posit unobservable entities.

E2.       We should not accept any claims that posit unobservable entities.

E3.       Therefore, no metaphysical inferences are sound.

2.1.1. Chakravartty’s “Zombie Argument” against E2

Z1.       If empiricism is true, then we should believe that science aims only at knowledge of the observable.

Z2.       Aims are unobservable.

Z3.       Therefore, if empiricism is true, then we should believe some claims that posit unobservable entities.

In other words, if empiricism is true, then E2 is false and the empiricists’ ban on metaphysical inferences is unsupported.

2.1.2. Chakravartty’s “Empirical Metaphysics Argument” (55-57)

EM1.   If empiricism is true, then observable entities must be distinguishable from both unobservable entities and “observable non-entities” (such as hallucinations, optical illusions, etc.)

EM2.   In at least some cases, distinguishing observable entities from unobservable entities and observable non-entities requires sound metaphysical inferences.

EM3.   Therefore, if empiricism is true, then at least some metaphysical inferences are sound.

2.2. Pro-Metaphysics

M1.      If successful science makes metaphysical assumptions, then some metaphysical inferences are sound.

M2.      Successful science makes metaphysical assumptions.

M3.      Therefore, some metaphysical inferences are sound.

2.2.1. Chakravartty’s “Reinterpretation Argument” against M2

R1.       If any putatively metaphysical assumption can be reinterpreted as an empirical assumption, then successful science need not make metaphysical assumptions.

R2.       Any putatively metaphysical assumption can be reinterpreted as an empirical assumption.

not-M2.   Therefore, successful science need not make metaphysical assumptions.[2]

2.3. Some Thoughts…

While it is historically accurate to describe empiricists as subscribing to a thin notion of the “empirical,” it is not clear why this is a necessary aspect of empiricism.

Furthermore, many empirical claims do not need to be justified via IBE of any sort. Presumably, they can be justified by observation.

If we combine these two points, then a liberal notion of the empirical provides a stronger anti-metaphysical argument than the one presented in §2.1: lots of things are observable and many of them can be known without any appeal to IBE. The remaining question would be if those that can only be known by IBEs are metaphysical inferences.

3. Epistemic Stances Regarding Scientific Ontology

Chakravartty restates some of his core ideas about stances, but emphasizes a few things that were understated in Chakravartty (2018):

  • Stances are non-propositional.
    • Propositions can be true or false and can be believed. Stances are neither. Instead of being believed, stances can be “adopted.”
  • They are orientations, clusters of attitudes, commitments, and strategies “relevant to the production of allegedly factual beliefs.” (p. 47; This seems a bit broader than an epistemic policy, but perhaps not.)
  • They can coexist with each other:

Where someone adopting the metaphysical stance may be tempted to affirm certain propositions regarding unobservable objects, events, processes, and properties, the holder of the empiricist stance does not deny such claims. Rather, the empiricist simply has no beliefs at all concerning such propositions… In this way the latter’s scientific ontology may be a subset of the former’s… (50)

4. “Metaphysical inference” versus “metaphysical inference”

Consider two IBEs:

Numbers IBE
The applicability of mathematics to science is a fact in need of explanation.
Numbers’ existence best explains why mathematics is applicable to science. [probably]
Numbers exist.
Gravity IBE
That apples fall is a fact in need of explanation.  
Gravity’s existence best explains why apples fall. [probably]
Gravity exists.

While most philosophers would agree that the “Numbers IBE” is a metaphysical inference, they might well disagree about whether the “Gravity IBE” is a metaphysical inference. Those who claim that Gravity IBE is non-metaphysical might claim that, at most, it is a metaphysical inference, but unlike the Numbers IBE, it is not a Metaphysical inference.

One might think that once we can pinpoint the “objective boundary” between little-m and Big-M metaphysical inferences, we can resolve some debates about scientific ontology. Chakravartty disagrees:

There is no boundary. Rather, there is something like a continuum ranging from lesser to greater magnitudes of metaphysical inference, and disagreement regarding which parts of this continuum are epistemically solid enough to serve as a basis for scientific ontology” (52).

5. Metaphysics without Scientific Ontology?

Most “professional” metaphysicians are not philosophers of science. (They are frequently called “analytic metaphysicians.”) They see themselves as doing something: (a) more “general” or “fundamental” than scientific ontology—but they also see what they’re doing as: (b) independent of scientific ontology. Chakravartty argues that these two claims are in tension.

            Specifically, if metaphysics is more general/fundamental, then it must show that it can subsume more specific/derivative scientific claims under its framework. However, this means that it depends on scientific ontology.

Chakravartty, Anjan. 2018. “Realism, Antirealism, Epistemic Stances, and Voluntarism.” In The Routledge Handbook of Scientific Realism, edited by Juha Saatsi, 225-236. Routledge.

[1] However, it can also be represented deductively:

It is rational to believe the best explanation of a fact in need of explanation.

P is a fact in need of explanation.

Q is the best explanation of P.

Therefore, it is rational to believe that Q.

Indeed, this might help us to more clearly see where stances come in—the “rational-to-believe parameter.”

[2] Note the switch to “need not” in this claim from the implication that successful science “(actually) makes” metaphysical inferences in the original pro-metaphysics argument.

Chakravartty: Week 1

Chakravartty (2017, Ch. 1, 2018)

Chakravartty pitches voluntarism as a response to the scientific realism debates. In particular, he sees voluntaristism as an explanation (and potential escape) from the seeming intractability of the scientific realism debates.

Background: The Scientific Realism Debates

Chakravartty offers voluntarism as a framework for diagnosing the seeming intractability of debates about scientific realism. In the philosophy of science, scientific realism is understood as a doctrine consisting of three claims:

  1. A metaphysical/ontological claim: The physical world exists independently of our minds.
  2. A semantic claim: Scientific theories should be interpreted literally, not as “secretly” being only about observable reality.
  3. An epistemological claim: Our best scientific theories provide approximately true descriptions of both observable and unobservable reality.

Antirealists deny at least one of these claims. Chakravartty focuses on those antirealists who deny either the semantic or epistemological claim. Critics of the semantic claim include deflationary views which “recast the study of what things exist, and what they are like, in terms of something else” (2017, 9), and instrumentalists, i.e., empiricists who “hold that scientific descriptions of unobservable things… are meaningless, strictly speaking” (2017, 17).

Most of the debate in the last fifty years or so concerns the epistemological thesis. Chakravartty focuses on constructive empiricism, which is typically taken to deny only the epistemological thesis. Specifically, constructive empiricists hold that our best scientific theories only provide approximately true descriptions of observable reality but are not (or at least need not be) accurate with respect to unobservable reality.

            The most prominent argument for the epistemological thesis is what is called the No Miracles Argument:

NM1. Science is empirically successful.

NM2.  The best explanation of science’s empirical success is that its best scientific theories provide approximately true descriptions of unobservable reality.

NM3.  For all x and y, if x best explains y and y is true, then (probably) x is also true.

SR.     So (probably) our best scientific theories provide approximately true descriptions of both observable and unobservable reality, i.e., the realist’s epistemological claim is true.

The most prominent argument against the epistemological thesis is what is called the Pessimistic Induction:

PI1.    Our best scientific theories in the past have turned out not to provide approximately true descriptions of unobservable reality.

not-SR.   So (probably) our best scientific theories in the present do not provide approximately true descriptions of both observable and unobservable reality, i.e., the realist’s epistemological claim is false.

Evaluate the No Miracles Argument and the Pessimistic Induction. Which premises do you accept and which do you reject?

Scientific Realism Debates’ Inevitability

One might think that scientific realism debates are misguided. After all, what better guide can we have to knowing whether our scientific theories provide accurate descriptions of the world than from science itself? Chakravartty (2017, 31) argues that “scientific ontology is underdetermined by the sciences and case studies thereof.” Reviewing three prominent kinds of scientific realism—explanationist, entity, and structural—he shows that while attention to scientific practice is necessary for adjudicating between different philosophical positions, it is not sufficient. In each of these cases, further philosophical argumentation about historical interpretation (for all three realist positions), language (in the case of entity realism), and structure (in the case of structural realism) is needed.

Do only realist positions face these problems? Chakravartty does not say much on this front.

Scientific Realism Debates’ Intractability

So, it appears that philosophical issues cannot be avoided. Why think these debates are intractable or irresolvable? After all, many philosophical debates are “perennial.” Chakravartty (2018) offers three reasons for this intractability (2018, 226). First, in addition to the “ground-level” disagreement—between, say, the realist’s epistemological claim and its empiricist counterpart—there are also differences in meta-philosophical commitments or “epistemic stances.” Second, these stances “concern how these ground-level positions, realism and antirealism, are formulated as views about scientific knowledge and how they are evaluated.” Third, whether realism or antirealism is correct is decidable only after one has adopted a particular stance. So, the debate bottoms out in these meta-philosophical differences in stances. If this is correct, then arguments such as the No Miracles Argument and Pessimistic Induction cannot possibly resolve the debate, since they aren’t resolving the different stances that are the true drivers of the debate.

Epistemic Stances

So, what exactly are stances? After reviewing some options, Chakravartty settles on the idea that stances are “epistemic policies… regarding which principles and methodologies are appropriate or inappropriate to producing knowledge” (227). Thus, epistemic stances bear directly on the debate about realism’s epistemological thesis.

            Chakravartty (2018, 229) considers three stances: the metaphysical, empiricist, and deflationary stances. For purposes of illustration, let’s consider the first:

The Metaphysical Stance:

M1      Accept demands for explanation in terms of things underlying the unobservable.

M2      Attempt to answer these demands by theorizing about the unobservable.

It is clear why realists, many of whom adopt the metaphysical stance, are drawn to the No Miracles Argument: that argument presupposes M1 and M2. By contrast, neither empiricists nor deflationists share the metaphysical stance’s epistemic policy regarding demands for explanation, so we can now see why they are unmoved by the No Miracles Argument.

            Furthermore, as M1 and M2 illustrate, epistemic policies are not descriptions that can be true or false. Rather, they are imperatives. This goes some way to accounting for the realism debates’ intractability. It is not altogether clear how one argues that one of these imperatives is better than another without begging the question. Following van Fraassen (2002), Chakravartty identifies two criteria of assessment. First, stances should not admit of “self-sabotage by one’s own lights,” meaning that an epistemic policy should not undermine the achievement of the very epistemic aims one seeks to fulfill. This is how van Fraassen and Chakravartty conceive of rationality. Second, one’s stance should reflect what one values epistemically.

Both of these aspects of stances are permissive, meaning that they allow for a multiplicity of rational stances. For instance, the metaphysical stance reflects a commitment to the rationality or epistemic value of accepting certain kinds of explanation, while the empiricist stance does not. Yet, both of these two stances are immune to self-sabotage and reflect their proponents’ core epistemic values, so both realists and empiricist stances seem legitimate.

As Chakravartty notes, some people find these constraints on stances to be too permissive. What do you think?


To be a voluntarist about x is to claim that x can be freely chosen. In epistemology, the controversial doctrine is voluntarism about beliefs. In short, many (if not all) of our beliefs do not seem to be things that we can directly control. (Think, for example, of your beliefs about what you see.) However, voluntarism about stances seems more plausible.

            Chakravartty then notes that one can rationally choose a stance by: (i) surveying the range of stances that are rational, i.e., non-self-sabotaging and (ii) adopting the stance that best accords with one’s epistemic values. From this he notes, “minimally, one can say that a good choice of stance is one that incorporates these two aspects, a negligent choice is one that neglects one or both, and a poor choice is one that for reasons of mistaken assessment or negligence results in the adoption of a stance that fails to pass the test of rationality or value matching” (2018, 234).

Is it really necessary for stances to be voluntarily chosen, or is it enough that different stances are permissible? What’s gained by the voluntaristic move?

Chakravartty, Anjan. 2017. Scientific ontology : integrating naturalized metaphysics and voluntarist epistemology. Oxford: Oxford University Press.

Chakravartty, Anjan. 2018. “Realism, Antirealism, Epistemic Stances, and Voluntarism.” In The Routledge Handbook of Scientific Realism, edited by Juha Saatsi, 225-236. Routledge.

van Fraassen, Bas C. 2002. The empirical stance. New Haven: Yale University Press.