Scientific Realism and Blocking Strategies*

ABSTRACT

My target is the epistemological dimension of the realism debate. After establishing a stance voluntarist framework with a Jamesian background, drawing mostly on Wylie, Chakravarty, and van Fraassen, I argue that current voluntarists are too permissive. I show that especially various anti-realist stances but also some realist and selective realist stances block themselves from refutation by the history of science. I argue that such stances should be rejected. Finally, I propose that any disagreement that cannot be resolved by this strategy frequently boils down to an epistemic value disagreement about balancing the truth-goal.

1. Introduction and the Stalemate

This paper concerns the epistemological dimension of the scientific realism debate. A short explication about what that amounts to is due. One needs to differentiate between characterising being an epistemological scientific realist and epistemological scientific realism. I understand being an epistemological scientific realist as being committed to believing that unobservable¹ entities and/or structures, as put forward by a proper subset of our scientific theories, exist. Epistemological scientific realism is the position that

1. those beliefs have some kind of rational standing (e.g. justification, warrant, reasonableness)² based on scientific investigation and

2. contending further that such rational standing is strong enough to epistemically obligate one to believe in said entities and/or structures.³

Chakravarty classifies the scientific realist similarly when saying ‘it is the naked fact that she believes that makes her a scientific realist’ (Chakravartty and van Fraassen 2018, 23). This is a classification with quite a low demand. Suppose I tell you to believe in electrons, but knowingly my belief is a pure leap of faith, then I am a minimal realist in this sense, but my realism does not have any strong epistemic commitment. In fact, I openly do not have any epistemological standing towards my belief.⁴ As such, the important epistemological questions are (i) and (ii)—the ones about norms of belief.

My explication of scientific realism in terms of some form of justification is common⁵ but it is even more common to explicate the rational standing in terms of knowledge (e.g. Boyd 1983; Hacking 1983; Ladyman 1998; Chakravartty 2017) even though I think this is quite problematic because it packs the goal (true believing) and the methods (justified believing) into one complex concept.⁶ Still, for the purpose of this article, I will not have any quarrels with ‘knowledge’ either.

Condition (ii) is to exclude such constructive empiricists from the realist camp who believe in unobservables but also believe that there is no rational obligation for this belief. One objection to condition (ii) might be that it also excludes some voluntarist realists, such as Chakravartty, so maybe (ii) is too strong. I will talk about voluntarism later. For now, this should make my target sufficiently clear.

Today, many philosophers of science think that the realism debate is at a stalemate or even a dead end (cf. Blackburn 2002; Magnus and Callender 2004; Monton 2007, 3) and this sentiment is nothing new (cf. Fine 1986, 136–150; Nagel 1961, 145; Wylie 1986, 287). Why is this the case? We should distinguish three dimensions where the disagreement can be stuck:

1. The Synchronic Evidence Dimension: empirical evidence in support for a given scientific theory

2. The Diachronic Evidence Dimension:

   1. Historical Evidence: present empirical evidence from episodes in the history of science about theory change

   2. Prospective Evidence: yet unavailable evidence from historical scientific episodes and future scientific developments about theory change

3. The Value Dimension: considerations about attitudes, commitments, values, epistemic standards, and epistemic goals

Note first that the application of synchronic evidence to the realism debate is very limited. It is the main concern for working scientists, but the philosophical debate goes a different route with some exceptions.⁷ This is simply the case because in the current discussion all rival accounts are constructed for synchronic empirical adequacy.

Next, consider diachronic evidence. Historical evidence is often associated with the Pessimistic Induction [PI].⁸ The interesting feature of it is that, contrary to the synchronic evidence dimension, it cuts through a wide variety of rival accounts. Here, at least to some extent, the disagreement is about what the evidence shows. This is a healthy aspect of the debate because it allows evidence-based comparisons.⁹ Still, various rival accounts claim the historical track record being on their side.¹⁰ In this paper, I argue that one can distinguish positions in the realism debate that are open to refutation by historical evidence or by the future progress of science, and those that are not open to such refutation, and we should reject all those positions that are not open to such refutation.

Before I get to this main point, I need to say a few words about the value dimension.

2. The Jamesian Truth-Goal, Stances, and Voluntarism

An important starting point for discussions of the value dimension is Alison Wylie’s (1986) ‘Arguments for Scientific Realism: The Ascending Spiral’ . She explicates the disagreement in the realism debate with the notion of epistemic risk—the risk of being in error. According to Wylie (cf. 1986, 290), anti-realism is based on a meta-philosophical decision to reduce epistemic risk because of the possibility of global error while realists are more risk-tolerant. When we talk about any risk, the question arises: Why are we risking something? Risk implies a trade-off. Wylie is not very clear about what is to gain for the risk seeker. One way to conceptualise the trade-off is to import the general thesis by Chris Swoyer, captured in what he calls the fundamental ontological trade-off:

This is the trade-off between explanatory power, on the one hand, and epistemic credibility, on the other; between a rich, lavish ontology that promises a great deal of explanatory punch, and a more modest ontology that promises more epistemological security and believability. (Swoyer 2008, 29)

I think Wylie had something similar in mind. Making it more explicit with Swoyer’s thesis gets us the following picture: Realists choose a higher epistemic risk. They potentially get an increase in explanatory power by sacrificing epistemic security. Anti-realists choose a lower epistemic risk. They gain epistemic security but sacrifice explanatory power. I think this is the most common way, especially on the realist side, about how to embed the realism debate in an epistemological trade-off. For instance, we find it explicitly in Forbes (2017); in Park’s defense of scientific realism: ‘Believe less. Explain less’ (Park 2017, 113); and predated to some degree by Van Fraassen (1989, 2002) and Chakravartty (2007, 2013).

Furthermore, the value dimension is intimately connected with the discussion of stances. A precursor of this idea is Carnap (1937, 3). According to Carnap and van Fraassen, empiricism is not a set of beliefs but rather something non-doxastic, non-cognitive, a collection of ‘attitudes, commitments, values, goals’ (Van Fraassen 2002, 48). As Chakravartty (2017, 47) puts it: ‘A stance is an orientation, a cluster of attitudes, commitments, and strategies relevant to the production of allegedly factual beliefs’. Teller (2004) introduced the analogy between stance and epistemic policy; Psillos (2021, 17) calls a stance a ‘blueprint for an epistemic policy’.

Recently, Chakravartty (2018, 232) articulated the realism debate in terms of the Jamesian ([1896] 2013) truth-goal (henceforth the truth-goal). It is characterised by two sub-values: believing truths and avoiding error.¹¹ One central feature of that characterisation is that those two parts pull in opposite directions. The more we value avoiding false beliefs at the expense of believing truths the higher our epistemic standards rise and the more often we withhold believing. The reverse gets us lower epistemic standards and a decreased likeliness to withhold believing. This can be called the fundamental epistemic trade-off and I will call the choice of how to do the trade-off balancing. I spelled the picture here out in terms of ‘full truth’.¹²

How does the Jamesian picture translate to the realism debate? Suppose one of our best-supported scientific theories shows prima facie an ontological commitment to electrons. Suppose further that electrons exist. Then realists will believe an (interesting and relevant) truth. On the other hand, suppose that electrons do not exist. Then realists will believe a falsehood. Presupposing that our current best scientific theories will be sometimes correct about unobservables and sometimes incorrect, realists will believe more interesting and relevant truths but also more falsehoods than anti-realists. This picture suggests that realists are more risk-tolerant, anti-realists more risk-averse, and selective realists try to walk the line in between. Stance choice depends on risk choice.

Recently in philosophy of science, Khalifa (2020) argued that one can reduce all epistemic goals of science to the veritist goal. Note that this commitment is very minimal. For instance, van Fraassen is committed to it by him aiming at truth about observables. Note further, that, more precisely, the goal has to be believing relevant truths and avoiding error. After all, irrelevant truths are not epistemically valuable (cf. David 2001; Khalifa 2020). Adding a relevance condition, introduces some further complication to the Jamesian trade-off. If by believing that p I can expect to believe a relevant truth within some acceptable range of error, then believing is at least epistemically permissible. If the payoff for believing that p is only an irrelevant truth, then the potential for error will make believing that p not epistemically permissible.¹³

Since I adopt the Jamesian trade-off, I need to respond to a likely objection. The realist usually adopts a notion of ‘approximate truth’ and not ‘full truth’. As such, since approximate truth involves some kinds of error, must there not be a trade-off between acceptable and unacceptable error instead? Following Musgrave (2007), my view is the following. If S tells me that they believe that theory T is approximate truth, then S should tell me instead in which limit, in which domain, or under what conditions they believe it is fully true. If they cannot do that, then they are not clear about what they believe. Suppose some S would tell you that the caloric theory is actually approximately true. What does S believe? What S should believe (given current science) is something like the following: Some parts or implications of the theory are fully true. For instance, one implication of the theory is that a cup of hot tea cools down in a cold room. This S believes. Some parts were fully false. For instance, there is a fluidal substance of heat that passes from warmer to colder bodies. This is fully false, and S does not believe it. Believing in the ‘approximate truth’ of the caloric theory means believing in the full truths of various parts of the theory and its implication. Realists believe that these parts are fully true but also believe that sometimes they are in error. The trade-off is between full truth and full error.

One can call the Jamesian picture a meta-stance. It is a general framework of instrumental rationality that embeds different stances. One picks out a specific stance in the Jamesian picture if one (i) commits to a specific balancing of the truth goal, and (ii) policies of how to achieve it.

With James, ‘decreasing epistemic risk’ means putting more weight on avoiding error. ‘Taking more epistemic risk’ means putting more weight on believing truths. Going back to Chakravartty for how to do balancing, he calls for permissivism (Chakravartty 2018, 233) and concludes to ‘subtract the usual judgment that at most one party to these disputes is, in fact, correct’ (Chakravartty 2018, 232). As such, Chakravartty ends up at a very permissive attitude,¹⁴ resembling some of van Fraassen’s voluntarism. This is straight up implied by the Jamesian picture since epistemic reasoning must underdetermine how much epistemic risk one is willing to take. The purely epistemological point of view is concerned with instrumental rationality, i.e. whether one takes the right means to one’s epistemic ends. This reveals that epistemology cannot reach out beyond itself to provide epistemic reasons concerning the balancing of the truth-goal itself. As such, I largely agree with Chakravartty’s permissivism and stance voluntarism on the value dimension of the debate. However, while agreeing with such voluntarism, I argue that Chakravartty and van Fraassen are too permissible about stance choice. Specifically, I argue that stances that block themselves from refutation by diachronic evidence, such as constructive empiricism and some version of classical realism, should be dismissed for more responsive alternatives.

3. Historical Evidence

For this, I want to analyze the function of historical evidence for the scientific realism debate. This pertains to Pessimistic Induction arguments (PI) (Laudan 1981). It also pertains to Unconceived Alternative arguments (UA) (Stanford 2006). UA arguments are based on the historically informed claim that there are frequently unconceived alternatives to successful scientific theories (Stanford 2006) or models, or explanations (Rowbottom 2016). As such, if the unconceived relevant kind of alternatives can radically alter confirmation values, then it is questionable to believe what our current best theories say about unobservables. Note that the novelty of UA is its unique combination of underdetermination and PI arguments. UA is not merely a thesis about some logical possibility of underdetermination, as frequently found in classical underdetermination arguments. It is a historically informed thesis, a new ‘induction over the history of science’ (Stanford 2006, 19).¹⁵

Both PI and UA are error-rate-type arguments with an empirical basis informed by the history of science. The attitudes of anti-realists towards historical evidence can be grouped into three categories: irrelevancy, asymmetricity, symmetricity. I will argue that only symmetricity is the appropriate attitude.

1. Historical Evidence is Irrelevant

Van Fraassen states: ‘I am quite proud never to have relied on the so-called Pessimistic Induction’ (Van Fraassen 2007, 347). It should surprise no one that constructive empiricists, and anti-realists in a similar spirit, do not base their arguments on the evidence from the history of science. After all, the point of constructive empiricism is that the aim of science is empirical adequacy. Starting from an empiricist stance, there cannot be any rational compulsion to go beyond the observable, so the argument goes, and thus no amount of empirical evidence, be it from the history of science or from current science, could ever change anything about this.

Concerning underdetermination arguments, Van Fraassen (2007) is likewise dismissive. In fact, Rosen (1994, 160–161) argues that the constructive empiricist is well advised not to use underdetermination arguments, given the underdetermination of scientific theories concerning empirical adequacy until now vs. full empirical adequacy (i.e. also in the future). As a result, Monton and Mohler (2021, section 2.1) calls both PI and underdetermination ‘poor arguments for constructive empiricism’. For the constructive empiricist, historical evidence is irrelevant—it can neither disprove anti-realism nor undermine realism.

2)	Historical Evidence is Epistemically Unsymmetrical

Other anti-realists use PI against the realist, while simultaneously blocking an optimistic version against themselves. A symmetrical treatment suggests itself especially when treating PI as a standard form of induction. If the empirical basis rarely shows continuities of unobservables in successful theories, we infer that our current successful theories likely will change too and believing in their claims about unobservables is then judged as unjustified. If, however, there are mostly¹⁶ continuities of unobservables of the relevant category¹⁷ in theories of the relevant type (e.g. mature and predictively successful theories), then we infer that we are justified in believing.

Already very early (arguably in Laudan 1981), however, PI was not treated as a standard form of induction. Historical evidence was supposed to be a heuristic collection of examples about why various realist claims are unjustified, as for instance the claim that only truth explains success, and Stanford (2000) treats his UA the same way. This move for PI got even more prominent after it was contended that it rests on fallacious reasoning, such as a base-rate fallacy (cf. Lewis 2001; Magnus and Callender 2004). Responding to this charge by following, for instance, Saatsi (2005), anti-realists got an additional reason to not think about PI as a standard form of induction, but rather as a heuristic collection of case studies to criticise the success-truth pathway.¹⁸ With this line of argument, anti-realists do not need to establish any sense of the error-rate. All they need to show is a few relevant examples of entity or structural discontinuities in the relevant type of theories and the relevant category of unobservables (PI), or a few examples where unconceived alternatives actually superseded successful theories (UA).

With this version of PI and UA, it is easy to block the optimistic side while still maintaining the pessimistic side. The idea of one branch of empiricist anti-realist philosophies of science always was to not allow for any claims about anything that goes beyond their preferred empirical basis: Stay at the phenomena!, Stay at observables!, Stay at experience!, Stay at sense data!, etc. Historical evidence could not possibly amount to a justification of anything that goes beyond that. After all, such anti-realists contend precisely that there cannot be such empirical evidence, and thus they categorically block themselves from refutation by the history of science. Such anti-realists use PI and UA against success arguments by realists. An optimistic induction against such anti-realists, on the other hand, is supposed to be futile.

3)	Historical Evidence is Epistemically Symmetrical

There are other anti-realists, however, the ones who were motivated by historical evidence to begin with, who view PI and UA as symmetrical. They are anti-realists because they think that the historical track-record does not sufficiently support the belief in unobservables even of our best current theories. Note that anti-realists of this type would change their stance if the historical track-record turned out to be better than they think. Such responsiveness is not in place for anti-realists of type (1) and (2). Arguably Kuhn would fall into category (3). He had a cumulative realist picture in his The Copernican Revolution (1957) but switched to various anti-realist conclusions in his Structure (1962) precisely because of his discovery of discontinuities in the history of science. Mizrahi (2015) conceives an optimistic and a pessimistic version of UA but argues that current historical evidence cannot judicate between them. Various selective realists base their view on a symmetrical treatment of historical evidence. For instance, Worrall (1989) argues from historical evidence of discontinuities for anti-realism concerning entities but conversely argues from continuities to realism about structures. Similarly, Kitcher (1993), and Psillos (1994, 1999) conclude realist optimism for various stable, invariant portions of theories, but anti-realist pessimism for the rest, and Chakravarty (1998, 2007) follows a similar strategy for detection vs. auxiliary properties.

4. Steadfast Anti-Realism

I want to pick out constructive empiricism specifically, to introduce the concept of a ‘blocking strategy’. Constructive empiricists contend categorically that there cannot be empirical evidence for unobservables sufficient for belief obligations, and thus categorically block themselves from refutation by synchronic and diachronic evidence. Such anti-realist positions can be labelled Steadfast Anti-Realism.

Consider the following thought experiment to make the steadfastness of the meta-philosophical value commitments of Steadfast Anti-Realism clear. Suppose our best scientific theories will continue to feature electrons in the next 1000 years. Furthermore, suppose we will have an excellent understanding of what it means that electrons exist, and this understanding will not change either. Despite that, if we follow the arguments of Steadfast Anti-Realists, such as constructive empiricist, we would still not be justified in believing that electrons exist regardless of the amount of empirical support for the theories and the continuity of entity-claims. Let us spin this argument even further. Imagine a counterfactual world with a nearly perfect purely cumulative advancement of science. There are no (Kuhnian) revolutions, one theory builds on the next; the track-record of scientific theories is perfect (or as perfect as imaginable) and no unobservable is ever dismissed without showing clear continuities to the successor theory. The Steadfast Anti-Realists must suppose that this is still not enough for belief permissions or at least for belief obligations (in the case of the voluntarist) in some unobservables. Any anti-realist that finds some unobservable claims, even in this hypothetical situation, epistemically obligatory to believe is already a Factual Anti-Realist, i.e. factually an anti-realist given the current evidence but potentially moving to the realist side if the diachronic evidence changes. This, however, completely changes the game when compared to Steadfast Anti-Realists. Factual Anti-Realists have to engage with diachronic evidence to back up their position. This opens up the possibility for an optimistic induction.

5. Steadfast Realism

In this section, I argue that blocking strategies are advanced on the realist side as well even though, nowadays, this becomes rarer. I argue further that those stances that build on blocking strategies should be dismissed.

Classical realists, such as the early Putnam (1975), view it as an important meta-philosophical practice to establish metaphysical claims as an explanation of our scientific empirical data based on abductive reasoning. On this basis, consider versions of classical realism, for which it is a sufficient condition for the justification of an entity claim that it is the best explanation of the data. Now suppose that our best scientific theories switch ontologies 1000 times in the next 1000 years. Suppose further that in the year 3020 the data of our best scientific theories is best explained by the existence of xyz. Given that the theory is mature and predicatively successful, we would have belief obligations or at least permissions in xyz.¹⁹ Such a version of classical realism, one that solely relies on inference to the best explanation (IBE), blocks itself from diachronic evidence as well.

Steadfast Positions either deny or affirm the existence of unobservables no matter what the diachronic evidence shows. Presented this way, both the synchronic evidence dimension, as explained in section 1, and the diachronic evidence dimension, as shown now, have very little to no weight on the stance choice of Steadfast Positions. There is a categorical difference in philosophical methods: the Steadfast Anti-Realist is committed to avoiding metaphysics at all costs; the Steadfast Realist is committed to finding metaphysical explanations at all costs.

Since the amount of evidence in such a thought experiment can be adjusted to any extremes and the Steadfast Positions remain steadfast, this should demonstrate that large parts of the fundamental disagreement in the realism debate is not motivated by diachronic evidence. Instead, a substantial amount of disagreement is due to a commitment to different epistemic values.

One might object that my characterisation of this scenario is straw-manning the realist. Would realists not say that theories that switch ontologies that often are simply not mature, and only from mature theories we should infer via IBE to the truth of its ontological commitments (cf.; Boyd 1983; Psillos 1999, 102–103; Putnam 1978). There are then simply no such Steadfast Realists. I reply, first, that this scenario is surely not unfair against some very prominent realists and defenders of IBE. Musgrave, for instance, straightforwardly defends the following principle: ‘It is reasonable to believe that the best available explanation of any fact is true’ (Musgrave 2007). Given this principle, by stipulation of my case, it is reasonable to believe in xyz in the year 3020.²⁰

Second, restrictions via the notion of maturity are tricky since maturity is a notoriously dogy term. Worrall (1989, 113) raises the ‘suspicion that the realist has supplied himself with a very useful ad hoc device’. But Worrall also gives an influential suggestion what ‘maturity’ could mean:

[A] realist should take it that a science counts as mature once it has theories within it which are predictive in this latter demanding sense—predictive of general types of phenomena, without these phenomena having been ‘written into’ the theory. (Worrall, 114)

This is a demand of maturity that my case can easily fulfil. That the underlying ontology switches all the time does not mean that there is not some continuity and progress on the predictive level on every step of the way.

Other realists are more demanding than Worrall. Psillos explicates the following criterion of maturity.

[It is] characterised by the presence of a body of well-entrenched background beliefs about the domain of inquiry which, in effect, delineate the boundaries of that domain, inform theoretical research and constrain the proposal of theories and hypotheses. (Psillos 1999, 102–103)

This is now a criterion that might not be fulfilled by my case, given that these background beliefs are about wide parts of the ontology and the constant switch in ontologies indicate a form of theory instability. On this, I agree that realists who make the maturity criterion that strong, are then not Steadfast Realists. I think that it is no accident that Psillos, who has a more demanding criterion for maturity, is also more selective in his ontological claims than classical realists as he advocates for differentiated believing dependent on the evidential support of each part of a theory (cf. Psillos 1994, 161). One expects Standfast Positions more on the very anti-realist or very realist end of the spectrum, since positions in between are often motivated by selectively choosing ontological commitments on the background of diachronic evidence. Psillos, as a selective realist, falls into this middle ground.

This gets me to the third reply concerning potentially straw manning the realist. It is indeed true that there is an asymmetry between classical realism and classical anti-realism. The thought experiment should have shown very well that at least constructive empiricism, and potentially various strains of classical empiricist instrumentalism are Steadfast Positions that are immune to the development of diachronic evidence. It is indeed not as clear to what extent this is true for versions of classical realism. Classical realists always were at least to some degree responsive to arguments from PI, as soon as they emerged. That is why justified beliefs are typically restricted to apply only to mature (as just discussed) and predictively successful theories. Such classical realists also infer only to approximate truth or truth in parts rather than full truth.

At least nowadays, realists are already somehow selective about what claims of current science to believe. When it comes to diachronic evidence, if there would be more and more problematic episodes of discontinuities, realists would just add more and more qualifications.²¹ Let us call this position, which I think is mainstream realism nowadays, Cautious Realism.

Steadfast Anti-Realist stances cannot be that nuanced. Diachronic evidence could in principle not warrant even very few very restricted claims about unobservables even with a perfect track-record of science. But then, if Steadfast Anti-Realists are this risk-averse, and there is also no theoretical reason why one would restrict one’s standards of epistemic risk to just the region of unobservables, vast other regions of kinds of hypotheses would also suffer from reversals and sceptical considerations. Such anti-realists are no longer merely anti-realists, but rather very risk-averse general investigator in all regions, observable and unobservable. Since wide-spread scepticism is an epistemically paralyzing position, I suggest rejecting such a stance.

This reorientation of anti-realists away from steadfast versions is to some extent on its way, starting with Stanford (2006), but I think, as argued here, we should be bolder in what that means for a dismissal of steadfast anti-realist accounts.

6. Starting to Break the Stalemate

If steadfast anti-realist is to be dismissed, what is the alternative? I suggest the following version, which is a form of anti-realism with a symmetrical treatment of historical evidence, I term factual anti-realism.

[FA] Factual Anti-Realism.

• (i-a) High-Standard Value Commitment. Factual anti-realists commit themselves to high epistemic standards for their ontological commitments. (These cannot be so high as to lead to radical scepticism.)

• (i-b) Moderate-Standard Value Commitment. Factual anti-realists commit themselves to moderate epistemic standards for their ontological commitments. (These cannot be as low as believing contradictions.)

• (ii) Factual Claim. Given (i-a) or (i-b), current diachronic evidence is not sufficient for obligations to believe in any unobservables (be it entities or structures) even if postulated by our current most successful scientific theories.

• (iii) Responsiveness Claim. It is possible that new diachronic evidence comes up that is strong enough to overturn (ii). If it is evidence for an entity or structure x, then, at least, one should give up anti-realism with regard to x.

Given (i)-(iii), FA comes in two flavours:

• [VFA] Voluntarist Factual Anti-Realism. Is committed to (i-a). Finds (i-b) permissible. Claims that given (i-a), (ii) follows, but given (i-b), (ii) does not follow. Is committed to (iii).

• [SFA] Strict Factual Anti-Realism. Is committed to either (i-a) or (i-b). Claims that given (i-b), (ii) follows. Is committed to (iii).

As such, VFA contends that current diachronic evidence cannot compel one to believe in any unobservables but permits one to believe in at least some unobservables. SFA contends that current diachronic evidence compels one not to believe in any unobservables.

My considerations provide two reasons why versions of FA are advantageous to Steadfast Anti-Realism. First, in scenarios with a very solid historical track-record, anti-realists do not have to bite the bullet of remaining anti-realist no matter what the evidence shows. They will give up being anti-realists based on diachronic evidence. Linking anti-realism to the diachronic evidence dimension makes for a stance that does not unreasonably block itself from diachronic evidence. Anti-realism will become empirically testable by a long-term research project in philosophy of science. Second, versions of anti-realism that are responsive to historical evidence are a genuine novel response to the evidence specific to the sciences and to the history and philosophy of science and not merely a radical sceptical epistemological stance that can be applied to any knowledge-seeking endeavour whatsoever without engaging with the domain-specific evidence.

Note that the same argument can be advanced against those instrumentalists who follow a similar argumentative strategy as constructive empiricists by blocking themselves from refutation by diachronic evidence. If they too conclude their anti-realism starting from similar a priori empiricist principles, then such instrumentalists should be dismissed on the same grounds. Those instrumentalists, however, who were motivated by diachronic evidence to begin with, are not in the boat with constructive empiricists. They would change their stance if the historical track-record improves significantly. They might then always have been Factual Anti-Realists and I hope I was able to make their commitments more salient and clear.

This analysis reveals a point where I disagree with Wylie’s analysis on epistemic risk. She views realism and anti-realism as ‘incommensurable modes of philosophical practice’ (Wylie 1986, 287). I think that the notion of ‘epistemic risk’ can make rival stances commensurable and can tell us precisely which parts are commensurable. First, it makes the accounts in the realism debate commensurable on an axis of risk-taking and risk-avoiding. On this part, Wylie would most likely agree, even though she did not spell it out explicitly. However, secondly, it also reconnects the debate back to the empirical evidence, and thus makes the accounts commensurable relative to empirical diachronic evidence. Finally, what might remain incommensurable is the epistemic value-decision itself. I will give some prospects about this in the last section. Before that, I want to analyze whether blocking strategies are also in place for selective realists.

7. Blocking and the Epistemic Structural Realist

This section is devoted to accounts with intermediary epistemic risk. Typically, versions of selective realism fall into this category. They can be viewed as walking the thin line between epistemic risk-tolerance and risk-aversion. As one example, I want to pick out epistemic structural realism (ESR). I will argue that some structural realists also follow a blocking strategy.

Psillos (2001) distinguishes between an ‘upward path’ [ESRup] and a ‘downward path’ [ESRdp] for structural realism. Similar to various anti-realists, the upward path (e.g. cf. Russell 1927; Weyl 1963; Maxwell 1970) starts from empiricist principles and tries to establish that our scientific theories can be a structural image of reality but not possibly more. Thus, it denies justified beliefs about entities in principle. This principled argumentative strategy is again unshakable by synchronic or diachronic evidence. One can run a similar thought experiment as the one I gave for constructive empiricism.

However, contrary to constructive empiricism, ESRup might be able to recover some connection to diachronic evidence. Suppose a situation with overwhelming diachronic evidence for structural discontinuity. ESRup could then argue the following. Even though science can only be a structural image of reality, unfortunately, given current science and how the world behaves, science cannot even deliver that. If, however, diachronic evidence improves, structural claims can become justified. This is a potential strategy for ESRup to not block itself from diachronic evidence to some degree. Note, however, that all this can do is allowing diachronic evidence to move ESRup closer to the anti-realist side by abandoning some or all justified beliefs in structures. If diachronic evidence would show an increase in entity continuity, this would not move the ESRup to the full-blown realist.

Today, the more widely held structural realist view, however, is characterised by a downward path, as advocated by Worrall (1989). The downward path starts from a classical realist position by accepting the existence and knowability of a mind-independent world. However, advocates of ESRdp think that PI discredits full-blown realists with their entity claims while structural claims do not share a similar fate. If this argumentative strategy of ESRdp has any epistemic bite, then it has to be sensitive to new diachronic evidence. If diachronic evidence would begin to show structural discontinuity, then this has to count as evidence against ESRdp by pointing in a more anti-realist direction, and if diachronic evidence would begin to show a continuity of entity claims, at one point, this has to count as evidence against ESRdp as well. Furthermore, if one would get a discontinuity in structural claims but continuity in entity claims this would have to move the ESRdp to a Hacking-style entity realist. The move in this direction is, however, surely the most curious of the presented option. It is, after all, highly controversial whether there can be radical changes in the structure of theories while preserving entity-continuity (c.f. e.g. Resnik 1994).

With this depiction of ESRdp, it is clearly susceptible to new diachronic evidence. It is also clearly dependent on an evaluation of epistemic risk. Consider Psillos’ (1994) attempt of showing that in theory-change from the caloric theory to thermodynamics the truth-content was retained. Or consider Nola’s (2008) attempt to show the continuity of the electron. If such continuities can be successfully demonstrated and one takes some epistemic risk, then the ESRdp must move from structural realism closer to classical realism. Furthermore, if the epistemic risk of some entity claims is on par with the epistemic risk of some structural claims, then it would be irrational to believe in the one but not the other, no matter where one sets the bar for epistemic risk. This implies that the structural realist has the specific task to demonstrate that every structural claim that the structural realist thinks is justified has a lower epistemic risk than any entity claim relative to the diachronic evidence. I think, structural realists generally agree with this, even though this commitment is quite under the surface. It is definitely part of the strategy of structural realists to show that the claims of entity-continuity, such as Psillos’, are all not warranted given the epistemic standards set for the structural realists.

If structural realists actually want to convince entity realists, then the argument has to be done on the floor of entity realists, i.e. given fairly moderate epistemic standards it has to be shown that arguments of the entity realists do not hold up. If structural realists cannot do that, they will be met with suspicions that they just set the bar of epistemic risk ad hoc high. Conversely, entity realists cannot convince structural realists by merely arguing on the grounds of fairly moderate epistemic standards for relevant entity continuities either. Entity realists need to show this by application of fairly high epistemic standards, only then structural realists need to be responsive. Otherwise, both positions are just self-justifying under their own epistemic standards.

This dialectic shows that my strategy of reconnecting the accounts in the realism debate to the historical evidence can partly break the stalemate. However, it also reveals that there will always be remaining disagreement that cannot be resolved by this strategy because some disagreement emerges from the adoption of differences in risk balancing.

Note that Psillos (2001, 18) starts by arguing that even ESRdp is characterised by a principled divide between knowable structure and unknowable entities. From this, Psillos develops all his objections against ESRdp. However, in his conclusion, he leaves the room for a version of ESRdp ‘as a modest epistemic thesis‘ (Psillos 2001, 23), and even though he does not spell out the details, I take it that my short explication of ESRdp would do justice to his demand because it does not draw said principled divide. Instead, in my explication, and Worrall (1989) might be read as agreeing, the structural realist conclusion, if it holds up, is drawn from diachronic evidence and can change with new diachronic evidence.

8. Where Does that Leave the Debate?

We are now left with three varieties of stances that are responsive to historical evidence: factual anti-realism, cautious realism, and some varieties of selective realism, such as ESRdp. There is, of course, still research to be done whether all those positions can actually account for all synchronic and diachronic evidence. Maybe Hacking-style entity realism cannot avoid the question of representation (cf. e.g. Resnik 1994), and thus the epistemic risk of believing in causally manipulable entities is still higher than Hacking claims. Maybe ESRdp cannot establish structural continuity as well as their proponents claim (cf. e.g. Chakravartty 2004, 164; Stanford 2003, 570–572), and thus the epistemic risk by believing in the structural portions of our current most successful scientific theories is actually not that low. Or maybe Psillos-style entity realism cannot establish entity-continuity as well as he says he can (cf. e.g. Chang 2002), and thus his position has more epistemic risk than says. Such arguments already received great attention and are very important to establish the actual level of epistemic risk of different accounts. However, such arguments are not in the scope of my paper. Instead, I focused on reconnecting the accounts to diachronic evidence.

9. Results and Prospects

I started with the Jamesian picture about epistemic value, aligning with the recent development of voluntarist epistemology in the scientific realism debate. I agree with current voluntarists that the epistemic trade-off can be struck so that epistemic standards are strict. Consequently, anti-realism can emerge as one permissible position. However, if it does, it is because the evidence says that this is the right verdict and not because we dogmatically insist that no evidence can ever proof realism right. This highlights that factual anti-realism is superior to steadfast anti-realist positions, such as constructive empiricism. Furthermore, I argued that steadfast realism, such as Musgrave’s account, are to be dismissed as well and proposed cautious realism instead. Lastly, I explicated a version of selective realism (ESRdp) that is responsive to diachronic evidence.

All of those positions are susceptible to new diachronic evidence, which makes this conceptualisation of the debate departing from the depicted steadfast stances. My conceptualisation rather makes realists, anti-realists, and selective realists comparable on an axis of epistemic risk relative to the diachronic evidence. The scientific realism debate should then be understood as a long-term research project. The longer mature science develops, the more diachronic evidence of theory-continuities and discontinuities can be gathered. Different accounts not only accommodate historical evidence better or worse, but they should also be interpreted as predicting how diachronic evidence develops and their predictions can be confirmed or disconfirmed by the evidence.

After dismissing dogmatic accounts, and integrating diachronic evidence properly, some disagreement will remain, which will mostly boil down to balancing. And here, in philosophy as in daily life, it is a matter of personal preference how much risk one is willing to take. This transfers the remaining stalemate to a handshake. Agents need to conditionalise on their values and cannot dogmatically insist that their value-choice is superior. As such, our theories of rationality should reflect an attitude of tolerance, such that enquirers are more open to rival accounts.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 I will utilise here the common distinction between observables and unobservables even though it is widely recognised that there is no such context independent distinction to be had. However, nothing in my argument depends on the choice of the empirical basis for the anti-realist. In fact, this should be read as a placeholder for ‘those portions of reality your preferred anti-realist deems problematic (e.g. those not sufficiently accessible by experience (e.g. Hume), or by detection (e.g. Bogen 2011), or by narrow observation (van Fraassen 1989), or by wide observation (Bueno Shalkowski 2020)) and those portions that are deemed unproblematic (e.g. those that can be seen with the naked eye)'.

2 I will use for now ‘justification’, but you can pick your preferred notion of rational standing here.

3 Following, for instance, Berker (2013) that epistemology is about what we ought to believe. Some philosophers think that there cannot be any obligations to believe because of the problem of doxastic involuntarism (see e.g. Alston 1988). It is outside the scope of the paper to taggle this argument. In the end, I agree with Feldman (1988) that even if we had not any control over our belief-forming processes, there are still epistemic belief obligations.

4 Something similar can be said for Rowbottom’s (2019, 461) suggestion of an epistemic realist thesis (ET): ‘A proper subset of science’s content, E, is approximately true (on a proper subset of theories of truth, T).’ Note first, that for a specification of a realist, it is not enough that (ET) is true. For any S to be a realist, S at least has to affirm or believe that (ET) is true. But again, I can believe (ET) without any epistemic reasons, which makes me a minimal realist, but not one that is epistemically committed to a lot.

5 E.g. ‘it is reasonable to accept […]’ (Musgrave 1988), ‘it is reasonable to believe […]’ (Musgrave 2007). ‘Why We are Justified to Believe in Colored Quarks’ (Massimi 2004).

6 For a different criticism see Rowbottom (2019, 463–465).

7 For instance, (a) selective realists frequently argue on grounds of synchronic evidence to dismiss some ontological commitments but not others (cf. e.g. Magnus Callender 2004), (b) it is frequently maintained by realists that theories under consideration need to fulfil some theoretical virtue, such as maturity, predictive success, or being past a ‘take-off point’ (e.g. Boyd 1983). Theories lacking such virtues will be discredited even by realists, and whether some theory is past the take-off point is a matter of synchronic evidence.

8 Going back to Poincaré’s Science and Hypothesis (1902) and Laudan (1981).

9 It also has many challenges. Major ones are interpretative problems on the semantic dimension of scientific realism. For instance, could we redescribe ‘caloric’ as badly described thermal energy rendering various claims by the caloric theory at least as approximately true? (see e.g. Psillos 1999, 110–124) These deep problems need to be engaged with if we give a bigger role to the history of science (as suggested in the present paper). However, this paper focuses on specific points of the epistemological dimension only.

10 Take, for instance, Kuhn (1962), Laudan (1981), or Stanford (2006) who draw various anti-realist conclusions, or selective realists who draw realist conclusions but for structures only (Worrall 1989), entities only (Hacking 1983), properties only (Chakravartty 2008), or only for various stable, invariant portions of theories (Kitcher 1993, Psillos 1994, Nola 2008).

11 For early accounts in epistemology see BonJour (1985, 7–8), Goldman (1979, 29–30), and Lehrer (1990, 112). For a defence see David (2001).

12 Since I adopt this trade-off, I need to respond to a likely objection. The realist usually adopts a notion of ‘approximate truth’ and not ‘full truth’. As such, since approximate truth involves some kinds of error, must there not be a trade-off between acceptable and unacceptable error instead? Following Musgrave (2007), my view is the following. If S tells me that they believe that theory T is approximate truth, then S should tell me instead in which limit, in which domain, or under what conditions they believe it is fully true. If they cannot do that, then they are not clear about what they believe. Suppose some S would tell you that the caloric theory is actually approximately true. What does S believe? What S should believe (given current science) is something like the following: Some parts or implications of the theory are fully true. For instance, one implication of the theory is that a cup of hot tea cools down in a cold room. This S believes. Some parts were fully false. For instance, that there is a fluidal substance of heat that passes from warmer to colder bodies. This is fully false, and S does not believe it. Believing in the ‘approximate truth’ of the caloric theory means believing in the full truths of various parts of the theory and its implication. Realists believe that these parts are fully true but also believe that sometimes they are in error. The trade-off is between full truth and full error.

13 This leads to a challenge pointed out by Grimm (2008): irrelevant truths are arguably justified. I think these problems can be bypassed by differentiating narrow justification and wider norms of belief acquisition.

14 Already present in Chakravartty (2007, 187; 2013, 31) but it appears to be now even more permissive.

15 See, for instance Lyons (2013) and Mizrahi (2015), arguing that UA needs to be empirically and historically informed.

16 Arguably, we do not long for certainty, so it is not demanded from the realist to show a perfect track-record.

17 ‘Relevant category’ because selective realists might want to restrict those inferences to stable subparts of a theories, e.g. structural portions.

18 Something similar can be said about the general problem of how historical case studies can function as evidence and how they can be systematised to even get something akin to an inductive inference of the ground (cf. Chang 2012; Kinzel 2015; Nickles 1995).

19 This is, of course, a quite far-fetched scenario but it is only presented to make the logical point coming up. One may replace the example with an example that explains a more realistic increase in the occurrence of radical theory changes of future scientific development. Such an example would serve the same purpose.

20 Something similar can be said about Boyd (1983) but I will not go into the details here.

21 This criterion suggests that there is not a clear-cut demarcation between full-blown realism and selective realism if one does not want to strawman ‘full-blown’ realists into stating that it is justified to believe all ontological commitments of current science.