Resisting the Historical Objections: The Selective Strategy

The selective strategy rejects PMI1 by holding that even in radically false and discarded theories there were some truths, which are still preserved today; hence even in current theories there are some (and presumably more) truths.

8.1 Referential Continuity and Entity Realism

Realists employed Kripke's and Putnam's (1975c, d) causal theories of reference to argue that discarded theories were not completely off-track, because they posited the same entities we now believe in, although with different descriptions and sometimes with different names. For instance, the term ‘ether' referred to whatever caused its introduction—say, the electromagnetic field (Hardin and Rosenberg 1982).

But this does not seem possible when the core descriptions associated to a term are completely wrong, or there exists nothing even slightly similar: there is nothing to which ‘phlogiston' or ‘caloric' might refer (Putnam 1978a: 25; Laudan 1984b: 160-161; Psillos 1999, 290-293); understanding ‘ether' as ‘electromagnetic field' may be overstretched (Worrall 1995). Holding that a term refers to whatever is the cause of the phenomena it is supposed to explain would trivialize reference: for example, Aristotelian natural places, Newton's gravitation force and Einstein's curvature of spacetime, all were supposed to be the cause of gravitational phe­nomena.

Psillos raised further problems (1999: 286-287), and proposed instead a causal-descriptivistic account, by which a term refers to the unique natural kind having the core causal properties assigned to it by the description, provided that the actual kind-constitutive properties are the causal origin of such description (ibi: 295). More recently, Schurz (2011) proved a correspondence theorem showing that a theoretical term originally intended to refer to an inexistent entity may indirectly refer to a real counterpart of that entity. For instance, in phlogiston theory ‘dephlogistication' indirectly refers to the process of oxidation.

So, noting that different theories of reference have been proposed, motivated by conflicting intuitions, Votsis (2011a) suggests that perhaps the very concept of reference is not a monolithic one. At any rate, what ‘reference’ means is conven­tional, and scientists may intend to pick their referents in different ways: no doubt Mendel understood ‘gene’ very liberally, as whatever played a certain causal role, while Bohr may have understood ‘atom’ as something very close to the description he gave. Besides, probably the crucial question is not whether T’s terms refer, but how much truth is found in T. It doesn’t help much that T’s terms refer, if everything T says on their referents is wrong. If a relaxed definition of reference is used, T can be radically false although its terms refer; if instead a strict definition is chosen, there can be a lot of truths in T even if its terms fail to refer (more on this below). So, reference may help, but it is neither necessary nor sufficient to resist the PMI (Alai 2006: 239).

According to a different approach, however false some theories may be, the entities they postulate must exist, because we currently manipulate them. For instance, in certain experiments electrons are sprayed to reveal the existence of quarks with fractional charges.

Musgrave (2006-2007) objected that “entity realism is a hopeless form of realism”, because the existence claims are empty without some description of properties and behaviour. Moreover, the manipulation argument is question beg­ging: who says you are actually manipulating an unobservable entity, rather than merely performing certain macroscopic operations with certain macroscopic effects? (see van Fraassen 1985: 298; Nola 2002: 9).

To avoid these problems the argument can be reinterpreted as an inference to the existence of such entities as the best explanation of their observed effects (Nola 2002: 9-14), but then it supports also some theoretical assumptions about them. Otherwise, it can help to solve the Duhem-Quine indeterminacy of empirical control: machines incorporate the less problematic of our beliefs, which can be used to test the more problematic ones (Giere 1988). Even so, however, entities and beliefs about them go hand in hand. Summing up, the partial truth of theories cannot be confined to the existence of the entities they postulate (see also Dorato 2007: 183-184; Nanay 2016).

8.2 Structural Realism

In opposition to entity realism, structural realism (StrR) holds that only structures can be known, while entities cannot (Frigg and Votsis 2011). According to different versions, we cannot know the individual entities, but we can know their properties and relations; or not their intrinsic properties, but their first-order relational prop­erties; or none of these, but the second-order structure of their relational properties.

This last was Russell’s (1927) and Carnap’s (1928) view (Ladyman 2014; French and Ladyman 2011).

Poincare adopted StrR in reaction to the PMI: as theory T₁ is replaced by theory T₂, and T₂ by T₃, etc., the entities postulated by T₁ are substituted by different ones postulated by later theories (for instance, ether was substituted by the electro­magnetic field); but the basic equations, tracking the underlying structure of things (e.g.

Lately this position has been advocated by Worrall: “Fresnel completely misidentified the nature of light, but nonetheless it is no miracle that his theory enjoyed the empirical success that it did... because Fresnel’s theory had... more or less the right structure”. Thus, “there was continuity or accumulation in the shift [from Fresnel to Maxwell], but. one ofform or structure, not of content’” (Worrall 1989. See also Worrall 1994, 1995: 92-94). Therefore, showing that the success of discarded theories was due to their structural claims, StrR also vindicates the NMA against Laudan’s MMT refutation. Besides, since incompatible but empirically equivalent theories may share the same structure, it has been suggested that StrR can also solve the underdetermination problem (Worrall 2011; Lyre 2011; French 2011).

More recently, the merely epistemic thesis that we can know only structures (EStrR), has been reinforced by the ontic thesis that there exist no objects but only structures (OStrR). The latter may be further detailed as claiming that (1) there are no individuals but only relational structures; or (2) relations do not supervene on the intrinsic spatio-temporal properties of their relata, or (3) individual objects have no intrinsic natures or properties; or (4) the identity of objects is ontologically dependent on their relations; or (5) individual objects are just constructs used to build approximate representations of the world (Ladyman 1998; French, Ladyman 2003a, b, 2014; Ladyman and Ross 2007).

OStrR is strongly suggested by contemporary physics: in the entangled states of quantum mechanics relations do not supervene on the properties of particles, and particles seems to lose their individuality, since there are no properties, even spa­tiotemporal, which allow to distinguish them from one another.

Moreover, the traditional ontology of individuals, intrinsic properties and rela­tions seems to be at odds with the nature of space, time and matter. The proper objects of contemporary physics are rather symmetries and invariants, and “ele­mentary particles are hypostatisations of sets of quantities that are invariant under the symmetry groups of particle physics” (Ladyman 2014), or excitations of fields.

Many objections have been raised against StrR: do the mathematical equations preserved in theory change tell something about the underlying structure of the world, or merely about empirical regularities? (Laudan 1981: 237). Isn't the retention of equations just a convenient and labor saving pragmatic feature of scientific practice, due to the conservativeness of the scientific community? (Fano 2005). That equations tell us something about the structure of the world can only be shown by the NMA, which however supports at the same time the theoretical claims about entities (Psillos 1999: 152). In fact, for StrR structures are the only responsible for success. But mathematical equations by themselves can license predictions only when theoretically interpreted and supplemented with auxiliary hypotheses; so, the success of predictions confirms also those theoretical hypotheses (ibi: 153-155).

Besides, we cannot distinguish “between the nature of an entity and its structure such that we can... know its structure but not its nature”, because “to say what an entity is is to show how this entity is structured” (ibi: 155-156). For instance, there was no structure of light on which Fresnel was right while being wrong on its nature: there simply were properties of light on which he was right and others on which he was wrong (ibi: 159).

Granted, the properties on which he was right were behavioral and relational properties (the ways of propagation) while he was wrong that the physical nature of light was molecules of ether.

Moreover, structural realists must specify what exactly is the “structure” pre­served across theories, and how is it represented. For Grover Maxwell, Worrall, and Cruse and Papineau (2002), it is the relevant equations, represented by the theory's “Ramsey sentence” (a formulation where all the theoretical terms are substituted by existentially quantified variables). However a number of problems arise concerning the adequacy of such representation (Demopoulos and Friedman 1985: 635; Psillos 1999: 63-69; Ladyman 1998, 2014). This is why, instead, Ladyman and Ross (2007) and French (2014) hold that what is preserved are symmetries and the associated group-theoretic structures, represented by the semantic formulation of theories.

Opponents of StrR also charged that it cannot account for the difference between physical reality and mere mathematical structures; that often also structure is lost in theory change; and that StrR only applies to physics (Ladyman 2014). Specific objections have been raised against the ontic versions of StrR, especially con­cerning the plausibility of the existence of relations without relata, and the extent to which physics univocally supports this view (ibid.). There is no room here to discuss them; besides, even if OStrR were wrong, one could still use EStrR as a form of selective realism able to resist the PMI.

Summing up, on the one hand the basic role of structures (in some sense of ‘structure’) in modern physics seems undeniable, and there are striking examples of structure preservation in theory change; on the other hand it has not been shown that no theory has ever correctly described the nature of entities, or that entities were never essential for novel predictions; moreover, although there have been attempts to apply StrR to biology (French 2011) and to the social sciences (Ladyman and Ross 2007; Kincaid 2008), it is far from clear that it can apply to all scientific disciplines. Therefore, while these questions are still hotly debated (Ladyman 2014), at present it may be reasonable to assume that StrR can account only in part for our realist commitments, and we should take a liberal position on which kinds of features (entities, or laws, or structures, or particular properties, etc.) theories can get right even if otherwise false.

8.3 Deployment Realism

Kitcher’s and Psillos’ “deployment” realism is not concerned with the kind of theoretical components which survive scientific change and deserve realist com­mitment (whether they are ontological or structural), as with their role: they are those essentially deployed in novel predictions.“No sensible realist should ever want to assert that the idle parts of an individual practice, past or present, are justified by the success of the whole” (Kitcher 1993: 142). In fact if a component C were not deployed essentially in deriving prediction NP, there would be a different component C' not implying C, from which NP could have been derived; hence success could equally be credited to C', and the truth of C would no longer be the only plausible conclusion of the NMA.

Lyons (2002) criticized deployment realism by listing a number of novel pre­dictions derived from components we now recognize as false. For instance, the idea of absolute acceleration was used in derivations from Newton's theory; the claims that phlogiston is the principle of heat and that “sulfuric acid was dephlogisticated sulfur” were involved in Stahl's prediction that the synthesis of phlogiston and sulfuric acid would result in sulfur; the postulate that charcoal is “high in phlo­giston” and that inflammable air is pure phlogiston, were used in deriving Priest­ley’s prediction that inflammable air would, like charcoal, turn calx into metal; the prediction that the rate of expansion is the same for all gases was derived from a number of false claims about caloric; etc. (Lyons 2002: 80-81).

But it has been replied that some of those predictions were true only under a charitable interpretation (e.g., by understanding ‘phlogiston’ as deprivation of oxygen), under which, however, also the claim used in deriving them turns out to be true. Other predictions were actually derived from false claims by chance, but this was possible because they were a priori probable (and we saw that only improbable predictions are evidence for truth). All the other false claims were not actually essential in the prediction, because they entailed some weaker and true claims which were sufficient to derive the same prediction (Alai 2014b). Lyons (2006) argued that the essentiality requirement should be dropped, but this would obvi­ously deprive the NMA of its cogency.

8.4 Semirealism

Another version of selective realism is Anjan Chakravartty's (1998) semirealism. Like entity realism, he holds that we can know the existence of the unobservable entities with which we establish causal interactions; but unlike entity realism, he also grants that we can correctly describe their “detection” properties: “Detection properties are causal properties one has managed to detect; they are causally linked to the regular behavior of our detectors. Auxiliary properties are any other putative properties attributed to particulars by theories” (Chakravartty 2007: 47). So, we should be realist about detection properties, but agnostic about auxiliary properties (see Ivanova (ed.)). Bence Nanay (2013) proposed instead singularist semirealism, according to which “science is mostly right, not only about which unobservables exist, but also about their property tokens, but not their property types”.

8.5 The Verisimilitude Research Program

Two traditional and influential approaches can also be considered as forms of selective realism. One is the “verisimilitude” research program initiated by Popper (Popper 1963; Oddie 1986; Niiniluoto 1987, 1998). Its key idea is that even false theories may be more or less “verisimilar”, or “close” to the truth, since they include some true statements, or false statements with some true content. More verisimilar theories have more truth content and/or less false content. For instance, (1) ‘All swans are white' and (2) ‘All swans are black' are both false, but part of the content of (1) is the entailed statement (3) ‘All swans except Australian swans are white', which is true and explains the predictive success of (1) (Musgrave 2006­2007). Also (2) entails a true claim, viz. (4) ‘Australian swans are black', but that is weaker than (3), so (1) is more verisimilar than (2). “Approximate truth is a species of partial truth, since the approximations in question are logical parts of what we began with. ‘It is 4 o'clock' logically implies ‘It is approximately 4 o'clock' as well as ‘It is 4 o'clock give or take 5 min'” (ibid.). Rescher (1987: Ch. 5) seems to follow a similar line when he distinguishes between forefront science, which is precise and never true, and “schoolbook science”, which though vague and imprecise includes the true core of the forefront science.

8.6 The Restricted-Domain Approach

The other traditional and influential idea which may be interpreted as a version of partial realism is that theories must not be taken as true for all the phenomena and with absolute precision, but only for certain intended ranges of phenomena, or levels of approximation, or domains of entities, which are defined by the theory itself.^[7] Therefore, even when rejected, they remain true within those limits. For instance, “The fact that we can use classical mechanics in creating many machines or for sending rockets into space certainly means that this mechanics is true of its objects and therefore ‘tells a true story' about certain aspects of reality” (Agazzi 2014: 310-311).

If taken to an extreme this position might imply that all theories are analytic (Kuhn 1962, ch. 9), or that they are reducible to their empirical claims, or that they don't describe the actual world, but different worlds of our making, as claimed by Goodman (1978, chs. I, VII); for instance, it might mean that phlogiston theory was true of phlogiston, ether theory was true of ether, etc. On the contrary, theories and laws are intended to be true—period, true of everything. For instance, ‘electrons have negative charge' is not intended to be true only of electrons, but of everything, for it has the universal form ‘Vx(Ex NCx)'. If the class of intended applications is delimited a priori, it precludes the extension of the theory to new phenomena, depriving it of fecundity and heuristic power; if instead it is delimited a posteriori, following empirical failures, then it is ad hoc, and the theory risks to lose its empirical content. In order to restrict the domain of a law we need good reasons, beside an experimental failure; in particular, we need to find a different (specialized) law, which however has a universal scope, and above all must be embedded in a different theory. Moreover, the theories which Einstein called “principles theories” don't delimit any field of intended applications.

This approach is fine, however, if interpreted as the selective realist thesis that theories which are radically false overall, nonetheless include descriptions—not merely empirical, but theoretical as well—which are approximately true for certain domains, or aspects of reality, or scales, or levels approximation. There is no delimitation of the intended applications, either a priori or a posteriori, and when a claim is empirically refuted it is declared false. However, it may be found that it was partly true, since it entailed a weaker but true claim. Still, the latter must be supported and explained by a new theory. For example, the assumption that mass is inalterable was proven false—false of everything. However, it entails the true claim that mass is approximately inalterable at low speeds; yet, this limitation cannot be explained by Newton's theory, but only by Relativity. This is why Newton's theory is considered false—period. The theory of luminifer ether was intended for all the phenomena of light, but it is true of none. Rather than conceiving theories as completely true of a restricted domain, they may be seen as partially true of a universal domain. Partial truth also explains why different theories can be all true in the same field: they do not concern different realities, but different aspects of reality (Agazzi 2014, 405).

8.7 Local Realisms?

In view of so many different ways and versions of realism, Magnus and Callender (2004) argued that we shouldn't look for “wholesale” arguments for realism, but for a case-by-case defense. Saatsi (2016) suggests that realists should give up the idea of a general “recipe... capable of distilling the trustworthy aspects of a theory, applicable to any good, predictively successful mature theory”. Instead, they should settle for an “exemplar realism” which focuses on specific, “local” reasons for realism. All these different recipes shouldn't be seen as competitors, but “as cap­turing the different possible ways that a theory can ‘get the world right'” (French, this volume). Yet, why are all these versions forms of realism, what do they have in common? (this was Plato's problem of the universal). For Saatsi it is the general idea that science is successful because it gets something right about the world, but this needs not be exactly the same in all theories, contexts or disciplines.

We noted that deployment realism is already enough flexible as to which kind of components can be right, focussing more on how the particular true claims can be identified. But also different identification criteria have been proposed, and perhaps they are compatible, or suited to different contexts: being essential to derive novel predictions (for deployment realism); being confirmed by indirect but theory-free observation (as explained above); being the “minimally interpreted mathematical parts” of successful theories (Votsis 2011b); being the minimal sub-theories which are presupposed by the successful predictions and not empirically refuted (Peters (2014); being involved in predictive success, resistant to hostile probing and with outside support (Cordero, this volume); attributing properties that are in principle observable, measurable by distinct independent methods and causally produce the observed data (Ghins, this volume).

Votsis (2011b) and Peters (2014) argue that in order to save the NMA from the MMT we should be able to identify the particular true components of discarded theories prospectively, from the authors' point of view. In fact, if we could identify them only retrospectively, as those which are preserved in today's theories, we would beg the question of the cogency of the NMA by assuming that the currently accepted theories are true (see Stanford 2006: Ch. 6).

However, if we could do this for past theories, we should also be able to distinguish in current theories precisely which claims will be preserved forever and which ones will be discarded, which is impossible (Alai 2016: § 3; Nickles, this volume). Yet, realists can still make their point if they can (a) argue in general that a successful prediction must be derived from some true assumption, even without being able to pinpoint exactly which one; (b) for any particular successful pre­diction NP, show that indeed the theory includes certain assumptions A₁^A_n from which NP can be derived, and that for all we know A₁^A_n may be true, because they fulfill the just mentioned criteria; and (c) explain away each putative coun­terexample of true predictions apparently derived from false assumptions as sket­ched above.

9