Scientific realism with correspondence truth: A reply to Asay (2018)

Characterizing scientific realism

In our original article, we raised the question of whether or not correspondence truth implies a commitment to metaphysical realism—the idea that things in the world exist independently of our beliefs about them. We acknowledged that because truth is basically a semantic issue, and realism is most fundamentally a metaphysical issue, the two need not go together. For instance, there is no inconsistency in asserting that the claim “realism is false” is true if and only if realism is false; thus, one can consistently deny realism while affirming correspondence truth.

However, since the core claim of realism is that scientific theories aim to identify and characterize such things as objects, properties, and causal mechanisms that exist in the world, which make up plausible truthmakers for sentences that feature in scientific theories (truthbearers), it is natural (but not logically necessary) to wed realism to correspondence truth, and, for this reason, we stated our preference for bringing them together under the umbrella of the broader notion of scientific realism. Below, we will in fact argue that, although correspondence truth does not require scientific realism, scientific realism does typically require correspondence truth.

We and Asay share with many realists the belief that scientific realism is first and foremost a metaphysical thesis, but we disagree on whether or not correspondence truth should be part of scientific realism proper. Not only does Asay want to remove correspondence truth from the ambit of realism, he also thinks that it is a mistake for realists to characterize their philosophy by focusing on the aim of science (Asay, 2018). Thus, our disagreement with Asay about the relation between realism and truth points to the controversial issue of how scientific realism should be characterized, an issue to which we turn now.

Realism as a family of doctrines

Although realism is customarily presented in a global form, intended to apply to all the sciences, Asay (2016) favors local formulations of realism that cater to the diverse natures of the various sciences. This enables him to readily reject realist debates about the aims of science, particularly the aim of truth. We too favor selective and local realist strategies, where appropriate. One of us has made selective use of both theory realism and entity realism to help shed light on different aspects of the conceptual foundations of psychometrics (Borsboom, 2005). The other has argued that a fine-grained understanding of the behavioral sciences will require local rather than global formulations of realism (Haig, 2014).

That said, we count ourselves among those realists who think there is merit in understanding realism as a broad family of doctrines (e.g., Hooker, 1987; Psillos, 1999; Sankey, 2008). Construed in this manner, realism, in its best form, is a general, systematic theory (Hooker, 1987) that combines the full range of relevant theses in a coherent manner. We think it stands to reason that the ability of realism to help make best sense of the varied and complex nature of science is maximized by appealing to our best formulation of general realism, not local realism. However, we hasten to add that full-blown realism will do much of its day-to-day work as a reservoir of theses from which one can selectively draw as an aid in formulating local realisms as specific occasions demand.

More often than not, three core doctrines are presented in expositions of realism: (a) metaphysical realism which, as just noted, maintains that there is a real world of which we are part, and which science investigates; (b) semantic realism, which is the claim that scientific theories should be interpreted realistically and taken at face value; and (c) epistemological realism, which asserts that both observable and unobservable features of the world can be known by the use of scientific methods. Sometimes, an additional thesis, (d) axiological realism, is added. This is a thesis about the aims of science, and standardly states that science primarily aims for true theories. Finally, the family of realist theses is occasionally extended to include (e) institutional realism. This thesis maintains that science is a human social endeavor and is subject to institutional as well as theoretical determinations.

Our particular concern here is with the theses of semantic and axiological realism, which have a direct bearing on the question of how truth relates to realism.

Truth in realism

Among other things, modern science articulates, and works toward satisfying, the various aims it seeks to realize. Just what these aims are, and how they should be characterized, is a matter of ongoing debate. For this reason, we think that an adequate realist philosophy of science should be critically aim-oriented. One long-standing view is that scientific realism itself is fundamentally a thesis about the aims of science. This is the thesis of axiological realism just mentioned, and it says that scientific inquiry aims to discover truths about the world. We hold a broader view of axiological realism than this, because we think truth is a cardinal aim of science but that it is not its only aim. For example, understanding, prediction, and control are also important aims, and whereas these aims are satisfied in good part by successfully constructing explanatory theories, they do not strictly require it across the board. Control, for instance, which is satisfied when we successfully apply scientific knowledge to improve human welfare, can be obtained by using theories that are not literally true, but that are good enough to obtain desired results—the use of Newtonian physics, which incorrectly posits absolute space and time but offers approximations that are highly useful for exerting control over natural phenomena, is a case in point. Although we agree with Asay (2014) that realism should not be characterized in terms of the axiological thesis alone, we do think that it should count as one of the major theses of realism, and this means that (correspondence) truth is to be counted among the goals of science.

In its typical formulation, semantic realism claims that scientific theories should be interpreted realistically in the sense that they should be taken at face value. By taking scientific theories at face value, the semantic realist holds them to be true or false (or approximately true or false). That is, theories are not merely inference tickets (Ryle, 1949) or economically compressed descriptions of empirical data, as some logical positivists thought. It is part of the doctrine of global semantic realism to claim that many factual and theoretical claims in our current science actually do refer (i.e., reference is not merely an idealized goal but is in fact realized), though we think that this is a claim that should be reformulated for disciplines that make modest progress; currently it could plausibly be argued that psychology has, as yet, produced few theories that adequately meet this goal.

Even with these brief remarks, it should be clear enough that truth is directly implicated in the two realist theses of axiological and semantic realism. In this sense, realism needs truth.

Even though scientists regularly speak of the truth (and falsity) of knowledge claims, and semantic realism standardly maintains that our best theories in the mature sciences are literally, or approximately, true, there are times when this characterization is inadequate, especially for the social and behavioral sciences. Following Mäki (2005), we think that a particular local formulation of realism is helpful here. The thought is that rather than take our best theories to be true, or approximately so, a realism that is sensitive to the growth of scientific knowledge should accept the view that our theories might well be true in the future, if not right now. This will certainly be the case for the vast majority of scientific theories when they are first conceived. Therefore, it is more realistic to nominate our theories as candidates for truth rather than ascriptions of truth. Consistent with this, truth should be understood as an orienting ideal (a point we emphasized in our original article; Haig & Borsboom, 2012), which we approximate by fashioning, justifying, and modifying our theories. Because we cannot expect immediate truth in science, truth should be understood as a distal goal, not a proximal goal. Having said that, as we suggested in our original article, many mundane statements in scientific papers are aimed at literal truth (e.g., “20 students participated in this research”) and are naturally constructed using correspondence truth (i.e., the previous sentence is true if there were in fact 20 students who participated in the research).

As a result of realism’s natural adherence to axiological and semantic theses concerning truth, we think that a suitable notion of correspondence truth should be considered a part of scientific realism. For, there will be facts of the matter that make our truth-nominated theories true or false, even though we might not have strong grounds for determining their truth value. What matters is that our theories be given a decent opportunity to be judged true. And here, a commitment to the thesis of institutional realism is relevant because considerable resources of human intellect, time, money, and other forms of institutional support are needed for inquiry to be undertaken successfully.

The merits of correspondence theory

So far, we have asserted the thesis that realism, broadly construed, requires correspondence truth. Despite his defense of deflationary theory, Asay could in fact be read as being in agreement with this thesis, albeit in the most minimal way possible. He states:

If the sentence “Infants do not develop object permanence before eight months of age” is true, it is true because infants do not develop object permanence before eight months of age. If the sentence “Humans possess an innate, genetically determined faculty for language” is true, it’s true because humans possess an innate, genetically determined faculty for language. Most deflationists concede that there is a very weak sense in which these two sentences “share” the property of truth: they’re both true (e.g., Horwich, 1990, p. 38). (Asay, 2018, p. 387)

In our judgment, this relatively rich reading of deflationism is very close to a correspondingly austere reading of correspondence theory. That is, if correspondence theory is read parsimoniously, as simply saying that “all true sentences have truthmakers,” and deflationary theory is read as saying that “all true sentences share the property of truth,” the gap that separates the positions is rather small, in our view.

Asay argues that, given this minimal investment, deflationary theory is able to “do the work” that correspondence theory is typically employed to do. The reason for this is that, as we stated in our original article, “[f]or concrete individual empirical propositions, the theories will give the same verdict regarding their truth” (Haig & Borsboom, 2012, p. 274) because they both commit to realism. This property is used extensively by Asay and, in our view, forms the backbone of his argument. However, by accepting the fact that true propositions share the “property of truth,” Asay is, in fact, moving very close to a correspondence theory. One only needs to ask: can we characterize the family of true sentences with respect to a common property? It is, in our view, entirely reasonable to answer this question affirmatively, and to state that what all these sentences share is a truthmaker; that is, there is some state of affairs in the world that makes them true.

Asay claims, without further argument, that such a construal “is quite different from how substantivists see the matter, and quite different from how [sentences] share other genuine properties, such as being in English and being examples in an academic paper” (Asay, 2018, p. 387). This statement is, however, contestable. Despite their different levels of abstraction, it is perfectly clear what sentences like “52 undergraduate students at the University of Amsterdam participated in our experiment” and “cognitive tests feature a positive manifold” share in terms of the correspondence theory; namely, that there are matters of empirical fact that make them true.

We think that this categorization of sentences into true and false in terms of their correspondence to states of affairs is extremely useful in our communication in general. The reason for this is that it allows us to communicate and theorize about large classes of sentences, which are held together by the correspondence relation. The case of fraud is illustrative in this respect. Asay suggests that correspondence is not needed in characterizing fraud, because any individual case can be identified without explicit recourse to truth. He says, for example, “I am … claiming that Professor Cheat is, with intention to deceive, asserting that 80% of the cotton-top tamarins in the study exhibited behavior pattern p following stimulus s when he knows full well that it is not the case that 80% of the cotton-top tamarins in the study exhibited behavior pattern p following stimulus s” (2018, p. 388). Therefore, “we don’t need to bring in talk of correspondence truth to understand what fraud is, as Haig and Borsboom do” (p. 388).

This, however, puts the cart before the horse. Of course, one does not need a full spelling out of the correspondence theory of truth before one can investigate a given case of fraud, just as one does not require a philosophical course in scientific realism before setting up an experiment. However, in order to characterize fraud broadly, it is extremely useful to have a concept like correspondence theory in place, because that concept can serve to group a large class of true sentences together without specifying their particular meanings. This was, in fact, a major point of the discussion in our paper: we did not claim that correspondence theory is useful in the identification of fraud but in its general characterization.

Philosophical theories about realism and truth are meta-theories that are used to talk about true sentences and theories as they function in science and society. In our view, the deflationist position confuses the particular with the general. Because in any particular case, deflationary and correspondence accounts give the same verdict, and because deflationary accounts use no explicit correspondence relation, the deflationist concludes that we do not need correspondence. However, this misconstrues an important role for theories of truth: they are useful in order to characterize general classes of sentences that are homogeneous with respect to a particular purpose (e.g., characterizing fraud) because they have the resources to identify what these sentences have in common.

Conclusion

In the present commentary, we have defended the thesis that scientific realism requires correspondence truth and that correspondence theory does have important resources in addition to deflationary theory by providing grounds to generalize over large classes of sentences. Thus, while we accept that a “rich” interpretation of deflationary theory (in which “[true] sentences ‘share’ the property of truth”; Asay, 2018, p. 387) is sufficiently close to an austere interpretation of correspondence theory (in which true sentences share the property of having a truthmaker) to be able to do much of the work required in dealing with individual sentences, we conclude that generalizations about these classes of true sentences are served better by a correspondence theory.