Chap. 7: Theory Change

As we have seen in the previous article, Chapter 6 of Modal Empiricism presents various criticisms of the realist arguments. However, one of the most influential reason to doubt scientific realism is the so-called pessimistic induction (Laudan 1981). It rests on the observation that many theories of the past, even successful ones, are now replaced by better ones that posit a completely different ontology, and so, we have good reasons to think that the same will happen with contemporary theories, no matter how successful they are. A related idea, also supported by the history of science, is that there are probably many yet unconceived alternatives to our theories (Stanford 2010).

The standard realist responses to the pessimistic induction consist in limiting realist claims: to “mature” theories only, to the posits that are essentially involved in novel predictions (Psillos 1999), to the entities with which we can causally interact (Hacking 1983), or to the structure of reality, and to argue that these are not lost in theory change.

According to structural realism, we can only hope to know the relational structure of reality, not its fundamental nature or ontology, or alternatively, this nature/ontology is entirely structural (Ladyman and Ross 2007 French 2014). This would solve the pessimistic induction, because one can claim that even if past theories were wrong about the ontology of the world, their equations got its structure right, and these equations are often used to build new theories, so they survive theory change.

Modal empiricism is not very far from structural realism in some respects. In chapter 6 of the book, I claim that an induction on models, which is required to justify the adequacy of theories and account for their novel predictions, presupposes that the world is structured in some way. However, in chapter 7 of the book, that I will present here, I argue that important differences remain between structural realism and modal empiricism, and that modal empiricism is better equiped to face the problem of theory change.

Structural Realism and Newman’s Objection

In order to respond to the pessimistic induction, the structural realist must explain in what sense there would be a structural continuity between successive theories, so that we could say that abandoned theories were already (approximately) right about (part of) the “structure of reality”. According to some authors, there is no such structural continuity between successive theories, or not always (Psillos 1999 ch. 7 Stanford 2003 Saatsi and Vickers 2011 …).

Structural realists can focus on well-confirmed empirical relations between phenomena to make their point. These empirical relations, at least, are likely to be retained in some form in new theories, otherwise these theories would not be empirically adequate. They can also appeal to Post’s principle of correspondence, according to which new theories should “degenerate” into old theories under conditions where the old theories have been confirmed by tests. For example, the theory of relativity “degenerate” into Newtonian mechanics in flat spacetime. The risk with this strategy is that by focusing too much on empirically confirmed structure, structural realism could collapse into a a form of empiricism, according to which theories merely “save the phenomena” (that is, the “structure” or regularities of our observations).

This problem is related to an influential objection against structural realism, dubbed “Newman’s objection” (in reference to an exchange between Newman and Russell). The objection goes roughly as follows: given a set objects, any possible relational structure between these objects “exists” mathematically speaking, in the sense that it can be defined in set theory. So saying that reality has a given structure (the one described by our theories) is not saying anything, unless this structure is qualified: this structure must “exist” in a stronger sense than mere mathematical existence. Saying that this structure is physical rather than mathematical is not saying much more. The problem is that if one qualifies this structure in ontological terms, in terms of natural categories for instance, then we are back to standard realism, and it is not clear that such structure survives theory changes. What old theories take to be natural categories, such as "caloric fluid", are often not considered natural anymore by new theories. On the other hand, if one qualifies the structure in epistemological terms (in terms of observational, measurable relations or categories), then this is nothing more than a version of empiricism, according to which the structure of our observations, or context-relative “phenomenological laws”, are preserved by our theories. It seems that there is no middle ground between full blown realism and empiricism.

Structural realists have proposed a way out. According to them, the only qualification of the structure that we need beyond its empirical interpretation is that it is “modal” or “nomological”. It describes relations of necessity between phenomena, akin to laws of nature, and this is the sense in which they are “real”, that is, they exist in a more robust sense than mere mathematical existence. This option corresponds to so-called ontic versions of structural realism.

At this point, it seems that the difference between modal empiricism and structural realism is very thin. Ladyman (2004) actually suggests that a version of empiricism committed to natural modalities would be nothing but structural realism. However, I think that important differences remain due to the fact that according to modal empiricism, possibilities are situated. This means that the structure of our theories does not necessarily correspond to a “real structure” that would be independent from our position in the universe. As explained in chapter 5, the necessity involved is not absolute, but relative. As argued in this chapter, I believe that for this reason, modal empiricism is distinct from structural realism, and this is precisely why it is not threatened by the pessimistic induction, while structural realism is.

“Real” Relations and Theory Change

In order for a structure to count as real, it must be “mind-independent”: it must not be relative to us and the way we represent it. Ontic structural realists often claim that reality is essentially structural, in the sense that relations are more fundamental than the objects they relate, which are merely determined by the relations in which they take part (French 2014). As just explained, their way out of Newman’s objection, that all structures exist mathematically speaking so that structural realism would say nothing interesting about the world, consists in claiming that the structure of reality is “modal”, that it correspond to the laws of nature (or something similar). A physical structure, contrarily to a pure mathematical structure, is potentially manifested by “existential witnesses”. I briefly examine these claims and some objections in the book.

The main problem faced by structural realism concerns, according to me, the interpretation of “modal”. If the structural realist refers to relations of absolute necessity, associated with fundamental laws of nature, then it is not true that old theories describe the real structure of the world. Old “laws” often become relative to a context. For example, Galileo’s law of free-fall, according to which all objects accelerate towards the centre of the Earth at 9.8 m/s², is only valid in a limited range of circumstances, when these objects are located at the surface of the Earth. It might be true that Newtonian’s theory “degenerates” into Galileo’s law when circumstances are specified, but this does not entail that Galileo’s law describes the “modal structure of reality”. The fact that we are located at the surface of the Earth is pure contingency, and Galileo’s law is relative to this contingency, so it is not “absolutely necessary”.

The same goes for most theories of the past: in light of new theories, their validity becomes relative. By induction on past theories, there is no reason to assume that our current theories describe relations of “absolute necessity”: they are probably relative to a limited range of contexts to which we happen to have access. So, if structural realism asserts that the equations of our theories correspond to relations of absolute necessity, it cannot respond to the pessimistic induction, contrarily to what is often claimed.

On the other hand, if the “modal structure” described by our theories is understood as relative to a context, then structural realism cannot claim to be a realist position. Relations of relative necessity cannot count as real, because they are relative to us, and in this case, to our context (flat space-time, the surface of the Earth, etc.). If this context was made explicit and integrated as a precondition for the application of the theory, then the theory could perhaps be said to correspond to relations of absolute necessity. But in general, we can only make this context a bit more explicit in light of new theories (for example, by saying that Newtonian mechanics is only valid in flat spacetime). This means that current theories are probably relative to a context, to conditions of applications, that we are yet unable to make explicit. All we can do is refer to our position. For this reason, our theories cannot be said to correspond to “real relations”.

This is where the main difference between modal empiricism and structural realism lies: according to modal empiricism, the modal structure described by our theories is not absolute, but relative to unknown conditions of application associated with our contingent position in the universe.

Relativity and Fundamentality

I examine in the book claims that are made by Ladyman and Ross (2007) that structural realism is compatible with a notion of relative necessity. These claims are made about specific sciences, such as biology or chemistry, assuming that the set of circumstances to which theories of these disciplines are relative can be made more or less explicit (by means of what they call "locators"). However, they tell a different story in the case of physics. Rightly so: if theories of physics were about relative necessity, structural realism could not be a realist position. So, the pessimistic induction still threaten structural realism.

I also examine various options for responding to the pessimistic induction that are available for structural realists, and scientific realists in general. One of them is to be a fallibilist. I think that this option is problematic, because strictly speaking, being a fallibilist entails assuming that our theories are probably false. They cannot be approximately true either, because relations that are relative to peculiar circumstances are no approximation of absolute relations. Being true “in a restricted domain” does not make sense for a realist with a robust notion of truth.

Other philosophers think that we are in a better position now that we were in the past, so that even if most past theories have been replaced by better ones, this might not be true anymore in the future (Ruhmkorff 2013 Fahrbach 2011 Devitt 2011 …). The main problem, I think, is that we lack an external viewpoint in order to make this judgment. A few centuries ago, the universe was thought to be much smaller than it is thought to be now, and scientists of the past could entertained the same idea. I see no reason to assume that we are in a different position now. For this reason, it is reasonable to be a modal empiricist.

Modal empiricism is compatible with a notion of progress: the range of contexts to which our theories are applicable becomes larger with the development of science. However, applying to a larger range of context does not mean being free of any context, nor having access to "most possible contexts", whatever that could mean.

We can see that the notion of realist truth, expressed in terms of correspondence with a mind-independent reality, plays a role in some of the arguments presented in this chapter. A pragmatist assuming a different notion of truth would not be affected in the same way by the pessimistic induction. In the final chapter of the book, I argue that adopting a pragmatist conception of truth and claiming that our best theories are true or approximately true is compatible with modal empiricism, and that modal empiricism actually provides a nice way of fleshing out the pragmatist notion of truth. I will present it in the next and final article of this blog.