Let is take stock. The arguments of the book can be seen as an upward path to modal empiricism: we start from an account of scientific practice so as to end with claims about what can be known about reality. After reviewing various accounts of theories and representation in chapter 2, I argue in chapter 3 that scientific models and theories convey norms for concrete applications. The question then bears on the aim of these norms. In chapter 4, I argue that this aim has to do with ideal success in all possible applications, and with unity. This prompts the question: how could such ideal success be justified? I argue in chapter 5 that assuming a notion of situated possibilities, it can be justified by induction on experience in the same way universal regularities are justified. In chapter 6, I argue that this does not justify scientific realism, because ontological commitments are still underdetermined by experience, and the successful extension of theories to new domains of experience needs not be explained by the idea that these ontological commitments are true: it can be justified by an induction on the models of the theory instead. This induction on model seems to require that the world is structured, and that our theories correspond to this structure, which makes modal empiricism similar to structural realism. However, I argue in chapter 7 that an important difference remains: according to modal empiricism, this structure is not absolute, but relative to our position in the universe. This is how modal empiricism is no threatened by problems of theory change, while structural realism is, contrarily to common assumptions.
The whole debate presupposes a robust, realist notion of truth. This is the kind of truth that, I believe, cannot be attained. Such notion, associated with semantic realism, is also accepted by anti-realists such as van Fraassen. According to him, theories indeed consist in literal descriptions of reality, but at the same time, we should not believe these descriptions: we should only accept them. Acceptance involves being committed to a research program, believing that all observable phenomena can be accounted for without giving up the theory, being immersed in the theory and talking as if theories were true, but it does not require actually believing that it is true (van Fraassen 1980 pp. 12, 88, 202).
Practically speaking, this notion of acceptance is not really different from belief, since it is not temporarily and guide our actions and discourse in the same way (Horwich 1991 Mitchell 1988). In particular, if we adopt a pragmatic conception of truth and belief, the two become indistinguishable. This gives us a motivation for adopting such a conception of truth, instead of maintaining a gap between acceptance and belief. This is the topic developed in chapter 8 of Modal Empiricism.
Correspondence and Pragmatic Truth
Realists typically identify truth with correspondence to reality. Although it seems intuitive, when we try to make sense of such a “transcendental” relation, all we can do is bring in more representations, and the relation of correspondence ends up being a relation between various representations (one of an agent’s belief state and one of reality for instance). Claiming that this represented relation itself corresponds to a “real” relation of correspondence leads to an infinite regress, and one could wonder if the approach makes sense at all.
Does it mean that truth is nothing but an internal relation of correspondence between our representations? We seem to lose sight of the world by adopting such a conception of truth.
This is where pragmatic truth enters the picture. According to the pragmatist, in order to understand truth, we should focus on the norms of inquiry, have a look at how our representations are used, and on what grounds they are accepted or rejected. In this respect, not all representations are on an equal footing. For example, in science, the relation between theoretical models and data models is asymmetric: theories are tested against data, and not the other way around. Representations that are closer to experience generally take (or should take) priority. Even when scientists cast doubt on experimental results because they contradict well accepted theories (which happened during the OPERA experiment), the resolution involves checking the experimental setup and the way data was produced, that is, bringing in more information from experience, and not simply adjusting the data to the theory by fiat. Furthermore, a scientific theory only gains this kind of relative authority on data when it is highly confirmed by past observations (for a similar view, see Israel-Jost 2015).
In light of this, it makes sense to identify truth with ideal empirical success: a belief is true if it corresponds to an ideal representation of its object, one that could not be defeated by any possible experience. This is the pragmatist notion of truth.
Why Semantic Realism?
Semantic realism is often uncritically accepted in contemporary philosophy of science because the alternative once proposed by logical empiricists turned out to be impracticable. Logical empiricists wanted to analyse the content of theories in terms of a vocabulary of observations, which looks like a reinterpretation. When a scientist talks about electrons, she seems to be talking about real entities that populate the world, and not about detection instruments or their outcomes. So, if we want to be faithful to the way scientists talk, we should accept that electrons exist. Note, however, that this is an argument against reductive semantics, not against pragmatic truth. The failure of reductive analyses does not entail that truth must be transcendent. More positive arguments for semantic realism are based on Kripke (1980) and Putnam (1975)’s externalism about meaning, but as I explain in the book, I do not think that they are conclusive (particularly when it comes to unobservable entities).
I believe that there are many reasons to reject a transcendent theory of truth. One of them has to do with the fact that theories are best understood as families of models, and that the domain of application of models must, in principle, be accessible to its users (Ruyant 2020). All user-centred aspects of representation presented in chapter 2 are relevant in this respect. Another reason is that the way scientist talk does not necessarily correspond to a realist semantics. Take quantum mechanics for example: there are many ontological interpretations of the theory (Belot 2012). A consistent semantic realist should say that they are all different theories, and that quantum theory is no theory at all. This conclusion is endorsed by some philosophers (Maudlin 2018). Coffey (2014) goes as far as saying that different metaphysical accounts of laws of nature should give rise to distinct physical theories. This is clearly at odds with scientific usage of the world “theory”, and physicsits do not seem as concerned as metaphysicians about finding the right ontology, or the right metaphysics for laws of nature. It seems to me that scientists consider theories to be equivalent if their applications cannot make any different in any possible context of use, that is, that the way they identify the content of theories is generally pragmatic.
Other observations cast doubt on the idea that ordinary scientific discourse is implicitly realist. For example, you can hear physicists claiming that Newtonian mechanics is true “within its domain of validity”. This relativisation of truth to a domain of experience is not clearly compatible with semantic realism. Finally, the standard textbook formulation of quantum theory explicitly mentions measurement, an epistemically loaded notion, and this does not seem to cause much trouble from a scientific point of view. Realists are unsatisfied, and propose various ways of eliminating this notion from the interpretation of the theory, but it is far from clear who, in these matters, is really interpreting the theory “at face value”, and who is imposing reinterpretations that fit their agenda.
Since semantic realism is largely responsible for all the difficulties of scientific realism, such as the problem of underdetermination and the pessimistic induction, it might be worth looking for an alternative semantics for scientific theories.
A Pragmatist Alternative
According to Peirce (1931 5.565), a true statement is a statement that would withstand doubt, were we to inquire as far as we fruitfully could on the matter. Pragmatic truth can be associated with a notion of ideal success. A complementary account associates truth with norms of assertion and inquiry: asserting something (and so, claiming that it is true) commits us to providing reasons in support for our claims and being held accountable for its implications, for instance (Misak 2007). Empirical adequacy is a notion of ideal success that plays a normative role for inquiry. It is fit to play the role of theoretical truth. If a theory correctly accounts for all possible manipulations and observations we could make in a domain of experience, then it would indeed “withstand doubt, were we to inquire as far as we fruitfully could on the matter”, and believing that a theory is empirically adequate is nothing more than believing that it would do so.
If we accept this, modal empiricism becomes the position according to which the aim of science is to produce true theories, now in a pragmatic sense of true. All the arguments of this book can be carried over to this new version of modal empiricism.
This notion of pragmatic truth does not have the same implications as the one that a semantic realist would adopt. However, it offers exactly the same linguistic resources. The modal structure of interpreted models can easily be described as a causal structure if one adopts a counterfactual theory of causation, for example. One could even identify the nodes of this causal structure with unobervable objects, even when the associated symbols are not directly interpreted empirically. They usually have a theoretical name, and one could attribute dispositions to these objects on the basis of their functional role in various models.
This kind of interpretation can help make sense of scientific discourse, where causal talk is ubiquitous, and where the existence of unobservable entities is not questioned. However, this discourse is no more metaphysically loaded. For a modal empiricist, the “causal relations”, “objects”, “properties”, “dispositions” or “laws” that can be read off of the structure of scientific theories do not describe the fabric of the world, because the modal structure of a model interpreted in context is ultimately a structure of possible manipulations and observations, within a limited range. This relativity to potential users entails that causal structures and objects are not “real” in the realist sense.
This anti-realist stance is not synonymous with lack of objectivity. Scientific objectivity is sometimes understood in terms of context invariance, which is a feature of modal empirical adequacy (accuracy in all contexts), even though this objectivity might not be absolute because of limitations in the contexts to which we have access.
An important aspect is that this notion of objectivity is orthogonal to the fact–value dichotomy. It rather follows the concrete–abstract dichotomy. If models are not purely descriptive at the local level, but also performative, the fact that a model is applicable across various possible contexts does not make this model descriptive. It only make this model more abstract: disconnected from both particular facts and values, and only concerned with generic types of facts and values. An abstract model is still indirectly performative, because it regulates its concrete applications.
Abstract values are more likely to be shared, which could be associated with a certain objectivity, this time understood in terms of value-neutrality. We could associate epistemic values or rationality with the most abstract kinds of values.
However, abstracting away from local values is not the same as adopting a purely descriptive approach, so we can have objectivity without adopting a realist perspective. Embracing more potential contexts actually brings us further away from local facts as much as it brings us further away from local values. Therefore, objective truth should not be equated with correspondence to the facts, but with ideal instrumental reliability when handling local purposes and local facts, whatever the purposes and facts involved.
This can be associated with a particular understanding of abstraction. According to a pragmatist stance, an abstract representation, such as a scientific theory, does not float freely in a Platonic space: it is indexical, and its content can only be interpreted in terms of its potential applications.
Modal Empiricism and the Role of Metaphysics
I believe that this kind of view can be used to propose a revisionary metaphysics for science.
Assuming that pragmatic truth also applies to metaphysical discourse would mean that this discourse must be connected to potential experiences, at least indirectly. Metaphysical claims enjoy a greater level of generality than any other, and I suspect that they could be reinterpreted in terms of reflexive principles that regulate the way we should represent the world. For example, the notion of metaphysical determinism should be reinterpreted in terms of the principled predictability of systems by agents representing them. I examine how this affects the debate on free will and determinism in the book, and how it connects it to more tangible and practically relevant considerations, while keeping it close to its traditional understanding. Similarly, the notion of metaphysical reduction could also be reformulated in terms of the compatibility of fine-grained and coarse-grained representations of the same phemomenon, thus connecting the metaphysical notion of reduction to potential representational uses. Again, I explain in the book that this pragmatist reformulation could be fruitful.
In this view, metaphysical questions are not purely empirical. They are so to speak “meta-empirical”, because they demand that we clarify the relationship between our concepts, representation and experience. One could say that a pragmatist approach blurs the distinction between metaphysics and epistemology. I doubt that a traditional metaphysician will be totally satisfied with these reformulations, but this is part of a pragmatist stance.
This stance is suspicious towards traditional metaphysics. It implies that abstract representations should not be interpreted purely descriptively, but also normatively: an abstract representation is a norm for possible concrete representations. Situated representation, potential or actual, should always take centre stage in philosophical analysis. Good concepts tell us how we should think in particular contexts. I strongly suspect that some metaphysical questions only occur because we take our abstract representations too seriously, and forget their normative dimension with regard to concrete representation. According to a pragmatist, even metaphysics should have indirect practical implications.