1. Introduction

1 The notions of preference and utility have had a strikingly enduring role in the history of economic thought, and they are still pivotal in most contemporary economic studies (Broome [1991], Hausman and McPherson [2006], Hausman [2012]). In the late 1930s, Samuelson famously advocated that economic studies should abandon these notions and develop a theory of consumer’s behavior “freed from any vestigial traces of the utility concept” (Samuelson [1938]). Paradoxically enough, far from contributing to the abandonment of the notions of preferences and utility in economic studies (Hands [2013]), this attempt historically inaugurated what was to become a prominent economic theory of preferences and utility: the “Theory of revealed preferences”. Samuelson’s attempt was anchored in what is now known as the weak axiom of revealed preference. The latter is sometimes presented as reducing the choice function to the consumer’s behavior in such a way that “the preference relation R has no independent meaning” (Hédoin [2016]). In the same vocabulary, Houthakker’s [1950] stronger axiom, the strong axiom of revealed preference, can be presented as allowing to capture the meaning of a choice behavior.

2 Obviously enough, in the above formulations, the notion of meaning is used in an intuitive, unclarified sense, rather than in a philosophical sense. They nevertheless suggest that the concept of meaning might have an interesting role to play in the elaboration and interpretation of the theories of preferences and utility – a conjecture that, as far as we know, has not received the attention it deserves in the literature. In this article, we would like to argue that there are strong (though largely unnoticed) conceptual links between the economic concept of utility and the philosophical understanding of the notion of meaning. Our aim is to provide the first elements of exploration of these conceptual links and to develop some of their prominent theoretical and practical implications.

3 The article is organized as follows. Section 2 introduces a general definition of meaning. According to this definition, any account of an act that can be endorsed by the actor performing this act qualifies as a meaning. Section 3 clarifies how utility qualifies as economic meaning according to our definition. Section 4 explores what we see as the main theoretical implication of our approach, by showing that this theory sheds new light on the debate on the justifiability of the axioms of the theory of choice. Section 5 moves towards more applied considerations and spells out three practical implications of this framework. Section 6 briefly concludes.

2. A general definition of “meaning”

4 In this section and the following one, we introduce a general definition of a central notion for our argument: the notion of meaning. Indeed, although the term “meaning” is widely used in everyday conversation and plays a key role in many philosophical theories, there is currently no consensual definition of the concept of meaning in the literature. We claim neither to review all the philosophical, linguistic and other theories of meaning, nor to introduce a complete new theory outcompeting all the others. We introduce a simple definition and we argue that it encompasses the most influential theories of meaning that have been produced in the past few decades. This section will thereby set the stage for the remainder of the argument by anchoring it in a clearly spelled-out definition.

5

Definition:
In this paper, we define “meaning” as any verbal, written or mathematized account of a given act, such that the agent or agents performing this act can endorse this account. For short: meaning is endorsable account.

6 In this definition, we define the notion that an agent i endorses an account a as encompassing two things. First, that i takes a to be a relevant, truthful or adequate (the three adjectives being considered equivalent for our purposes) account of his/her action. Second, that i can grant to a a role in the conscious determination of his/her own future behavior.

7 Based on this definition of endorsing, we then define meaning in explicitly modal terms. Neither do we claim, as an indicative definition would, that a meaning is an account that happens to be endorsed, nor do we claim, as an imperative definition would, that a meaning is an account that should be endorsed. What we claim is that a meaning is an endorsable account – our definition hence takes meaning to be a modal notion. In other words, what we claim is that an account of an action qualifies as a meaning as long as, if some specific (yet unspecified) conditions held, then it could or would be endorsed. Like any modal notion, meaning is therefore anchored in counterfactual reasoning. It can happen that an account a is endorsable by i, but i nonetheless fails to endorse it, because of bad-faith, a lack of lucidity or information, or any other distorting factor. If i were lucid, truthful and informed enough, he would endorse a. The importance of such counterfactual idealizations in theories of meaning is largely acknowledged in the literature in philosophy of language, and has been emphasized most prominently by Habermas [1984].

8 An important objection to such a modal definition would be to retort that it is bound to be too vague unless it specifies exactly which counterfactuals (or, equivalently, which specific conditions) are to be considered in the presupposed idealization. This would be too exacting a requirement, however, for two reasons:

9

• First, the precise nature of the counterfactuals to take into account may vary from one case to another: our definition is purposefully very general, and aims at encompassing a large variety of accounts of behavior, whose endorsability might hinge upon very different parameters determining very different kinds of counterfactual idealization. Recall that our point in spelling out a definition of “meaning” was to clarify it to enable us to use philosophical insights on meaning to enrich economic reasoning. A definition that would be too specific in delineating precisely the counterfactuals to take into account could cage the notion in an exceedingly narrow definition and in turn estrange it from the philosophical understanding of the notion.
• Second, the endorsability criterion as we conceive of it is not a turnkey practical criterion. It is rather a conceptual criterion whose specification is the crux of most philosophical debates on meaning. By saying that, we do not mean that this criterion is bound to be empirically irrelevant. We rather mean that there is room for conceptual discussions about which kind of account should be considered endorsable and which should not: one does not always has to work with one specific individual i in the flesh to be able to usefully contribute to debates about endorsability.

10 By defining meaning as endorsable account, in an explicitly modal but nonetheless general sense, we accordingly claim to fulfill three objectives. First, we claim that this definition captures a general structure, which is shared by all theories of meaning despite their differences. Second, we claim that it does so by providing an explanatory content (through a reference to endorsement and modality) to the otherwise often undefined term “meaning”. Third, we claim that this definition manages to remain general enough to leave room for both philosophical and practical debates about the relative merits of different approaches to meaning and behavior.

11 In a sense, our definition above echoes Weber’s: “we shall speak of “action” insofar as the acting individual attaches a subjective meaning to his behavior] [1978, p. 4]. But, as argued by Habermas [1984], Weber’s theory is buttressed on an exogenous, intuitive and non-clarified notion of meaning. By contrast, our definition does not take any exogenous notion of meaning as given. We rather explicitly take meaning as the object of our definition. Moreover, as Habermas also noticed, Weber’s approach is predicated on a specific theory of consciousness, whereas our definition is not anchored in any such theory.

12 That said, before proceeding, it is useful to emphasize several noticeable aspects of our definition.

13 First, we take meaning to characterize actions, acts, behaviors or patterns of behavior, rather than words, sentences or other linguistic entities [1].

14 Second, in our definition, we purportedly lump together the notions of act, actions, behavior and patterns of behavior to emphasize that our general definition does not presuppose any specific theory of the ontology of this domain of object (Livet [2017]).

15 Third, we take meaning to be materialized in verbal, written or mathematized accounts, which implies that we take meaning to be neither an ineffable entity nor an entity necessarily expressed in a specific linguistic medium such as ordinary language. We use the general and admittedly vague term “account” in order to avoid “representation” and its unfortunate connotations associated with the so-called “modern” philosophy of the subject (Habermas [1994]).

16 Fourth, we stress in this definition that the agent (s) concerned is/are not necessarily involved as individual, insulated subject (s). A candidate meaning can be an account of a collective action performed by a group, and the relevant agent to endorse or reject the account can be a group as a whole, which deliberates and collectively expresses or reveals a collective willingness to endorse or reject the account. Thoroughly exploring how individuals partake in such collective processes and how and to what extent and in which sense individual willingness translates or is aggregated into collective willingness is not the focus of this article. Our point is simply that our general definition does not presuppose an atomistic, individualist ontology; it rather can encompass theories of collective agency, and can account for the importance of supra-individual entities such as language in the genesis and nature of meaning. Similarly, our general definition neither presupposes nor exclude that the agents at issue can be single, dual or multiple selves.

17 Fifth, related to the fourth point, we stress that our general definition does not presuppose any theory of the ontological status of mental states or other psychological entities. Articulated as it is, our definition can therefore encompass both so called “internalist” accounts of behavior in terms of mental states or other psychological entities and so-called “externalist” accounts.

18 We claim that this definition encompasses the most influential theories of meaning produced in the past few decades in the philosophical literature. However, it is not a place in this article to develop the corresponding purely philosophical demonstrations [2]. Let us simply illustrate the logic underlying our contention in the case of intentionalist theories of meaning, because these are certainly the most commonsensical approaches to meaning. Intentionalist theories claim that meaning is first and foremost a matter of what a speaker intends by using a given linguistic tool in a given context. This approach to meaning parallels the ordinary attitude that consists in accounting for what one said by spelling out what one meant by saying it, possibly so as to justify what one said. Spelling out one’s intention or imputing an intention to someone is a prominent means for anyone to account for a linguistically mediated action, in such a way that the actor can endorse or reject this account. Reformulated in this way, the intentionalist approach to meaning therefore appears to be a specific variant of our general definition.

3. How utility qualifies as economic meaning

19 We now come to the core of our argument, which is that the economic notion of utility is also subsumed by our general definition of meaning. For that purpose, terminological clarifications are in order. In the economic literature, the notions of choices, preferences, utility, welfare and wellbeing are strongly associated, but the nature of the links between them can be variously interpreted. Clarifying these links is pivotal to unfold our argument unambiguously.

20 We will take for granted here that “welfare” and “wellbeing” are synonymous. This assumption could be disputed from a philosophical point of view, because one might argue that wellbeing is a more general notion. But in the present article we don’t think that this philosophical discussion is useful, and we leave it aside. We also take the notion of choice as a basic, non-problematic one. Again, from a philosophical point of view this assumption is not innocuous, but the philosophical discussion of the ontology of choices falls beyond the scope of this article.

21 We therefore focus our terminological clarification on the concepts of preferences, utility and welfare. The link between choices and preferences can be understood in what we will call a “substantial” and an “analytical” sense. In the substantial sense (thereafter: prefer^s), the link captures the idea that economic agents choose what they prefer, in a sense of “prefer” which is exogenously given, for example, by an exogenously given theory of motivation, theory of practical reason or ethical theory. By contrast, in the analytical sense (thereafter: prefer^a), saying that individual i prefers x to y is simply a reformulation of the fact that, when facing an alternative between x and y, s/he chooses x. According to these definitions, individual i can prefer^sy to x, but nonetheless choose x rather than y and, in that sense, be said to prefer^ax to y. One can accordingly understand the links between preferences, utility and welfare in a purely analytical sense: in this sense, the fact that i chooses x rather than y is reformulated by saying that i prefer^ax to y, and that x brings to i more utility^a than y, and that i’s welfare^a is increased by his choosing x rather than y. In this analytical case, the bridge between the various formulations is each time a matter of pure terminological substitution. By construction, the prefer^a relation can be interpreted as a complex bundle of so-called “inner preference” over states of affairs and beliefs about the likelihood that these states of affairs will occur (Hausman [2012]). But the reformulation of choices in terms of preference^a places itself upstream this possible decomposition. “Inner preferences” can be inferred from preferences^a only given premises concerning beliefs (in a process that Hausman [2012] terms “belief-dependent revealed preferences”). We forcefully emphasize that, as a consequence, preference^a does not capture anything closely akin to what the terms “preference” and “prefer” capture in ordinary language. Similarly, it is not by any means guaranteed that welfare^a captures anything akin to what the term “welfare” refers to in ordinary language or in philosophical theories. It is not in any way guaranteed either that welfare^a captures a concept coherent enough for rigorous propositions to be articulated in terms of welfare^a (except for propositions simply translating propositions originally articulated in terms of preference^a), and normative implications can consequently not be derived from propositions in terms of welfare^a (a point forcefully emphasized, using another terminology, by Hausman [2012], chap. 7). Alternatively, one could define, independently from choices, substantial notions of welfare, utility and preference, and make the substantial claim that economic agents strive to maximize their welfare^s through maximizing their utility^s gains by choosing what they prefer^s. This analytical vs. substantial dichotomy allows defining what one might call two “threads of concepts”: choice-preference^a-utility^a-welfare^a vs. choice-preference^s-utility^s-welfare^s. The first thread is a purely translational one, where the terms “preference”, “utility” and “welfare” are not used in their usual sense and the conceptual coherence of the corresponding concept is not guaranteed. The second threat embodies full-fledged theories claiming to explain or account for choices in terms of concepts of welfare, utility or preference more tightly linked with the usual semantic content of the corresponding terms. Notice that these definitions raise questions such as: is it possible for the two threads to cross or merge, for example, by defining “utility” in an analytical sense translating a “preference” understood in a substantial sense? In the remainder of this article, we will leave aside such intricate issues and simply take the two “pure” threads into account.

22 It is important here to see that the possibility to use the concepts of preference, utility and welfare in the analytical way as we define it is not ruled out by Sen’s ( [1973], [1974]) criticism of the Samuelsonian theory of revealed preferences (a theory to which we will come back extensively in section 4). In his criticism, Sen argues that several patterns of preferences can lead to identical patterns of observable choices, which implies that, contrary to what Samuelson assumed, observing behavior does not provide enough information to infer preferences (an argument further elaborated by Davidson [1980]). This point does not block the possibility to define preferences^a on the basis of choices. It simply implies that any given pattern of choices can be represented by infinitely numerous preference^a orderings. Sen even explicitly admits the possibility of a translation from the choice vocabulary to the preference one through a “redefinition of the word “preference”” away from its “usual sense” ( [1973], p. 243). His argument (Ibid., sections II-IV) takes “preference” in its ordinary sense, but it does not block the possibility to use the term in a preference^a sense. In his words “The difficulty arises in interpreting preference [defined as the binary representation of individual choice] as preference in the usual sense with the property that if a person prefers x to y then he must regard himself as better off with x than with y” ( [1973], p. 253). Using our terminology, Sen claims here that difficulties arise when one jumps from preference^a to utility^s and/or welfare^s.

23 Our analytic vs. substantial distinction is also reminiscent of Broome [1991], who argues that in contemporary economic discourses cohabit two meanings of the term “utility”. The first one, which he calls “technical”, defines utility as “that which represents preferences”. The second, so-called “underground” sense makes it synonymous of “good”. In this article, Broome does not talk about the link between choices and preference, he is rather concerned only with the link between preferences and utility. If one admits that Broome also understood the link between choices and preferences in what we call the analytical sense (that is, if one admits that Broome talks about preference^a – a conjecture that is admittedly not buttressed on any claim of Broome’s, who appears to be agnostic concerning the link between choice and preference), then his “technical” sense of utility is what we term utility^a. Indeed, he writes: “of any pair of alternatives, the [utility] function assigns a greater utility to the one that is preferred”. His “underground” sense is a version of our utility^s, anchored in an ethically grounded substantial theory.

24 In the following, we argue that, if the links between choices, preferences, utility and welfare are understood in an analytical sense, utility theory can – at least provided that some important conditions are realized – be encompassed in our general definition of meaning. By contrast, we argue that a substantial understanding of these links makes it more difficult, but certainly not impossible, to interpret utility as encompassed by our general definition.

25 In the analytical approach, utility modelling encompasses two elements: a representation of an action or a pattern of actions by a utility function, and an assumption of the classical axioms of the theory of choice. Because there are several variants of utility theory, characterized by different sets of axioms, and because a given set of axioms can also to some extent be formalized in various terms, let us choose here one specific, but emblematic formulation (Varian [1992], p. 95):

26 We consider an agent i facing a set X of states of affairs. Let ≽_i represent i’s preferences over X. The classical axioms of the theory of choice are usually stated as follows.

27

• Axiom 1: An agent’s preferences^a are complete: ∀ x, y ∈ X, either x ≽_i y or y ≽_i x, or both.
• Axiom 2: An agent’s preferences^a are reflexive: ∀ x ∈ X, x ≽_i x.
• Axiom 3: An agent’s preferences^a are transitive: ∀ x, y and z ∈ X, if x ≽_i y and y ≽_i z then x ≽_i z.

28 Associated with a Continuity condition [3], these axioms guarantee that preferences can be represented by utility functions (Debreu [1959]). Hence the dual content of the utility^a concept: at once representation and assumption of axioms.

29 Defined in its duality as above, utility provides a (mathematical) account of actions or bundles of actions. According to our general definition of meaning, it can therefore provide a meaning if it can be understood, and eventually endorsed or rejected by the actors whose actions are modelled by the utility function. We claim that the latter is the case because, or rather as long as, the classical above-mentioned axioms encapsulate simple properties characterizing choices, which can be articulated in ordinary language terms, such that the agents can adjudicate the soundness of their assumption.

30 Notice that the latter claim (that the axioms are credible when expressed in terms of choice) stands in contradiction with Sen’s [1973] argument that the axioms of the theory of choice owe their whole credibility to an implicit psychology of preferences conceived as mental states. But this argument of Sen’s is refuted by the so-called “neo-samuelsonian” approach to revealed preference theory (Hédoin [2016]). This approach takes advantage of Dennett’s [1987] theory of intentionality to conceptualize preference-based representation as a matter of delineating consistent patterns of choices by ascribing to economic agents a specific way to frame the choice problem they face (Ross [2005]). Our point here is not to champion this neo-samuelsonian approach: we simply use it to show that, Sen’s arguments notwithstanding, the axioms of the theory of choice can be conceived as simple and understandable even when articulated in terms of choices or, equivalently, of preferences^a.

31 In this reasoning, the dual structure of utility modelling and the representation theorems articulating the two aspects of the structure play a crucial role, because each aspect echoes an aspect of our general definition of meaning: on the one hand, the representation, and on the other hand, the endorsability requirement. This is reminiscent of Cozic and Hill [2015], who recently emphasized the semantic importance of representation theorems, but their argument was not anchored in an explicitly defined, philosophically qualified notion of meaning. Here we proceed one step further, by presenting a notion of meaning robust enough to flesh out this semantic role.

32 Our claim is hence that, thanks to the easy understandability and intuitive content of the axioms of the theory of choice, especially when they are articulated in terms of choices or preferences^a, utility-based modelling can provide a representation of behavior that can be understood, criticized or wholeheartedly endorsed by economic agents. This idea has two aspects. First, at a general level, we claim that because the above axioms are, in essence, easily understandable, utility^a can, in general, be conceived as a meaning, which we will term here “economic meaning” [4]. Second, more specifically, this very idea allows to define a standard to adjudicate whether specific instances of utility representation qualify as meaning. Indeed, utility representation comes out in various guises: utility can be defined on sets of states of affairs (as in the case of so-called “inner preference”), on lotteries, etc., and it can be anchored in marginally different axioms (axioms of expected utility theory, rank-dependent utility theory, prospect theory, etc.). The different variants of utility theory, based on specific axioms, defined on specific entities, applied to specific agents in a specific context, will vary in the extent to which they provide endorsable accounts of behavior – and therefore in the extent to which they qualify as economic meaning.

33 With this approach, we hence claim to provide a rationalization scheme for the practice of utility modelling. We claim that conceiving of utility modelling as a specific instance of our general notion of meaning captures a crucial part of the current, actual practice of economists using utility theory, especially when preference and utility are understood in the analytic sense. We thereby claim that this notion captures the bulk of current practices. But we also know that some specific applications will fall outside the scope of this notion, and we provide a borderline test, usable to characterize, on a clarified basis, what makes these applications different from the bulk of current practices.

34 If utility is understood in the substantial sense mentioned above, the requirement epitomizing our borderline test is more difficult to satisfy. Indeed, in the substantial case, utility modelling provides representations of behavior buttressed on the idea that economic agents choose what they prefer^s and thereby maximize their welfare^s. Here again, this approach provides an account of acts or behaviors that could be encompassed in our general definition of meaning if it demonstrably is endorsable by the actor (s) performing the act at issue. This putative endorsability is determined in part by the specifications of the exogenous substantial theory of welfare, utility and preference at issue. In the economic literature, hedonistic theories of welfare and motivations are prominent candidates to provide such exogenous substantial theories. For example, Kahneman et al. [1997] and, Kahneman & Sudgen [2005] argue that using the concept of utility means using the same informational basis as hedonistic utilitarianism. Similarly, Kolm [2004] criticizes utility-based approaches by claiming that using the concept of utility confines the analysis to taking economic agents’ pleasures or happiness and pains into account. Although most authors are more cautious and eschew drawing an explicit link between utility and pleasures and pains, it is not uncommon in the literature to read arguments introducing surreptitiously that methods based on the usage of utility carry ethical (often hedonistic) implications, as if that were self-evident and indisputable (Hausman [2010]). Adjudicating the credentials of these accounts and similar ones falls beyond the scope of the present article. For our purpose here, suffice it to notice that, because, as extensively studied by Fumagalli [2013], questions such as “does one always choose according to one’s preferences?” or “is one better off when one’s preferences are satisfied?” invariably surface when considering utility modelling, it would be hazardous to take for granted that accounts of behavior in terms of welfare^s, utility^s and preference^s are endorsable by agents. However, there is no principle reason to think that substantial approaches can never be endorsable.

35 Our approach therefore provides a rationalization scheme for these various understandings and uses of the concept of utility in economic theory, allowing to discuss their commonalities and to criticize some of their applications, without dogmatically excluding any approach as a whole.

36 Before closing this section, it is useful to stress that the endorsability criterion, lying at the core of our definition of meaning, does not require that the account whose endorsability is at issue exactly mimics the agent’s own prior understanding of her/his acts. Indeed, as noticed by Hausman ( [2012], p. 4), “[p] eople think about what they do and explain why others do things in terms of reasons”, whereas utility models and other economic approaches typically strive to account for actions in terms of beliefs and desires. There seems to be gap between first person accounts in terms of reasons and third person accounts, including utility-based ones, in terms that are different from the one used by actors themselves. However, the fact that agent i spontaneously or antecedently conceived of her/his actions in terms of reasons does not prevent her/him to endorse another account of her/his actions, articulated in other terms that s/he might conceive as providing a complementary or equivalent light on her/his own actions (this possibility is largely admitted, since it is a prerequisite for the very plausibility of so-called folk-psychological accounts of behavior). This point echoes what we said above concerning the conceptual aspect of the endorsability criterion. When one produces a utility model, it is not always possible to empirically test its endorsability with the agents whose actions are modelled in the flesh. This does not turn endorsability into an inapplicable criterion. Endorsability can be discussed at a theoretical level, or based on empirical generalization obtained with other individuals with similar relevant features in similarly relevant situations, possibly at other time points, etc.

4. Theoretical implication: the justifiability of the axioms of the theory of choice

37 In this section, we argue that understanding utility as economic meaning is useful to strengthen the positive justifiability of the axioms of the theory of choice.

38 To that end, let us start by explaining the structuring notions of the debate. A widespread argument to justify the usage of the axioms of the theory of choice is that they are minimal. For example, Kreps [1990] claims that they encapsulate commonsensical requirements (let us call this stance “the simple defense”). Indeed, the terms “choice” and “preference”, in which the axioms are formulated, seem to be simple and unequivocal because they are widely, and seemingly innocuously, used in ordinary conversation. Spelled out in these terms, the requirements encapsulated in the axioms appear unproblematic. However, a classical counterargument (let us call it “the empirical criticism”) to this defense is that myriads of concrete examples show that the axioms of the theory of choice are grossly at variance with empirical reality. This empirical criticism rejoins the literature on bounded rationality (Simon [1955]) and on decision biases (Kahneman, Slovic & Tversky [1982]), demonstrating the need to enrich models with empirical specifics. However, for the purpose of assessing the justifiability of assuming the axioms these approaches are inconclusive. Indeed, the axioms of the theory of choice are not empirically hypothesized, but rather stipulated [5] (Mongin [2000], [2003]). Factual counterexamples therefore do not disqualify them as they would disqualify empirical hypotheses. The simple defense and the empirical criticism are therefore both inconclusive because they treat axioms as empirical hypotheses; to pinpoint this common weakness, in the remainder of this article we will term these approaches “empiricist”. On the other hand, the simple claim that the axioms are stipulated rather than empirically hypothesized (a stance that we will call the “theoretical rebuttal”) obviously fails to justify this stipulation.

39 Empiricist approaches and their theoretical rebuttal draw a polarized picture. On one pole, empiricist approaches admit that the point of utility-based modelling is to catch a correct picture of the way economic agents behave or will behave. They therefore understand utility as a purely descriptive and/or predictive notion. And in so doing, they fail to provide a convincing account of the permissibility or impermissibility of stipulating the axioms of the theory of choice by ignoring that these axioms are stipulated. On the other pole, the theoretical rebuttal manages to account for the fact that the axioms are stipulated only by taking the point of utility to be to delineate the properties of what a rational behavior would look like (in particular, by identifying optimal solutions, that is, solutions that a rational behavior would capture). It therefore accounts for the fact that axioms are stipulated by being anchored in a normative approach to utility, and accordingly fails to account for the permissibility or impermissibility of stipulating the axioms because they are simply not concerned by empirical relevance.

40 This question of whether axioms that appear unrealistic can be justifiably postulated is actually a specific version of a more general debate on the status of models and their assumptions that have a long history, dating back at least to the classical economists. In this historical debate, the latter by and large defended that axioms are minimal, thereby siding with what we termed “empiricist” approaches (Caldwell [1982]).

41 Friedman [1953]’s instrumentalist argument that unrealistic hypotheses are innocuous and even recommendable if they allow producing valid predictions about economic phenomena has dominated and in effect shut down this debate over the second half of the twentieth century. Instrumentalist arguments can draw on the epistemology of physics to claim that “preference” and “choice” are “theoretical terms” (on a par with, e.g., “atom” or “magnetic field”). The axioms would capture features that theoretical terms owe to their relation with all the other concepts constituting the conceptual network within which they are used (Carnap [1966], Lewis [1970]). In this approach, there is not necessarily any simple, unequivocally identified relationship between, on the one hand, the theoretical terms and their constitutive properties (captured by the axioms) and, on the other hand, concrete ordinary idioms, objects, and patterns of behavior (Schmitz [2009]). Adjudicating the credentials of such an approach in its general application to economic modelling falls beyond the scope of the present article. For our purposes, suffice it to say that, in the case of utility axioms, the notions integrated in, or associated with the axioms also play a crucial role in the connection between theoretical models and empirical reality. This is true both in the analytical and in the substantial interpretations of the choices-preferences link. Whereas no one ever claims to directly observe atoms and magnetic fields, choices are observed in market studies, experimented in choice experiments, and reported in consumer surveys. Similarly, preferences are purportedly stated, for instance in contingent valuation surveys, or revealed, like in experimental economics. Conceiving choices and preferences as theoretical terms would entail losing a prominent and extensively exploited connection between theoretical construction and empirical reality. For economists aiming to study empirical reality, this would be a major loss.

42 The instrumentalist approach therefore fails to provide a satisfactory explanation of the justifiability of the axioms because it cannot explain both why:

43

1. Axioms can have counterexamples, and
2. The features that they capture appear to be commonsensical and minimal.

44 In addition to the incapacity of the instrumentalist approach to account for point (ii), the reliance of this approach on the notion that economics is first and foremost a predictive activity started in the 1990’s to be seen as a weakness, amounting to ignoring that a large part of economics is more concerned with theorizing, modelling and measuring rather than predicting.

45 This prompted a revival of the debate, resulting in the emergence of two competing theories: isolationism, mainly advocated by Cartwright [1989] and Mäki ( [1990], [2005], [2009]), and fictionalism, introduced and defended by Sugden [2000]. In their application to the justifiability of stipulating the axioms of the theory of choice, fictionalism and isolationism provide two competing solutions to bridge the gap identified above between the two poles of empiricism and its theoretical rebuttal. In what follows, we first explain how our framework allows elaborating a solution to the problem tackled by fictionalism and isolationism. We then present the fictionalist and isolationism frameworks, and investigate how these approaches relate to ours.

46 Sen’s ( [1973], [1974]) criticism of the Samuelsonian theory of revealed preferences provides the entry point for our theory of utility as economic meaning to provide an account of the justifiability of stipulating the axioms of the theory of choice satisfying conditions (i) and (ii). As already mentioned, Sen argues that several patterns of preferences can lead to identical patterns of observable choices. In other words, observing behavior does not provide enough information to infer preferences. One has to impute interpretative preferences to agents. According to Sen, economists usually perform this interpretative imputation unreflectively, falsely assuming that the preferences they impute are entirely determined by what they observe. Sen denounces this illusion. But he does not explain how a more reflexive economist should proceed. He claims that, when studying a given behavior, economists implicitly make sense of this behavior. In other words, when an economist imputes preferences and stipulates axioms, what s/he truly does is making sense of what s/he observes. But because Sen does not buttress his argument on a clearly spelled out theory of meaning, this conclusion is bound to remain elusive.

47 The notion that utility is economic meaning provides the rationale needed to escape the dead-end of Sen’s reasoning. In this approach, preference imputation is always a matter of accounting for actions in terms that can be endorsed by the actor (s) performing these actions. This step provides the bridge between empiricism and its theoretical rebuttal by taking utility in its dual nature. On the one hand, a utility-based model is a representation of a behavior that strives to accurately depict it. On the other hand, it does it by proposing a rationalization, embodied in the stipulation of the axioms, which is such that the agent (s) performing the depicted behavior can use it as a normative framework to account for his/their past and current behavior, and to rationalize his/their future behavior. This approach thereby articulates normative and descriptive dimensions in its understanding of utility. And in turn it thereby accounts for the permissibility or impermissibility of stipulating the axioms by setting limits to this stipulation: stipulating the axioms is permissible so long as it allows formulating both a correct representation of behavior and a normative framework that the agent (s) performing the behavior at issue can endorse as relevant a rationalization. Interestingly, the reasoning that we just unfolded as a complement to Sen’s argument could have been used to complement in the same way Ross [2005]’s theory of the economic agent, despite the deep differences between the two approaches (Hédoin [2016]). This illustrates that the feature of utility theory that we are interested in, and that we pinpoint by saying that utility is economic meaning, is a very fundamental one, placing itself upstream this debate between Sen’s critic of revealed preference theory and the neo-samuelsonian reply. In their application to concrete situations, both Sen’s psychological and Ross’s externalist understandings of the concept of preference can face the tribunal of endorsability, and therefore qualify or not as providing a basis for economic meaning.

48 Reformulated in more general terms, the argument that we just unfolded states that making sense of a behavior on the one hand, and judiciously stipulating axioms and imputing preferences on the other hand, are one and the same thing. Making sense of an action is recognizing that, for a given set of preferences to be imputed, the axioms of the theory of choice can be stipulated. What we call in informal discourse “making sense of an action”, is nothing more or less than accounting for this action by imputing preferences, chosen judiciously enough to render the stipulation of the axioms of the theory of choice endorsable by the agent concerned.

49 For the purpose of justifying the stipulation of these axioms, a major strength of the utility as economic meaning theory is that it lives up to the two requirements articulated above. We have seen that: (i) a satisfactory explanation would elucidate why the axioms have counterexamples. The explanation is straightforward in this approach. Some choices are simply not meaningful. The set over which they are made does not belong to a meaningful pattern of alternatives. They do not make sense for the agent her/himself, for example because the actions in question were unusual for her/him and s/he did not acted in a thoughtful way. Consequently, they do not make sense for the observer, who is at a loss trying to reconstruct a unitary meaning in what s/he observes. This explanation of (i) ultimately depends on the fact that, in this approach, the axioms are valid not universally but rather only on the restricted domain of the patterns of actions that are considered meaningful. This approach also explains that (ii) when a pattern of actions is meaningful, the content of the axioms appears to be minimal and commonsensical. This is because making sense of an action is not only what the observer does when s/he imputes preferences, but also what every agent does in everyday life to make sense of his or her own and others’ actions.

50 Let us now examine how this solution to the problem of the justifiability of the stipulation of the axioms of the theory of choice relates to the solutions provided by isolationism and factionalism. Isolationism holds that economic models strive to isolate causal factors in order to study their specific working. By “isolate [ing] a very small set of features from a very large set of features” (Mäki [1992], p. 321) they can appear unrealistic, but this is pivotal to their capacity to capture causal mechanisms and, in fine, to gain realist relevance. Mäki [2005] at times defended this view by analogizing it with experimental practices, which likewise manipulate elements to construct controlled environments. This defense however faces difficulties because experimentation arguably owes part of its scientific relevance to the way it handles constraints associated with the fact that causal factors are often not separable in empirical reality, whereas modelling simply assumes the possibility to isolate causal factors (Knuuttila [2009]). Symmetrically, modelling brings in mathematical constraints suggesting that the properties of models and the very assumptions they make might be first and foremost reflections of requirements of mathematical tractability rather than independently motivated isolations of causal factors (Knuuttila [2009]). Partly as a response to such weaknesses, the fictionalist view denies that economic models are elaborated, as isolationism claims, by abstracting elements from empirical reality to isolate individual causal factors. Using the example of Schelling’s [1978] checkerboard model of segregated neighborhoods, Sugden [2000] argues that understanding the genesis of this model as an abstraction process is artificial and incredible. In this view, a model is simpler than reality but is not in any sense the result of a simplification of reality. It is rather the product of a construction of an alternative, fictional, simpler system. It obviously raises the question of how such alternative systems can be informative or have explanatory bearings to study empirical reality. According to Sugden [2000], models seen through these lenses can produce knowledge about the real world through inductive inferences: they can be seen as instances of categories having as other instances systems existing in the real world. Sugden then anchors the possibility to identify the relevant classes and fictional and real concretizations thereof in the “credibility” of models. Sugden however does not spell out more precisely what he means by “credibility”. Our framework of utility as economic meaning usefully supplements his approach by claiming that a credible model is one that is endorsable by the agents whose behavior it models. To come back to Schelling’s seminal checkerboard example, this notion means that, if presented and explained the model, the agents would agree that this model cogently identifies and represent their “micromotives”. We do not mean to claim that endorsability is all there is to credibility, but only that endorsability is a source (possibly among others) of credibility. We therefore claim that our theory provides a specification of the fictionalism framework fleshing it out and thereby strengthening it in its application to the elaboration of representations of behavior through utility functions.

51 To sum up, in this section we have argued that, when a given utility modelling fulfils the requirements to qualify as a version of economic meaning, then in that case the stipulation of the axioms of the theory of choice is justified.

5. Practical implications

52 In this section we turn to more practical implications, pertaining to empirical economics.

53 The first implication of the proposed approach is that preference imputation and the ensuing economic modelling are valid only when it is reasonable to conjecture that the actions observed have a certain unitary meaning – that is, according to our general definition, when one can justifiably postulate that the utility-based account can be articulated in such a way as to be endorsable by the concerned actors. On this account, economic methods are not applicable in all kind of situations and to all kinds of objects simply by virtue of their formal generality.

54 Let us illustrate this point by looking at the example of hedonic valuations of environmental attributes. Hedonic valuations are methods used to deduce the demand for environmental quality from the demand for concrete marketed goods of various environmental qualities (Haab & McConnell [2002]). The current literature admits that hedonic valuations are sensitive to information (Kask & Maani [1992]). This reflects the notion that, being utility-based, hedonic valuations are based on the assumption that they model rational actions, and the behavior that it is rational for an actor to perform depends on the information he has. A consistent application of hedonic valuation would therefore include, as a preliminary first step, a verification that the actors studied have a reflexive awareness of their relation to the environmental attribute to be valued. By contrast, simply applying a utility model without starting by this preliminary verification would imply affirming that utility axioms can be postulated all the time, which in turn would be implicitly admitting that valuations based on uninformed, unreflective preferences are just as valid as valuations based on informed, reflective preferences. This sounds absurd, and most economists would reject this implicit admission; but as a matter of fact, the preliminary verification referred to above is rarely performed in practice. Our approach provides a simple means to perform this verification: narrate to the actors the content of the axioms of the utility model that you plan to use to model their behavior, if they endorse this account, go ahead and implement the utility model; if they don’t, the utility-based approach is not applicable.

55 A related implication is that the sets of actions that can be accounted for in a utility approach are not fixed a priori, once, and for all. These sets evolve as agents get used to the actions they perform, develop a reflexive analysis of their own actions, acquire more information on their nature, consequences and implications, and enrich their perceptions and conceptualizations with cognitive, emotive, and symbolic resources. In that sense, the meaning of an action is dynamic. Similarly, as emphasized by Ricœur [1975], the meaning of ordinary speech acts is often ambiguous and dynamic, which is not at all detrimental to ordinary communication.

56 Let us now take the case of shrimp farming as a concrete illustration of the possible implications of this dynamic approach for hedonic valuations. Shrimps have many attributes potentially determining behavior on markets: taste, color, harmony with other goods in local markets in coastal areas, etc. Shrimp farming also has major detrimental environmental effects (Martinez-Alier [2003]). This is yet another objective attribute of shrimps that one might be tempted to study with a hedonic approach. But it might be that consumers are informed on the detrimental consequences of shrimp farming and are strongly against such practices, but still consider that their behavior on markets is not the proper way to express these preferences. Those consumers could, for example, prefer to express this stance by being active in non-governmental organizations to protect natural habitats impacted by shrimp farming. The standard approach to such a situation is to maintain that, whatever agents might say or think, market behavior does reveal their (potentially partly unconscious) demand for environmental degradation. Our approach rather suggests that if we are to identify what is the demand for this attribute, a crucial interpretative step is required to identify which pattern of behavior of these agents (on the shrimp market or in their relations with environmentalist organizations or both) is relevant to capture the demand for this attribute. In the terms of the framework developed here, this in turn means identifying the formulation of the utility axioms that the agents can endorse.

57 In the wake of the above argument, our approach paves the way for a subtler analysis of behavior, preferences, and demand, by generalizing the idea introduced above. The fact that a given attribute exists in a given object or set of objects is not sufficient for the demand for this attribute to be readable off behavior with regard to this object or set of objects. The reason is that among the various elements that virtually can partake in the meaning of an act, only a subset actually does. The hermeneutic literature calls “thematization” the function selecting actual elements of meaning from the larger set of possible elements (Habermas [1984]). Scanlon’s [1998] notion of the “structure of reasons” captures the same idea. Thematization is a ubiquitous feature of cognitive processes, not a pathological feature of erroneous or incomplete ones (a point forcefully made by Habermas and Scanlon). The performance of certain thinking experiments can even be shown to unavoidably require a thematization. For example, Meinard and Grill [2011] argue that thinking about abstract goods like justice or biodiversity involves excluding some considerations as being irrelevant, which is by definition a thematization.

58 Thus, the following picture emerges. A given speech act can have several meanings depending on who utters it, who hears it, in which circumstances, what are the relevant symbols and the cognitive, emotive and symbolic resources that speakers, hearers, and spectators associate with it. Similarly, a given action can have several meanings depending on wide and heterogeneous series of contextual elements that cannot be identified a priori. When an economist studies a given pattern of behavior, what s/he has to do first is, therefore, to perform the interpretative task of hypothesizing an identification of the relevant meaning of the behavior at hand on what s/he observes.

59 This is certainly what most economists already do and what we are doing here is, in that sense, just rationalizing current practices by articulating the underlying logic that we are all more or less consistently already following. In the shrimp example, this interpretative exercise must lead to admitting that the observation of behavior on the market is not necessarily relevant to study demand for the environmental effects of shrimp farming. The agents do not thematize the relevant elements in their market behavior. Whereas current hedonic models postulate that the agents’ attitude toward environmental degradation are invariably present in the determination of their behavior and at most can be outweighed by attitudes toward other aspects, in our approach it is possible that the corresponding elements are not the-matized at all, and that preferences for environmental quality therefore do not partake in the determination of behavior on the market.

60 An additional strength of our approach is that it conceives of the dynamics of meaning as a dialectical interplay involving speakers’ intentions, hearers’ reactions, and observers’ reports (Ricœur [1983]). The notion of thematization introduced above is useful to flesh out this dialectical notion of meaning. Indeed, the cognitive basis of this dialectics is that hearers’ reactions and observers’ reports can lead speakers to modify their current thematizations, thereby leading them to adjust their acts and/or the meaning that they assign to their own utterances or acts.

61 This notion has sweeping implications for the practice of economic modelling. Indeed, in this approach, an economic model and its results are nothing else than elaborate observers’ reports. If this information is diffused and can be understood by economic agents, then they can use it to adjust their acts and/or the meaning they assign to it. To come back to our shrimp farming example, this means that the very act of elaborating an economic model of preferences for shrimps and publishing it can lead to a modification of the elements that consumers thematize when acting on markets. This in turn modifies the elements that should be integrated in the utility models. Notice, however, that the precise nature of such modifications is difficult if not impossible to foresee. It might be that consumers will become aware of the environmental consequences of buying farmed shrimps and therefore come to take this attribute of shrimps into account in the determination of their behavior on markets. But it might just as well be that it will motivate them to act against shrimp farming practices at a political level while still denying that their behavior on markets has any link with their environmental preferences.

62 The dialectical notion of the meaning of an act thereby strengthens, and substantially complicates, the mechanisms through which economic modelling can have performative implications – that is, can modify the very object that it aims to model (Callon [2007]).

6. Concluding remarks

63 In this article, we have articulated a new approach to utility theory. In this approach, using utility is justifiable so long as a preliminary interpretative analysis supports the idea that it is possible to make sense of the pattern of actions to be modelled. In emphasizing the importance of meaning and interpretation, this approach points toward a re-conceptualization of our discipline that some economist might be reluctant to endorse. Formalization and mathematization, although certainly indispensable, are not enough to secure the relevance of the economic framework (Grill [2015], p. 218-262). As economists we cannot ignore the intrinsically interpretative and meaningful nature of our object (the behavior of human agents), even if this comes at the price of admitting the tentative and interpretative nature of our results. This idea is hardly new and has instead been around at least since Dilthey. In this article, our aim was to buttress on a firm epistemological basis.

64 To conclude, let us emphasize that our approach opens an interesting field of inquiry related to a prominent methodological and ethical debate in the field of normative economics, launched by Sen [1979], on the concept of “welfarism”. Sen famously defined the latter as the doctrine stating that “the goodness of a state of affairs depends ultimately on the set of individual utilities in that state, and – more demandingly – can be seen as an increasing function of that set” (p. 446). Fleurbaey [2003] subsequently felt a need to distinguish “real welfarism”, where utility function “measures subjective utility or satisfaction” (p. 375), from “formal welfarism”, which satisfies the same axioms as real welfarism, but where utility function “may measure any objective or subjective notion of well-being and it is provided by moral philosophy or the social decision-maker” (p. 375). The theory of utility as economic meaning introduced in this article illustrates that, despite the clarification performed by Fleurbaey, the very meaning of utility is much more problematic than is usually admitted. Indeed, according to our reasoning, within what Fleurbaey calls “real welfarism”, the “subjective utility” that he refers to can be understood either in an analytical or in a substantial sense. Still according to our reasoning, depending on the endorsability of the utility-based representation at issue, it can be encompassed and qualify or not as “economic meaning”. Our approach therefore unveils that, upstream the distinction that Fleurbaey takes to be the most important one, other distinctions, which we claim are even more important, need to be clarified.