An Empiricist Critique of Constructive Empiricism: The Aim of Science
Abstract and Keywords
This chapter examines the aim of science according to constructive empiricism. It argues that empirical adequacy is not the only scientific non-instrumental theoretical value. A tautologous theory is empirically adequate, and yet such a theory would not be valued in science. It further argues that the scientific evaluation of theories is context-dependent.
Bas van Fraassen elaborates an ‘empiricist’ framework in which to conduct philosophy of science. Within it, he articulates an account of science he calls ‘constructive’. The core of constructive empiricism is the thesis that science aims to give us theories that are empirically adequate.1 Although I shall argue that this thesis, which I call ‘AimCE’, is ambiguous, and that the empiricist should reject it under the most important interpretations, my critique is far from destructive. I am sympathetic to empiricism, and the alternative account I develop on behalf of the empiricist concedes nothing to scientific realism's contention that science aims to give us theories that are true.
I begin by observing that van Fraassen takes AimCE to be an explicit claim about the main constitutive criterion by which scientific activity is evaluated, and an implicit claim about the main constitutive criterion by which theories are evaluated scientifically. Then I draw explicitly distinctions (p.84) he does not. Weak/strong versions of these claims employ a weak/strong notion of empirical adequacy. Immodest versions hold that the main constitutive criterion is the only constitutive criterion; modest versions admit additional constitutive criteria. Van Fraassen appears to reject weak immodest AimCE (a species of formalism) and to embrace either strong immodest AimCE or the species of weak modest AimCE holding that the only additional constitutive criterion is the provision of theories that are empirically informative (a species of substantivalism). But neither alternative sits comfortably with everything he says (Section 1.1).
Further unclarity in AimCE stems from ambiguities in the notions of ‘scientific activity’ and ‘criterion of success’. Narrow/broad versions of it construe scientific activity narrowly/broadly, while versions that are non‐comparative/comparative and absolute/context‐sensitive involve, respectively, notions of success that are non‐comparative/comparative and absolute/context‐sensitive. How these ambiguities are resolved determines where AimCE is best located among different strands of the literature concerning the aim of science. Non‐comparative, absolute, broad AimCE gels with the strand concerned with the nature of a theory whose provision would complete the entire scientific enterprise; non‐comparative, context‐sensitive, narrow AimCE better fits the strand focused on the constitutive goal of individual scientists or small research groups. But both interpretations are awkward, an interpretation on which AimCE is comparative being encouraged by the unnaturalness of construing the criterion of empirical informativeness non‐comparatively. Van Fraassen shows no sympathy for a context‐sensitive notion of empirical adequacy, but he does not exclude versions of AimCE according to which a context‐sensitive notion of empirical informativeness serves as a constitutive criterion of success. It is difficult to judge which disambiguations pertaining to these distinctions best capture the version of AimCE uppermost in his mind (Section 1.2).
I begin the second part of the chapter by arguing that both absolute and context‐sensitive criteria of success are constitutive of science. Those that are most fundamental are the absolute criteria that determine a comparative success relation over theories viewed sub specie aeternitatis, but a context‐sensitive criterion for the non‐comparative success of scientific activity in the narrow sense is also constitutive. Respectively, versions of AimCE that are absolute and comparative, and non‐comparative and context‐sensitive, offer accounts of these criteria (Section 2.1a). Defining empiricism (Section 2.1b), I assess them in (p.85) turn from the empiricist's perspective. I first argue that absolute comparative AimCE is too austere—the empiricist should include simplicity, explanatory power, internal coherence, and unification among constitutive theoretical values (Section 2.2a)—and that both formalist and pragmatist versions of it should be rejected: science values all empirical information (Section 2.2b). I then argue that even when context‐sensitive non‐comparative AimCE is modified so as to accommodate the additional constitutive theoretical values empiricism should recognize, the condition the most natural version of it claims to be necessary and sufficient for scientific activity to be successful as such fails to respect the competitive and communal nature of science. Though it serves as the agent of scientific activity's goal, the alternative condition I propose is heavily externalist (Section 2.3).
In several respects, van Fraassen clarifies the thesis that science aims to give us theories that are (merely) empirically adequate (rather than true). He explains that ‘science’ refers to ‘scientific activity’, that is, ‘[t]he activity of constructing, testing, and refining scientific theories—that is, the production of theories to be accepted within the scientific community and offered to the public’, while the ground of the presumption that science in this sense has an aim is said to be that ‘[i]t is part of the straightforward description of any activity, communal or individual, large‐scale or small, to describe the end that is pursued as one of its defining conditions.’ Since ‘[i]n the most general terms, the end pursued is success,’2 a claim about the aim of science amounts to a claim about ‘what counts as success in the [scientific] enterprise as such.’3 Clearly, an aim or criterion of this kind is constitutive, and not merely epistemic or instrumental: roughly, it is neither mere evidence for, nor a mere means to, success; it constitutes success.4 The thesis does not imply that the aim it (p.86) specifies, and hence the criterion of success for scientific activity it affords, is unique: van Fraassen (1980: 8) does ‘not deny that there are other subsidiary aims which may or may not be [mere] means . . . [For] everyone will readily agree that simplicity, informativeness, predictive power, explanation are (also) virtues.’ The thesis is concerned, therefore, to identify the main constitutive aim of scientific activity. Officially, it is agnostic as to whether subsidiary aims are mere means to achieving this or some other constitutive aim.
Once these explanations are made explicit, constructive empiricism's core thesis becomes:
Van Fraassen takes AimCE to imply that the main constitutive criterion of success for theories themselves is empirical adequacy: for he takes the activity of constructing, testing, and refining scientific theories to be successful as such iff it yields theories that are successful by scientific criteria.5 I shall call criteria of success for theories ‘theoretical values’. So he takes AimCE to imply that empirical adequacy is the main constitutive theoretical value.
(AimCE) The main constitutive aim of, and hence the main constitutive criterion of success for, the scientific activity of constructing, testing, and refining scientific theories, is that this activity gives us theories that are (merely) empirically adequate.
I agree that the evaluation of scientific activity in van Fraassen's sense by constitutive criteria is exactly parallel to the evaluation, by constitutive criteria, of the theory or theories the activity provides. When further clarifying, and evaluating, AimCE, I shall also focus on its implicit claim about constitutive theoretical values.
AimCE remains no clearer than the notion of empirical adequacy, but van Fraassen (1989: 228) defines a theory to be empirically adequate iff the empirical (p.87) structure that consists of all and only the observable parts of the world is embeddable ‘in some single model of the world allowed by the theory’.6 Empirical adequacy in this sense is so weak a property, however, that even tautologies possess it: since a tautology ‘allows’ all models, if there is a model into which all the parts of the world that are both actual and observable can be embedded, a tautology allows it. So let empirical adequacy in this sense, together with the versions of AimCE that employ it, be ‘weak’. Furthermore, let AimCE be ‘immodest’ when combined with the view that science has no constitutive aim other than the main one, and ‘modest’ when it is combined with the view that science has additional constitutive aims.
In maintaining that weak empirical adequacy is the only constitutive theoretical value, immodest weak AimCE denies, in particular, that the evaluation of a theory by constitutive criteria is sensitive to the theory's logical strength or ‘informativeness’.7 It is a species of what I call the ‘formalist’ view that science itself places no value on information: science merely constrains and manages our pursuit of information about matters that interest us.
Although van Fraassen (1980: 8) states that AimCE is agnostic as to whether science has a constitutive aim other than the main one, and the weak version is generated by his own definition of empirical adequacy, on other occasions he is less guarded about formalism. Taking a broader perspective, I find considerable attraction in an interpretation on which he is a formalist who embraces immodest weak AimCE. But this interpretation doesn't fit everything he says. In particular, later in The Scientific Image he contrasts empirical informativeness with what he terms ‘pragmatic’ virtues like simplicity and explanatory power, and he speaks of ‘the’ aim of science as being to give us theories that are ‘empirically adequate and strong’.8 This latter remark, especially, puts anti‐formalist—or, as I prefer, ‘substantivalist’—versions of AimCE firmly on the exegetical agenda. More specifically, where information that concerns only (p.88) observables is ‘empirical’, it invokes an ‘empirical’ substantivalism according to which science values information, but none that is not empirical.
The relation van Fraassen's (1980) advocacy of AimCE bears to his apparent empirical substantivalism is unclear. Is his intention that the former should incorporate the latter? On the one hand, it would seem not: since weak AimCE is compatible with formalism and reflects his official definition of empirical adequacy, it alone is not committed to substantivalism. On the other hand, it would seem so: for he seems inclined to employ a notion of empirical adequacy relevantly stronger than weak empirical adequacy, and to hint that it is in terms of this notion that AimCE should be understood. For example, he is among the authors of the following statement:
Here, he says explicitly that he has not expressed an opinion as to whether ‘empirical adequacy is within the reach of science,’ and he implies both that for all he knows it isn't, and that constructive empiricism is not committed to the claim that it is. This much is absurd, however, unless either a stronger notion of empirical adequacy is employed, or constraints are tacitly invoked that make the scientific pursuit of weak empirical adequacy more difficult than it would otherwise be.
Van Fraassen articulates part of his controversy with the scientific realist in terms of the aim of science, saying ([in The Scientific Image] p. 12) that it is ‘to give us theories which are empirically adequate’, whereas for the scientific realist it is to ‘give us . . . a literally true story of what the world is like’. On first glance this may seem to suggest that van Fraassen thinks empirical adequacy to be a reachable aim for science. But of course that is not implied at all. In fact, he nowhere says that empirical adequacy is within the reach of science—nor that it is not. It is simply an issue van Fraassen does not address and need not address in order to make his point against the realist. Perhaps the most unambiguous way to state this point is thus: even if empirical adequacy should be an attainable goal for science, this does not mean that truth is attainable as well.9
One can see how the latter interpretation would go. What would otherwise be a trivially easy pursuit of weak empirical adequacy is hampered by a need for empirical information generated by our own interests (on the formalist view) or by science itself (on the substantivalist view). But the following consideration favours the former interpretation. Whereas the notion of empirical adequacy that occurs in AimCE is twin to the notion of truth that occurs in the scientific realist's account of the main aim of science, nuances in van Fraassen's (p.89) formulations hint that the main aim scientific realism specifies is stronger than any aim achieved by the provision of mere tautologies. On a natural reading, the aim of giving us, ‘in its theories, a literally true story of what the world is like’ is stronger than the aim of giving us theories that are literally true. The proposition ‘something is spherical or nothing is spherical’ is literally true. But does it give us a ‘literally true story of what the world is like’? Surely not: it does not give us any story of what the world is like. A theory that gives us a ‘literally true story of what the world is like’ must be not only true, but informative. That it should not conflict with the world does not suffice; it must capture the world (at least in certain important respects). These hints approach bald assertion when van Fraassen (1980: 3) writes:
Here the switch to a stronger aim seems undeniable: the aim according to scientific realism is not provision of theories that are true per se, but provision of theories that state truths about certain weighty matters; the aim according to constructive empiricism is not provision of theories that are weakly empirically adequate per se, but provision of theories that state certain truths about what is observable. Only theories that are informative can achieve either aim.
It may at first seem trivial to assert that science aims to find true theories. But coupled with the . . . [view that theories are to be interpreted literally] the triviality disappears. Together they imply that science aims to find a true description of unobservable processes that explain the observable ones . . . Empiricism has always been a main philosophical guide in the study of nature. But empiricism requires only to give a true account of what is observable.10
The tension between van Fraassen's official definition of empirical adequacy and his unofficial remarks invites consideration of ‘strong’ versions of AimCE couched in terms of a strong notion of empirical adequacy that somehow incorporates empirical informativeness. Of various alternatives, those most akin to the definition of weak empirical adequacy, and most natural, have the form:
I suspect a notion of strong empirical adequacy having the form (F), or at any rate something similar, is at issue when van Fraassen and his fellow authors (p.90) distance constructive empiricism from the claim that empirical adequacy is an achievable aim.
(F) ‘x is strongly empirically adequate’ =df ‘x is weakly empirically adequate and x expresses such and such empirical information.’
Two formulations reconcile AimCE with the view that weak empirical adequacy and empirical informativeness are the only constitutive theoretical values. One is the species of modest weak AimCE according to which science has exactly two constitutive aims—the main aim of providing theories that are weakly empirically adequate, and the subsidiary aim of providing theories that are empirically informative. The other is an immodest version of strong AimCE according to which science has but one constitutive aim—the provision of theories that are strongly empirically adequate. Neither alternative fits perfectly everything van Fraassen (1980; 1997) says.
On their most natural interpretations, the difference between these versions of AimCE is more than verbal. Each notion of strong empirical adequacy having the form (F) is non‐comparative. Since the notion of success for which a non‐comparative property is most naturally taken to be a criterion is itself non‐comparative, strong AimCE is most naturally interpreted as claiming that possession of a certain non‐comparative property is the main constitutive criterion for the application of the non‐comparative predicate ‘is successful’. In contrast, on the most natural interpretation, weak modest Aim specifies weak empirical adequacy and empirical informativeness as the key ingredients in constitutive criteria for the application of the comparative predicate ‘is more successful than’. For whereas van Fraassen's talk of empirical informativeness is most naturally construed as talk of the comparative relation ‘is more empirically informative than’,11 it is incoherent to offer a comparative relation as a criterion for a non‐comparative property. If, for example, tallness is offered as the sole subsidiary criterion of attractiveness, and attractiveness is taken to (p.91) be a non‐comparative property that partitions the class of people into two, then tallness cannot be (a nominalization of) the relation expressed by ‘x is taller than y’; on pain of incoherence, it has to be the non‐comparative property expressed by the non‐comparative ‘x is tall’. A further advantage of a comparative interpretation is that it gives point to the distinction between main and subsidiary criteria. Since weak empirical adequacy and empirical informativeness can only to serve to determine a comparative success relation between theories upon being weighted, the point of calling the former the ‘main’ criterion would be to indicate that it counts more heavily in the weighting.12
Let a comparative/non‐comparative version of AimCE employ a comparative/non‐comparative notion of success. At least from a substantivalist point of view, sticking to van Fraassen's official definition of empirical adequacy practically forces an interpretation on which AimCE is not only weak but comparative: for empirical informativeness is most naturally construed as a comparative notion whose form prevents it from serving as a criterion for the application of a non‐comparative predicate. Conversely, adopting a strong notion of empirical adequacy in the light of his unofficial lead practically forces an interpretation on which AimCE is not only strong but non‐comparative: for his remark that he is agnostic as to whether empirical adequacy is achievable only makes sense if empirical adequacy is a non‐comparative property.
The questions as to whether AimCE is weak or strong, modest or immodest, comparative or non‐comparative, are relevant to the issue as to how it should be located in the literature on the aim of science. One strand of this literature focuses on the nature of a theory that would ‘complete’ science by making all (p.92) possible subsequent scientific activity redundant. For example, Newton‐Smith (1981: 30–1) writes:
Since van Fraassen must hold that provision of a theory that expresses all truths solely about observables would complete science in this sense—he denies there are constitutive theoretical values such a theory might fail to exhibit perfectly—and would be successful as such, one might wonder whether AimCE expresses a view about how science might be completed.
[I]magine achieving what the instrumentalist takes to be the goal of science. Suppose we have a black box into which we can feed an observational characterization of the state of any physical system at any moment of time and which correctly predicts the state of that system at any specified future moment and retrodicts the state at any past moment. If the instrumentalist were correct in his claim about the aim of science, this would represent the completion of the scientific enterprise.
A difficulty for this interpretation stems from the fact that AimCE is not really of the right form to serve as a(n implicit) specification of (even necessary features of) a completing theory. Like the formulations van Fraassen offers of the main aim according to scientific realism, AimCE employs a plural term: it says science aims to give us ‘theories’ that are empirically adequate. This use of the plural would be surprising were his concern Newton‐Smith's: for it is commonly thought that only one theory could complete science. Although it is possible that he intends to hedge his bets over whether it is a plurality of theories that would be completing, I think this unlikely. Admittedly, whether only a plurality of theories would complete science is a live issue if constitutive theoretical values include not only empirical informativeness but, for example, simplicity, and, furthermore, it is possible for there to be genuinely polyadic relations, that is, relations born to at least one plural subject. For in that case we could imagine there to be jointly completing theories T1 and T2 such that any attempt to merge into a single theory the empirical information they express jointly would be so much at the expense of simplicity that by constitutive criteria, for every theory T3, taken plurally T1 and T2 are more successful than T3. I don't see how this issue could be live for van Fraassen, however. If weak empirical adequacy and empirical informativeness are the only constitutive theoretical values, any theory carrying exactly the empirical information that T1 and T2 carry jointly would be at least as successful by constitutive criteria. So on his view it is a necessary condition of science only being completed by a plurality of theories that there be a set of theories such that no single theory (p.93) carries exactly the empirical information the theories in the set express jointly. But what could prevent the existence of a theory with this property?
A better exegesis must give a different role to van Fraassen's use of the plural. An alternative proposal takes it to signal a comparative reading of AimCE. Whereas on the first proposal, AimCE's use of ‘theories’ stems from the fact that the possibility is envisaged of a single instance of scientific activity having a plural outcome, on this proposal it stems from the fact that two instances of scientific activity, each of which has a single outcome, are being compared.13 But an ideal interpretation would accommodate the best features of both proposals. Like the first it would construe AimCE's account of the main aim of science as a specification of a constitutive criterion for the application of a non‐comparative predicate ‘is successful’. Like the second, it would retain the idea that each instance of scientific activity has a main aim that is singular (namely, provision of one theory that is empirically adequate). The only way to achieve a synthesis is to deny that AimCE presumes there is a single constitutive criterion C such that, for each instance S of scientific activity, S is successful as such iff S satisfies C. Of course, the question arises as to why van Fraassen would employ the singular phrase ‘scientific activity’ were his real subject plural in this way. But a ready answer is provided by the awkwardness of the phrase ‘instance of scientific activity’ and the like.
This third proposal works best when it exploits an ambiguity in the notion of ‘science’. The notion of science, and hence of scientific activity, can be understood widely or narrowly. At it widest, it refers primarily to what I take to be uppermost in Newton‐Smith's mind when he speaks of the ‘scientific enterprise’—namely, the lengthy historical process that occurs in the West and extends from the present day at least as far back as the seventeenth century, and, hopefully, far into the future. At its narrowest, it refers to small‐scale activities of individual scientists or research groups. This is not to say that only science in the narrowest sense has instances: there have been separate scientific traditions. But because philosophy of science accords the Western tradition such prominence, it is easy enough to think of science in the widest sense as unique. Moreover, even if the actuality, or at any rate possibility, of entirely (p.94) separate scientific traditions is admitted, it is also natural to think there is a single theory whose provision would complete each of them. Construing science as scientific activity in the widest sense facilitates the thought: one activity, one material aim.
In contrast, it is obvious that scientific activity in the narrowest sense has many instances, and hardly less so that, in practice, there is no one material criterion necessary and sufficient for their being non‐comparatively successful. The scientific activity that comprised Newton's dynamical researches could hardly be judged unsuccessful merely because the resulting theory is hopelessly inadequate in comparison with the theory (or theories) that would complete the entire scientific enterprise. On the contrary, that activity was exemplary. Nor does this judgement commit us to judging successful all scientific activity having the same outcome as Newton's. Although someone well‐versed in general relativity and quantum mechanics, but entirely ignorant of the history of physics, could rediscover Newton's theory, their so doing would be so far from being exemplary that one might justifiably deem it unsuccessful.
When the scientific activity constituted by Newton's dynamical researches is judged successful, the sufficient criterion employed is context‐sensitive. This context‐sensitivity finds no echo in the tradition to which Newton‐Smith's discussion of the aim of science belongs: a constitutive criterion whose satisfaction would render any further scientific activity redundant is not sensitive to context! But constitutive context‐sensitive criteria of success are invoked in another strand of the literature on the aim of science. I have in mind the strand in which one finds claims of the form ‘science aims to provide us with theories that have increasing F.’14 A claim of this form does not have to be taken to specify a constitutive context‐sensitive aim: it may be given a comparative reading on which it says that, independently of context, the (main) constitutive criterion of one instance of scientific activity being more successful than another is that the theory provided is more F. More (p.95) natural, however, is a non‐comparative reading on which it says that at each context, the (main) constitutive criterion for scientific activity at the context to be non‐comparatively successful is that the activity provides a theory that is more F than any rival theory that is relevant at the context (the presumption being that at different contexts different rival theories are relevant). Since AimCE makes no mention of theories that are ‘increasingly’ such and such, it does not fall explicitly into this strand of the literature. I do think it sits more comfortably here than in the first strand, however. Its talk of ‘theories’ in the plural would then be perfectly natural: if, constitutively, science has different (material) aims at different contexts, in practice only a multitude of theories will achieve them.
This exegesis also encounters an obstacle. If whether or not an instance of scientific activity is successful as such is sensitive to the context in which the activity occurs, at least one constitutive criterion of success must be similarly sensitive. But weak empirical adequacy, and notions of strong empirical adequacy of the form (F), are absolute. Nor has empirical informativeness been presented as being context‐sensitive.
Still, at the very least, context‐sensitive notions of empirical informativeness, and, even, of empirical adequacy, lie just beneath the surface. Consider van Fraassen's (1980: 12) alternative definition of empirical adequacy as ‘[saving] . . . all the phenomena’.15 He emphasizes that he intends the quantifier ‘all’ to be unrestricted: in particular, it is not to be restricted to phenomena that are sometime observed. But since the domains of ordinary language quantifiers are readily subject to tacit context‐sensitive restrictions—we say, for example ‘he told everyone to leave the room’—the linguistic context, and even the historical one,16 points in the direction of a context‐sensitive notion.
Let a restriction on the quantifier in the phrase ‘all phenomena’ that is induced at a context c determine the phenomena that are ‘relevant’ at c. Then a context‐sensitive notion of weak empirical adequacy is definable as follows: for each context c, T is weakly empirically adequate at c iff the structure that consists of all and only the phenomena relevant at c is embeddable in some single model of the world allowed by the theory. And notions of context‐sensitive strong empirical adequacy could receive definitions of the form: for each context c, T is strongly empirically adequate at c iff T is weakly (p.96) empirically adequate at c and T expresses such and such empirical information about the phenomena relevant at c.
Let context‐sensitive versions of AimCE specify at least one constitutive aim, and, hence, at least one constitutive theoretical value, that is context‐sensitive. At the very least, context‐sensitive versions of AimCE have some exegetical credibility. Although those using a context‐sensitive notion of empirical adequacy do violence to van Fraassen's texts, those according to which the only constitutive subsidiary criterion is a context‐sensitive notion of empirical informativeness do not.
AimCE is a claim about criteria of success for science (qua activity) and, implicitly, for theories (the products of scientific activity). It is not entirely clear about what the criteria are being offered for—is it comparative or non‐comparative success of science, and is it science in the narrow or broad sense?—or what these criteria are—are they weak or strong, modest or immodest, comparative or non‐comparative, absolute or context‐sensitive? As such, it is unclear about the question addressed and the answer given.
To clarify the philosophical issues, I shall first identify the two most important questions about constitutive criteria of success. Then, since I wish to develop an empiricist critique of the answers given by (the relevant versions of) AimCE, I shall say what I take empiricism to be.
(a) Of the issues my discussion of AimCE invokes, the most fundamental is broached by the absolute/context‐sensitive distinction. Absolutists maintain that there are no constitutive context‐sensitive criteria of success: they hold that constitutive evaluation transcends all contexts. At the opposite extreme, contextualists maintain that there are no constitutive absolute criteria of success: they hold that context pervades all constitutive evaluation. A hybrid view according to which there are both absolute, and context‐sensitive, constitutive criteria of success, lies between these extremes.
Since weak empirical adequacy and (literal) truth are absolute, on the most natural interpretation—indeed, officially—AimCE and the scientific realist's claim that science aims at the provision of true theories, are both (p.97) anti‐contextualist.17 So the debate van Fraassen describes as being at the heart of the philosophy of science presupposes that contextualism is false. This presupposition is reasonable, however: contextualism, which should not be confused with, and which receives no support from, radical historicism,18 is unappealing. Independently of what evaluative criteria appear constitutive, that science provides absolute criteria by which to evaluate its activities and products is plausible.
Once contextualism is rejected, the question ‘What are the fundamental absolute criteria constitutive of science?’ becomes pressing. The right answer is that they are comparative theoretical values. They are theoretical values because the constitutive absolute evaluation of scientific activity is derivative: constitutively, scientific activity is successful to the extent that the theory or theories it provides is successful. They are comparative because the non‐comparative absolute evaluation of a theory or theories is derivative: a theory is constitutively absolutely successful just in case constitutively, nothing is more successful than it.19
Van Fraassen's tendency to equate the constitutive criterion of success about which AimCE expresses an opinion with the end or telos of scientific activity gives some reason to think he would not agree with the second part of this answer.20 It follows by elimination, however. It cannot be that there is a constitutive absolute criterion for non‐comparative success, but none for comparative success: scientists often make judgements about theories of the form ‘x is more successful than y’ without invoking context. Nor can it be that the constitutive absolute comparative criteria are derivative: only a trivial comparative criterion is definable from a non‐comparative one.21 Finally, it is gratuitously speculative (p.98) to hold that both comparative and non‐comparative constitutive absolute criteria are alike fundamental; a purportedly fundamental absolute non‐comparative criterion is rendered superfluous by a fundamental absolute comparative one, while scientific practice can be no guide to its nature: in practice, scientists never judge a theory to be absolutely successful—that is, completing (of the entire scientific enterprise).
Consequently, van Fraassen's tendency to express his insight that there are criteria of success constitutive of science as the thought that scientific activity has a telos is, at best, potentially misleading. Science does not have a ‘naive’ telos—in effect, an absolute non‐comparative property whose pursuit gives point to scientific activity. A naive telos is constitutive of an activity only if the known metaphysical impossibility of realizing it would render further pursuit of the activity (other than as a means) irrational. But while a theory whose provision realized a naive telos of science would have to be maximally successful by the absolute comparative criteria that are constitutive, it is clear that an agent who came to know that it is metaphysically impossible to provide a theory that, constitutively, is absolutely maximally successful, would not be a logical inference away from discovering that further scientific activity would be irrational.22 Science is more like analytic philosophy than chess in this respect. The fundamental criteria constitutive of philosophical research are criteria of comparative success. In philosophizing, one tries one's best (by those criteria), but so doing is no more indicative of a naive telos than is the (non‐comparative) aim of writing the perfect paper: both are completing ends in name only.
Here then is one question to be asked—what are the constitutive absolute criteria for the comparative success of theories?—and the answer(s) to (p.99) it given by (absolute, comparative) AimCE assessed (Section 2.2). Whether there is another depends on whether there are any constitutive context‐sensitive criteria of success. It is unclear what van Fraassen believes regarding this issue. Since weak empirical adequacy is absolute, the matter turns on his attitude to empirical informativeness: for he definitely holds that no theoretical value other than these two is constitutive. If he espouses immodest weak AimCE, then he is an absolutist: for this species of formalism holds that the only constitutive theoretical value is weak empirical adequacy. If he embraces empirical substantivalism, the matter is less clear cut: the species holding that science values all empirical information is absolutist, whilst the pragmatist species according to which science values just the empirical information we value has both absolutist and non‐absolutist versions, depending on whether the notion ‘empirical information we value’ is context‐sensitive.23
Absolutism is surely false, however. The absolutist must hold that the criteria employed in the judgement that Newton's dynamical researches were successful are either not context‐sensitive, or not constitutive. But neither option is attractive: the former is at odds with the fact that a twenty‐first‐century child prodigy who came up with Newton's theories of motion and gravitation independently would not have engaged in similarly exemplary scientific activity, while the latter effects a massive abstraction that reduces science to the bare bones and emasculates the scientific agent by relegating his own perspective and cognitive goals to the periphery. The correct position in the debate between absolutists and contextualists is the hybrid compromise: both absolute and context‐sensitive criteria of success are constitutive.
Constitutive context‐sensitive criteria having been admitted, the question arises as to whether it is comparative or non‐comparative criteria that are basic. In this case, I think that it is the criteria of non‐comparative success that are basic. Taking them to be so permits a literal interpretation of the attractive idea that the practice of science is goal‐driven: an agent of scientific activity strives for a context‐sensitive goal whose realization is a constitutive necessary and sufficient condition of success. So here is a second question to be asked—what are the constitutive, context‐sensitive criteria employed (p.100) when we judge theories and the scientific activity that provided them non‐comparatively successful?—and the answer(s) to it given by (context‐sensitive, non‐comparative) AimCE assessed (Section 2.3).
(b) I shall argue that the answers given by the relevant versions of AimCE to the two questions I have singled out as being of most importance should be rejected by one who shares van Fraassen's empiricist perspective. It is especially important, therefore, to clarify what this perspective involves.
Whereas van Fraassen (1985: 286) identifies empiricism initially with a certain thesis (that empiricists believe), more recently he has argued that it is a stance (empiricists adopt).24 I shall remain neutral on this issue: as I employ the term, ‘empiricism’ is either a thesis or stance—I use the term ‘doctrine’ to cover both alternatives—regarding certain matters. But which? Again, his answer has evolved. Sometimes, and especially in his earlier writings, empiricism amounts to sympathy towards certain kinds of reasons for belief (or high credence), and hostility towards other kinds.25 On other occasions, it amounts to sympathy towards certain theoretical values, and hostility towards others (explanatory power in particular).26
The claim that for an empiricist there are no constitutive theoretical values other than empirical adequacy and (perhaps) empirical informativeness is threatened with triviality if empiricism is understood in the second way. But it is certainly substantial, and explicitly advocated by van Fraassen, when empiricism is understood in the first way, and this is the way I shall understand it. More specifically, I follow van Fraassen (1985: 286; 1989: 8) in taking empiricism to be the epistemological doctrine that:
Like him, I take (E) to express a negative doctrine about epistemic reasons of a certain kind. Initially, I take this to be:
(E) Experience is the sole legitimate source of information about the world.
(p.101) Since claims about the aim of science are not explicit epistemological theses, the impact of empiricism on the debate to which AimCE contributes is indirect. Indeed, although van Fraassen (1985: 286) claims ‘the empiricist critique of knowledge undercuts all grounds for scientific realism,’ even the route to a refutation of scientific realism's claim that science aims to give us theories that are literally true is perilous! Suppose with (E1) that there can be no epistemic reasons for giving high credence to a theory concerning unobservables. Why should this in itself prevent science from valuing truth above empirical adequacy, and, hence, from employing truth as a constitutive criterion of success for theories, and the truth of a theory as a criterion for the success of the scientific activity from which the theory results? It would prevent science from so doing were there a guarantee that for every constitutive non‐comparative criterion of success, in some circumstances, and with respect to some item, on occasion there are epistemic reasons for giving credence to the proposition that the item satisfies the criterion. But there is no such guarantee. In particular, the fact that scientific activity is rational activity is powerless to provide it. Practical rationality embodies no general requirement that in every case one's non‐comparative values must be such that one can have epistemic reason for giving high credence to the proposition that some optional act will realize them. The most that could be required is that rational activity only pursues non‐comparative values such that (it is reasonable to believe that) the activity increases the chance of their being realized. (E1) alone does not prevent an activity, whose sole constitutive aim is the provision of informative true theories concerning unobservables, from satisfying this minimal requirement.
(E1) With respect to any theory T such that T might be weakly empirically adequate and yet untrue, nothing—in particular, no degree of simplicity, or of explanatory power, or of internal coherence and so on—is an epistemic reason for believing T true.27
If empiricism is to aspire to undercut scientific realism with quite the finality van Fraassen claims, it must be identified with a doctrine stronger than (E1). One might try:
(p.102) Yet even empiricism in the sense of (E2) fails to ensure that an activity for which truth about unobservables is a constitutive theoretical value violates the canons of practical rationality just broached. Unlike (E1), (E2) does have the consequence that if an agent gives equal credence to two empirically equivalent theories T1 and T2, nothing can constitute an epistemic reason for his subsequently giving more credence to T1 than to T2. But even this doesn't entail that it is impossible for an agent to do something that gives him epistemic reason to raise his credence in the proposition that he believes something (contingently) true about unobservables: effecting a transition from the state of having no beliefs about unobservables to the state of having some beliefs about them will do that, for example. Moreover, there is no question of (E2) precluding the existence of agents who engage in an activity for which, constitutively, informative truth about unobservables is a constitutive theoretical value. Unless it is irrational to reject (E2), which of course van Fraassen (rightly) does not maintain, such agents might even be rational: brought up in a culture in which, for example, children are regularly instructed that there are epistemic reasons for believing certain theories that make strong claims about unobservables, they reject (E2) without a second thought and (rationally) dismiss the complaint that theoretical values constitutive of an activity they savour violate canons of practical rationality.
(E2) With respect to any two theories T1 and T2 that are empirically equivalent, nothing—in particular, not the fact that T1 is more simple, or more explanatory, or more internally coherent, or more unifying and so on than T2—constitutes an epistemic reason for giving more credence to T1 than one gives to T2.
Notwithstanding van Fraassen's bullish remarks about its efficacy, it follows that empiricism alone can have no objection to the contention that what certain folk are up to is schmience, an activity for which the sole constitutive criterion of success is the provision of true theories that are informative about both unobservables and observables alike. This point is a modest one, however: all it means is that the argument from empiricism to a denial of realist claims about constitutive criteria of success (and hence constitutive theoretical values) must proceed case by case. Empiricism must allow that it could turn out that constructive empiricists are right about the criteria of success constitutive of what we do (science), but wrong if they go on to reject realist claims about the superficially similar activity engaged in by certain rational extra‐terrestrials (who in fact pursue schmience). It isn't quite the case that the reverse could also be true: it couldn't happen that we don't do science! But empiricism must also allow a close relation: it must allow that it could turn out that the scientific realist is right about science (and hence us), but wrong if he goes on to assert that science is also the superficially similar (p.103) activity engaged in by certain extra‐terrestrials (for in fact these agents pursue the activity constructive empiricists mistakenly take science to be).
The first important question to be addressed by one who acknowledges that there are criteria of success constitutive of science was seen to be this: what are the absolute theoretical values to be utilized in a constitutive absolute comparative evaluation of theories? Comparative, absolute AimCE addresses this question. In effect, all versions of it hold that these values include weak empirical adequacy. The formalist version holds that there are no others; substantivalist versions admit just one other—empirical informativeness (in certain respects).
Let us say that empirical substantivalism is ‘global’ iff it holds that science values all empirical information, and that it is ‘local’ otherwise. I first argue that in one respect the global empirical substantivalist version of comparative absolute AimCE is too austere: there are constitutive theoretical values it does not recognize (part (a)). I then argue that in no respect is it insufficiently austere: global empirical substantivalism itself is correct, since the information science values is exactly the information that is empirical (part (b)). Both arguments assume empiricism.
(a) In van Fraassen's writings, AimCE's main opponent is scientific realism, and the main ground on which the battle is fought is the question why, as ‘everyone will admit,’ certain additional features of theories—such as ‘simplicity, . . . predictive power, [and] explanation’—are also ‘virtues’.28 Anti‐empiricism permits a scientific realist explanation of this fact: (at least some of) these features are epistemic reasons for higher credence in the truth of theories that better exhibit them, and truth is a constitutive theoretical value. But although empiricism precludes an explanation of this kind, it is not the case that comparative absolute AimCE emerges as soon as the scientific realist's explanation is rejected: of the alternative explanations empiricism permits, only some support it.
Three of the alternatives that do support it take such virtues to be constitutive means to constitutive criteria of success other than truth, mere (p.104) epistemic reasons for constitutive criteria of success other than truth, and non‐constitutive criteria of success. But while Van Fraassen (1980: 87–96) advocates (at least) one of these alternatives,29 a fourth is better. This is the simple alternative of taking such features to be constitutive theoretical values. Empiricists may adopt it, and there are no disadvantages, and some advantages, to so doing. To show this, I shall focus on the case of explanatory power first.
Since the empiricist merely denies that there is an epistemic guide to the truth about unobservables, he may take explanatory power to be a constitutive theoretical value. In particular, so doing would not undermine his contention that there is no epistemic guide to truth about unobservables: for it is not the case that belief that T1 has greater explanatory power than T2 requires that T1 be given higher credence of being true; whether a theory explains some phenomenon, and if so how well it explains it, are matters independent of whether the theory is true.30
Van Fraassen argues that two considerations should persuade the empiricist not to take this option, however. His first argument concludes that explanatory power is not a constitutive theoretical value because physicists do not insist that ‘an explanation of the [implied correlations between distant particles] must be found which fits in with quantum theory and does not affect its empirical content at all . . . [indeed] scientists [often refuse] to enlarge their theories in ways that do not yield different (or further) empirical consequences.’ But although the historical point is surely correct,31 the inference is a non sequitur. (p.105) The supposition that explanatory power is a constitutive theoretical value permits a perfectly good explanation of why scientists behave in this way. Because the constitutive value of explanatory power is small in comparison with the value scientists place on empirical informativeness, devoting precious cognitive resources to developing (and even learning) more explanatory but empirically equivalent theories would be inefficient: in practice, such resources are better devoted to the pursuit of greater empirical informativeness within the confines of the constitutive criterion of weak empirical adequacy.32
This explanation has a significant advantage over the explanations available to comparative absolute AimCE: it respects intuitive properties a theory would have to have in order to complete the scientific enterprise (in its entirety). If there are no constitutive theoretical values other than weak empirical adequacy and empirical informativeness, then any weakly empirically adequate theory that states all truths solely about observables would be completing: no theory or theories could be more successful, constitutively, than it. Yet it is wrong to suppose that the discovery of a theory with these features would automatically render further scientific activity redundant. If the theory did extremely badly by the criterion of explanatory power, the search would still be on for a single theory doing better by this criterion.33 Since there is no question, in such a case, of explanatory power serving as a constitutive means in pursuit of weak empirical adequacy or greater empirical informativeness, one could only deny that explanatory power is a constitutive theoretical value by denying that further pursuit of it would be scientific.
This is the attitude van Fraassen (1980: 95) displays when he speaks of (possible) theories that are empirically equivalent to quantum theory, but which unlike quantum theory explain the quantum mechanical correlations, (p.106) as ‘metaphysical extensions of [quantum theory that are] . . . philosophical playthings only.’ It is unattractive. In practice, pursuit of a theory that is empirically equivalent to one already in our possession, but more explanatory, is no doubt not the best science. But the idea that it is not science is far too stringent. In the case envisaged, at least some theorists who helped produce a theory that is weakly empirically adequate, and maximally empirically informative, could be expected to take up the challenge to produce a more explanatory theory. Moreover, they would proceed much as before. It is unmotivated and arbitrary to insist that while their earlier pursuit of explanatory power would have been science their subsequent pursuit of it would not be.
In effect, van Fraassen's (1985: 287) second argument tries to counter this charge. It holds that explanatory power cannot be a constitutive theoretical value because it is not objective:
This argument fails in two respects. Firstly, while van Fraassen's (1980: chapter 5) discussion of the pragmatics of explanation illuminates pragmatic aspects of the way in which language functions so as to express explanations (and the why‐questions they answer), it falls far short of showing that (scientific) explanation itself is a non‐objective, pragmatic notion.34 Secondly, his presumption that all constitutive theoretical values are objective is untenable. As Laudan (1984) emphasizes, dissensus among the scientific community is no less in need of explanation than is consensus amongst it, and supposing that some constitutive theoretical values are not objective yields at least a partial explanation. Since dissensus makes a vital contribution to scientific progress,35 it is surely naive to insist that its explanation should invoke none but factors that are extra‐scientific. Accordingly, one may not presume that unless explanatory power is objective it cannot serve as a constitutive theoretical value. Since some element of non‐objectivity in scientific theory‐evaluation is inevitable,36 whether explanatory power is fully objective has no bearing on the issue. (p.107) Mutatis mutandis, this defence of the view that (comparative) explanatory power is a constitutive (comparative) theoretical value is appropriate to the other features van Fraassen classifies as ‘subsidiary’—simplicity, predictive utility, and unification—and to a fourth feature, internal coherence.37 These are all features scientists deem virtuous. Although empiricism denies they ever yield epistemic reasons for believing a theory about unobservables, it is not obliged to deny they are constitutive theoretical values: the extent to which a weakly empirically adequate theory possesses any of them does not depend on whether the theory has the additional property of being true. Moreover, in so far as they are independent, they provide compelling criteria in addition to explanatory power for the evaluation of weakly empirically adequate theories that express all truths solely about observables, and it would be arbitrary to deny that their so doing is constitutive.
[O]n my view of explanation, what is the best explanatory account for one community of scientists may not be that of another . . . So [explanatory power] does not seem to describe, either on my view or on that of various realists, a worthy aim for science.
In one respect, then, the species of weak, absolute, comparative AimCE according to which (closeness to) weak empirical adequacy38 and empirical informativeness are the only absolute constitutive criteria of comparative success for theories is too austere: even the empiricist should take these criteria to include (comparative) simplicity, explanatory power, predictive utility, and internal coherence. The further question arises, however, as to whether in another respect it is too rich. Is it right to maintain that science values all empirical information? Indeed, does science value any?
(b) The answer given to this question by the formalist is negative: science values no information. This answer appears in its best light when seen as a response to the following dilemma: science does not value all information, since a theory that completes science need not be omniscient; but nor does science value only some information, since there is no non‐arbitrary distinction between the information science values and the rest. Formalism results when both horns of the dilemma are conceded.
Immodest weak comparative absolute AimCE—the species of formalism according to which the provision of theories that are weakly empirically adequate is the only constitutive criterion for the absolute comparative (p.108) evaluation of scientific activity—can be seen to fail prior to showing how substantivalism escapes this dilemma. It implies that constitutively, tautologies are as successful as any theory and more successful than any theory that is not weakly empirically adequate. Since scientists judge many informative theories that are not weakly empirically adequate to be more successful than tautologies, a proponent of immodest weak AimCE must take this judgement to employ criteria other than constitutive criteria of success. Indeed, he must resort to a brute insistence that it employs criteria of success external to science: it would be absurd to take informativeness to be evidence for, or a means to achieving, weak empirical adequacy.39 But this tactic is self‐defeating. By requiring a distinction among criteria of success that scientists actually employ that would be at least as arbitrary as any distinction between information that is, and information that is not, of interest to science, he would undermine the dilemma that is the sole support of formalism, and, hence, of his own position.
To refute immodest weak AimCE is not to refute formalism itself: in principle, formalism could advocate constitutive theoretical values other than weak empirical adequacy, and these could underwrite the judgement that, for example, Newton's theory of gravity (together with the standard auxiliaries) is more successful, constitutively, than a tautology. Because it is so compelling to think that it is the (empirical) informativeness of Newton's theory (together with the standard auxiliaries) that is crucial to this judgement, however, the formalist must place enormous weight on the dilemma he directs at substantivalism.
Empirical substantivalism rejects the second horn of the dilemma: it holds that science places no value on information that is not empirical. But with what right? After all, for reasons rehearsed above (in Section 2.1b), empiricism itself is compatible with the supposition that informativeness per se is a constitutive theoretical value. Again, however, to say that empiricism can accommodate this supposition is not to say that it should do so, and in this instance I am more sympathetic to van Fraassen's view. Scientific practice gives no reason to suppose that informativeness per se is a constitutive theoretical value. Once the criteria of explanatory power, simplicity, predictive utility, and unification (p.109) are factored out, I doubt that the history of science harbours cases in which the best explanation of a scientist's judgement that T1 is more successful than T2 is simply that he employs informativeness per se, rather than mere empirical informativeness, as a criterion of success.
Still, the dilemma with which the formalist confronted substantivalism is easily modified so as to focus exclusively on empirical substantivalism: science cannot value all empirical information, since a theory that completes science need not be omniscient about observables (so global empirical substantivalism is incorrect); but nor can science value only some empirical information, since there is no non‐arbitrary distinction to be drawn between the empirical information science values, and the empirical information science does not value (so local empirical substantivalism is incorrect).
Although local empirical substantivalism has its attractions,40 it suffers an intolerable defect of principle. If a more empirically informative theory T1 is claimed to be better by constitutive criteria than T2, there is no force whatsoever in an objection that holds, merely, that none of the additional empirical information T1 contains is of any scientific interest. Of course, strengthening a theory just by adding some arbitrary piece of empirical information need not improve it by constitutive criteria. But this is not because the information added might be of no scientific interest. It is because arbitrary information can reduce a theory's simplicity and internal coherence.41
The proponent of empirical substantivalism is better advised, therefore, to opt for the global variety. He is thereby obliged to rebut the first horn of the dilemma with which the formalist confronted him. To do so, he should concede that a theory that completes science need not be omniscient about observable matters, and then reconcile this fact with his claim that science values all empirical information. The latter task is easy. The reason such (p.110) a theory need not be omniscient about observables, even though science values all empirical information, is simply that empirical informativeness may conflict with other theoretical values that are constitutive. In particular, it may conflict with simplicity and internal coherence.42
The second important question to be addressed once constitutive criteria of success are admitted was seen to be this: what are the constitutive context‐sensitive criteria reflected in the judgement that the scientific activity that culminated in Newton's theories of motion and gravity was (non‐comparatively) successful as such?
This question is addressed by context‐sensitive non‐comparative AimCE. In effect, all versions of it hold that the main constitutive criterion for scientific activity at a context to be non‐comparatively successful is that the activity should provide us with a theory that is weakly empirically adequate with respect to the phenomena that are relevant at the context. The least demanding holds that context‐sensitive weak empirical adequacy in this sense is the only constitutive criterion. All versions also agree that a necessary condition of success is empirical informativeness with respect to the phenomena that are relevant at the context: substantivalist versions take this condition to be constitutive; formalist versions hold that it is not constitutive. One substantivalist version holds that weak empirical adequacy with respect to all phenomena is a constitutive criterion of success. What such versions of AimCE amount to depends on what account is given of what it is for a phenomenon to be ‘relevant’ at a context.43
(p.111) The thesis that weak empirical adequacy with respect to the phenomena relevant at a context is the sole constitutive criterion of success for scientific activity at the context can be rejected irrespective of how the notion ‘relevant’ is explained. No species of formalism is able to accommodate satisfactorily all the features of our judgement that the scientific activity that comprised Newton's dynamical researches was successful. This judgement doesn't just employ a criterion of (possibly restricted) weak empirical adequacy: if it did, we would be committed to saying that those researches would also have been successful if they had issued in mere tautologies. But nor do we employ in addition a criterion of informativeness that is non‐constitutive: if we did, we would be free to drop it so as to frame a purely constitutive evaluative judgement of the activity amounting to Newton's dynamical researches, and then we would again have the result that that activity would still have been successful as such even if it had issued in nothing but the most banal of tautologies.
So let us turn to substantivalist versions of context‐sensitive non‐comparative AimCE. Weak or strong, modest or immodest, it is tantamount to:
If (S) is read literally, nothing can be made of the main/subsidiary distinction in this case: (S) specifies two conditions that are individually necessary, and jointly sufficient, for constitutive non‐comparative success.
(S) For all contexts c, the main constitutive criterion for the non‐comparative success of scientific activity at c is that it should provide us with a theory that is weakly empirically adequate (at c), and the sole subsidiary constitutive criterion is that it should provide us with a theory that is fully empirically informative with respect to all the phenomena that are relevant at c.44
The inadequacies of (S) are stark. In one respect it is far too weak: at c, one might satisfy the condition it specifies as necessary and sufficient for success by simply listing all the phenomena that are relevant at c. The obvious way to try to strengthen it is to appeal to the additional constitutive theoretical (p.112) values the empiricist should recognize.45 Being comparative, these values alone cannot be treated as further necessary conditions, however. Since they determine a comparative success relation between theories, the most direct way of modifying (S) so as to utilize them yields:
(S1) has a virtue (S) lacks: it makes explicit the fact that scientific activity is sensitive to the competition. Whereas (S), and hence context‐sensitive non‐comparative AimCE, erroneously suggests that whether or not scientific activity is successful as such turns on a relation between the theory the activity provides and a part of the observable world selected by the context, (S1) respects the competitive and essentially communal nature of science.
(S1) For all contexts c, scientific activity at c is non‐comparatively successful as such iff it provides us with a theory that is (a) weakly empirically adequate, (b) fully informative with respect to all the phenomena that are relevant at c, and (c) such as to be more successful than all the rival theories that are relevant at c.46
In another respect, however (S), and hence (S1), is far too strong. It makes a demand even Newton's dynamical researches failed to satisfy: for those researches did not result in a theory that, coupled with the standard auxiliaries, is weakly empirically adequate. Nor does the judgement that Newton's dynamical researches were exemplary by scientific standards commit one to the view that the theory that resulted from them is fully informative with respect to all the relevant phenomena. But it isn't necessary for the empiricist to weaken the necessary conditions of success in (S) so as to accommodate these points. For he will hold that suitable weakenings of them are already incorporated into the condition by which (S) was strengthened so as to yield (S1). Accordingly, both respects in which (S) is deficient can be remedied by replacing (S) with:
(S2) is a schematic account of necessary and sufficient conditions for non‐comparative success as such that is neutral between empiricism and (p.113) anti‐empiricism. These doctrines need differ only over the absolute criteria that determine the relation of constitutive absolute comparative success between theories.
(S2) For all contexts c, scientific activity at c is non‐comparatively successful as such iff it provides us with a theory that stands in the constitutive absolute comparative relation ‘is more successful than’ to all the rival theories that are relevant at c.
Let us say that a necessary and sufficient condition for an instance of scientific activity to be successful as such is its ‘goal’. Since scientific activity in the narrowest sense is rational activity, its goal is also the goal of the agent (qua scientist). Formally, the goal (S2) assigns the agent of (an instance of) scientific activity at a context c is to produce a theory that improves upon rival theories that are relevant at c.
But how does the context in which scientific activity in the narrowest sense occurs determine which rival theories are relevant? When Newton's dynamical researches are judged to constitute scientific activity that is exemplary as such, the rival theories that are relevant are taken to be those then extant in the scientific community of which Newton was a part. So the notion of relevance has diachronic and synchronic dimensions: at a context, the goal of scientific activity in the narrowest sense is to improve upon rivals that, at the context, are neither future nor outside the agent's community. What counts as the scientific community at a context falls somewhere inside two extremes. At one extreme, the community coincides with the agent, so that the relevant rivals are restricted to the rivals of which he is cognizant. Supposing that the goal at a context is the provision of a theory that improves upon the theories known to the agent makes successful scientific activity too easy, however. This judgement might seem harsh, for example, on the precocious schoolboy who, upon learning Newtonian mechanics, Maxwell's equations, and the Michelson‐Morley experiment, sets himself the task of bettering Newton and reinvents Special Relativity. But it should be upheld. Science is an essentially communal activity. To engage in it, an agent must adopt the goal of bettering the rival efforts that have already been made in the community of scientists to which he belongs. At the other extreme, the scientific community at a context is taken to comprise all scientists contemporaneous with the context. But requiring scientific activity to have as its goal the bettering of all rival scientific activity would take anti‐individualism too far. When Newton embarked upon his dynamical speculations, although science obliged him to try to better dynamical theories then extant in the West, it did not oblige him to try to better the efforts, for example, of any extra‐terrestrials there might then have been.
The condition that is necessary and sufficient for the success as such of scientific activity at a context is to this extent externalist: the agent might be (p.114) unaware of rival theories that are relevant at his context. It is externalist in another respect too. Even if the agent knows all and only the relevant rivals, he can only evaluate his theory in the light of the pertinent phenomena he knows about. But the comparative success relation between theories holds contingently upon all the phenomena. Hence, even the scientific activity of an agent who is aware of all the rival theories T1. . .Tn relevant at his context, and of all the pertinent phenomena P1. . .Pn that are known at his context, and who succeeds in articulating a theory T* that better captures the Pi better than do any of the Ti, might be unsuccessful as such: in the light of further phenomenon unknown at his context, it might be that one of the Ti's is in fact more successful than T*.
This much externalism might seem unfair. It is one thing to judge scientific activity at a context on the basis of rivals known about at the context, since the communal nature of science renders it proper to demand that a scientific agent be familiar with the relevant science of his day. It is another to judge scientific activity on the basis of phenomena with which even the scientific community to which the agent belongs is entirely unfamiliar. I do not think the account should be made more internalist, however. The correct thing to say about a case like the one I describe is that the agent's scientific activity is not successful as such, although it seemed so to those who are ignorant of the phenomena that favour a theory that is a relevant rival.
Although the context‐sensitive necessary and sufficient condition for the constitutive non‐comparative success of scientific activity is heavily externalist, it remains a genuine goal: the scientific agent strives for it. In so doing, he will try to produce a theory that, when judged in the light of the pertinent phenomena he knows about, seems more successful than the rival theories he knows about. In a sense, from his own point of view, he can do no better. Nevertheless, a more internalist objective of this kind is only a means to the constitutive goal.
I have argued that scientific activity (in the narrow sense) has a context‐sensitive goal, but that science does not because the absolute constitutive criteria of success that govern it are comparative. Although the context‐sensitive goal is defined in terms of the absolute criteria, in a way it is more fundamental: to (p.115) understand science one needs first to know what goal is adopted by agents who pursue scientific activity.
How these conclusions engage AimCE is difficult to judge, since van Fraassen's advocacy of it seems to point in opposite directions. The non‐comparativeness of weak empirical adequacy, and the connection he makes between constitutive criteria of success and the notion of a telos, pull one towards a non‐comparative interpretation. But the apparent comparativeness of empirical informativeness pulls one towards an interpretation on which the intended thesis is comparative AimCE. Moreover, even on a non‐comparative interpretation it is unclear whether his focus is upon the idea of an absolutely completing theory (or theories), or the context‐sensitive goal of scientific activity: AimCE's talk of the provision of ‘theories’ in the plural suggests the latter, but the absoluteness of empirical adequacy and (apparently) empirical informativeness suggests the former.
My account of the context‐sensitive goal of scientific activity is neutral between empiricism and anti‐empiricism: it only becomes empiricist once empiricist constitutive absolute theoretical values are plugged into it. I argued against formalism and local empirical substantivalism (especially the pragmatist version) that these values include empirical information as such. Again, it is unclear how this conclusion engages AimCE: I do not know whether van Fraassen is a formalist, or a pragmatist, or a global empirical substantivalist. One thing that is certain, however, is that when I argued that the empiricist should include explanatory power, simplicity, unification, and internal coherence amongst the constitutive absolute theoretical values, I took issue with his preferred development of AimCE.47
Grimes, T. (1987), ‘Explanation and the Poverty of Pragmatics’, Erkenntnis 27: 79–92.
Kitcher, P., and Salmon, W. (1987), ‘Van Fraassen on Explanation’, Journal of Philosophy 84: 315–30.
Kuhn, T. S. (1973), ‘Objectivity, Value Judgment, and Theory Choice’, in Kuhn 1977, chapter 13.
(p.116) Kuhn, T. S. (1977), The Essential Tension (Chicago: University of Chicago Press).
Ladyman, J., Douven, I., Horsten, L., and van Fraassen, B. (1997), ‘A Defence of van Fraassen's Critique of Abductive Inference: Reply to Psillos’, The Philosophical Quarterly 47: 305–21.
Laudan, L. (1984), Science and Values (Berkeley: University of California Press).
McMullin, E. (2003), ‘Van Fraassen's Unappreciated Realism’, Philosophy of Science 70: 455–78.
Maher, P. (1993), Betting on Theories (Cambridge: Cambridge University Press).
Newton‐Smith, W. H. (1981), The Rationality of Science (London: Routledge and Kegan Paul).
Popper, K. (1972), Objective Knowledge (Oxford: Clarendon).
van Fraassen, B. (1980), The Scientific Image (Oxford: Clarendon).
—— (1983), ‘Glymour on Evidence and Explanation’, in J. Earman (ed.), Testing Scientific Theories (Minneapolis: University of Minnesota Press).
—— (1985), ‘Empiricism in the Philosophy of Science’, in P. M. Churchland and C. A. Hooker (eds.), Images of Science (Chicago: University of Chicago Press), 245–308.
—— (1989), Laws and Symmetry (Oxford: Clarendon).
—— (2000), ‘The False Hopes of Traditional Epistemology’, Philosophy and Phenomenological Research 60: 253–80.
—— (2002), The Empirical Stance (New Haven: Yale University Press).
(1) Van Fraassen (1980: 12). Van Fraassen defines constructive empiricism as holding, in addition, that acceptance of a theory involves as belief only that the theory is empirically adequate. I think this second thesis less fundamental (and I think he does too). In ‘An Empiricist Critique of Constructive Empiricism II: Theory‐Acceptance’ (forthcoming) I argue that it is false, and that, pace van Fraassen, it is not a corollary of the thesis about the aim of science to which I give primacy.
(4) The distinction between criteria constitutive of scientific evaluation, and mere means to, or evidence for, constitutive success, generalizes van Fraassen's (1983: 165–6) (non‐exhaustive) distinction between ‘informational’ virtues of a theory, which amount to the theory's telling us ‘what the world is like’, and ‘confirmational’ virtues of a theory, which are ‘features that give us more reason to believe . . . [it] (or some part of it) to be true’. I preface my explanation in the text with ‘roughly’ so as to avoid commitment to the claim that necessarily, satisfaction of one constitutive criterion of success suffices for being successful. I use ‘constitutive aim/criterion’ for ‘aim/criterion constitutive of science’ unless I explicitly say otherwise.
(5) Cf. van Fraassen's (1989: 193) remark that ‘the aim of science is not truth as such but only empirical adequacy . . . Acceptance of a theory involves as belief only that the theory is empirically adequate . . . To put it another way: acceptance is acceptance as successful, and involves the opinion that the theory is successful—but the criterion of success is not truth in every respect, but only truth with respect to what is actual and observable’ (his italics, my underlining). Note too the ease with which, in the previously quoted remark, he qualifies his claim about the aim of science by observing that there are ‘theoretical virtues’ other than empirical adequacy.
(6) See too van Fraassen (1980: 12). In more traditional terms, the analogue of this definition is that a theory is empirically adequate iff it does not entail any false proposition only about observables. Although the notion ‘observable,’ and in particular van Fraassen's view of its epistemology and extension, is controversial, in what follows I take it for granted. So doing is legitimate because my goal is an empiricist critique of AimCE.
(7) ‘Informativeness’ is ambiguous between ‘contentfulness’ and ‘true contentfulness’. In what follows, I use it and its cognates in the former sense.
(10) The italics are van Fraassen's. The underlining is mine.
(11) What non‐comparative property could it naturally be taken to be? The property of expressing some empirical information about some observables?! Moreover, if empirical informativeness is taken to be a non‐comparative property that, like weak empirical inadequacy, is a constitutive criterion for applying the non‐comparative predicate ‘is successful’, modest weak AimCE would have to be explained as stating that these criteria are individually necessary and jointly sufficient. For what could it mean to say that there are exactly two constitutive criteria for the application of a predicate F, the main criterion G and a subsidiary criterion H, unless what is meant is that a necessary and sufficient condition of something's being F is that it is both G and H? But this explanation ignores, and therefore renders otiose, the main/subsidiary distinction. That it does so tells against an interpretation of modest weak AimCE on which it is required.
(12) Admittedly, thus interpreted AimCE is absurd, if giving more weight to the ‘main’ criterion means that weak empirical adequacy is ‘trumping’ in the sense that no scientific activity that yields a theory that is not weakly empirically adequate is more successful as such than any scientific activity that does yield such a theory: the provision of a tautology is certainly not more successful in this sense than was Newton's provision of his theory of gravity (together with the usual auxiliary hypotheses). It might be taken to mean that to be more successful than a weakly empirically adequate theory (such as a tautology), a theory that is not weakly empirically adequate must be very empirically informative. But this too is false: a ‘theory’ that is merely close to being weakly empirically adequate doesn't have to bear much empirical information in order to be more successful by constitutive criteria than a tautology. Better still, then, is a non‐literal interpretation on which the main constitutive criterion is said to be the comparative relation ‘closer to being weakly empirically adequate than’. The point of the qualification ‘main’ can be taken to be that this relation is heavily weighted in comparison with the other constitutive comparative criteria.
(13) A comparative interpretation has its drawbacks. In particular, it doesn't gel with van Fraassen's (1989: 189) elucidatory remark that ‘[i]t is part of the straightforward description of any activity, communal or individual, large‐scale or small, to describe the end that is pursued as one of its defining conditions.’ Whereas the ‘end’ of an activity is something that can be realized so as to complete it, comparative AimCE does not specify an end in this sense.
(14) Compare such claims as ‘science aims to give us theories that have increasing verisimilitude’ and ‘science aims to give us theories that have increasing problem‐solving capacity.’ This second tradition is not always singled out, and claims that belong to it are sometimes erroneously construed as belonging to the first. This happens, for example, when they are viewed as mere weakenings of what their proponents take to be untenably strong accounts of what is required for a theory to complete the entire scientific enterprise. This interpretation renders them absurd, however: the scientific enterprise is not completed, for example, when two theories are successively formulated, the second of which has greater verisimilitude than the first.
(15) Here, ‘phenomenon’ means ‘fact about observable thing or event’.
(17) AimSR is the thesis that the main aim of science is the provision of theories that are true. Among the interpretations of van Fraassen's texts considered, the one that does most violence to his texts takes his notion of ‘empirical adequacy’ to be context‐sensitive empirical adequacy.
(18) Radical historicism is the view that no criteria of success are constitutive of science at all times. Historicism says some are; ahistoricism says none are. This issue is independent of the one debated by absolutists and contextualists. In particular, contextualism is compatible with ahistoricism, and absolutism is compatible with radical historicism.
(19) I say ‘nothing’ rather than ‘no theory’ so as to respect the possibility that ‘is more successful than (by constitutive criteria)’ is genuinely polyadic, in that one or both terms of it may be plural.
(21) Where C is the sole criterion of non‐comparative success, the best one can do is: ‘x is more successful than y’ =df ‘x is C and y is not.’ One should not suppose that a better definition is provided by ‘x is closer than y to being C,’ for this is not in terms of C. The mere supposition that C is the criterion of non‐comparative success need not be attended by a natural comparative notion of closeness to being C. Nor, in cases where some such notion is natural, need it contain a requirement that this notion serves as a criterion of comparative success: if the constitutive criterion of non‐comparative success is hitting the bullseye, there is nothing untoward in a practice that relegates ‘is closer to hitting the bullseye’ to a non‐constitutive criterion of comparative instrumental value. (While the ultimate goal of a serious athlete is winning Olympic gold, I have heard such athletes say that winning silver means nothing to them.)
(22) This conclusion has consequences for how to resolve the internal tension in the species of weak, modest, non‐comparative, and absolute AimCE that takes empirical informativeness to be the sole subsidiary constitutive theoretical value. Since resolving it by giving up the intuitive comparativeness of empirical informativeness results in a doctrine that does not address a question that is fundamental, it is better resolved by giving up the non‐comparativeness of AimCE.
(23) I have been unable to ascertain where van Fraassen stands on this issue. Compelled to bet, I would say he is either a formalist or an advocate of the context‐sensitive pragmatist species of empirical substantivalism.
(27) (E1) should be taken to be stronger than the doctrine that nothing is an objective epistemic reason of this kind, since van Fraassen (2000: footnote 19) maintains that no epistemic reasons for believing anything are objective. One way to think of an utterance of the form ‘there are no epistemic reasons for believing p’ as (merely) expressing a stance, instead of stating a thesis, is to view it as being just an expression of disdain. (Cf. van Fraassen 1985: 255.) A similar qualification applies to (E2) below.
(29) Van Fraassen (1980) does not distinguish between these alternatives. The distinction arises from the observation that a feature that does not function as a constitutive criterion of success may serve some other function that is nevertheless constitutive. In particular, if a feature F is but one possible means to realizing another feature G that is constitutive of successful H‐ing, the utilization of F as a means to G may or may not be constitutive of H‐ing. The first and second alternatives are closely related. A constitutive means to constitutive success must be (taken to be) an epistemic reason for constitutive success. Some epistemic reasons are candidates for being constitutive means; others, in particular evidential effects of constitutive success, are not.
(30) Nor need one believe a theory in terms of which one explains phenomena. One would only sincerely assent to the question ‘Does Newton's theory of gravitation give the correct explanation of the tides?’ if one accepted Newton's theory. But this much implies that when giving explanations one cannot utilize theories one does not believe true, only if accepting a theory involves believing it. Once an appropriate conceptual distinction between acceptance and belief is drawn, however, this supposition is undermined. (I develop a novel defence of the distinction between belief and acceptance in ‘An Empiricist Critique of Constructive Empiricism II: Theory‐Acceptance’ (forthcoming).)
(31) Van Fraassen (1980: 95). I doubt no scientists ever ‘enlarge’ their theories simply to achieve greater explanatory power. Indeed, I think some of the cases van Fraassen has in mind are of the form ‘scientist X's proposed empirically equivalent but more explanatory theory is not taken up by the scientific community.’ However, I take his argument to be that such cases would not arise were explanatory power a constitutive theoretical value.
(32) I have formulated this explanation in such a way as not to presuppose that the value scientists place on empirical informativeness is constitutive. It is thereby made available to the formalist. It becomes more compelling, however, if formalism is rejected and the value scientists place on empirical informativeness taken to be constitutive. (I argue that formalism should be rejected in part (b) of this section.)
(33) Reflecting on what would happen if a theory with perfect predictive power were developed, Newton‐Smith (1981: 31) says that ‘the scientific enterprise would continue in the face of this awesome achievement. No doubt some would abandon science and no doubt society would lessen its monetary contributions to science, but science would not end. For we do not wish merely to predict, we also want to explain.’
(36) Whereas substantivalism must weight informativeness against some such criterion as closeness to being weakly empirically adequate (or true), the weighting is surely not objective. But the opposing thesis—formalism—is false. (See part (b) of this section.)
(38) Since weak empirical adequacy is non‐comparative it is ill‐suited to the task at hand. Rather, it is the comparative closeness to weak empirical adequacy that is amongst the absolute criteria that, once weighted, determine an absolute comparative success relation between theories.
(39) Its absurdity invites the question as to why van Fraassen (1980: 8) includes informativeness in the list of subsidiary criteria that ‘may be . . . [mere] means to [achieving or promoting the main aim]’ (my italics). Could he be thinking of the main aim as the provision of theories that are strongly empirically adequate, and that the provision of theories that are non‐empirically informative is a means to this end?
(40) Certainly local non‐empirical substantivalism has found adherents. Maher (1993: 214) claims that science values some information, but that there is no information such that science values it, while Popper (1972: 191) hints at a pragmatist local substantivalism when he claims that ‘it is the aim of science to find satisfactory explanations, of whatever strikes us as being in need of explanation’ (my underlining).
(41) Similarly, I think Maher (1993: 243) is wrong to suggest, when arguing that weak empirical adequacy is not a constitutive theoretical value, that ‘an individual who . . . cares only for whether . . . theories fit observable phenomena that occur within a billion light years of us . . . [need not have] values [that] are unscientific.’ If God assured us that some phenomenon more than a billion light years from us falsifies quantum mechanics, by the canons of science this would reveal a defect in quantum mechanics.
(42) Since global empirical substantivalism cannot resist the first horn of the dilemma in this way in the absence of constitutive theoretical values other than (closeness to) weak empirical adequacy and empirical informativeness, the version of comparative absolute AimCE said in part (a) of this section to give an inadequate account of what would be sufficient to complete science is vulnerable from the opposite direction: it cannot give an adequate account of what would be necessary to complete science either if the thought that a competing theory need not be omniscient about unobservables is not only plausible, but true.
(43) I do not know where van Fraassen stands with respect to these alternatives. Indeed, it is unclear even whether he intends AimCE to address the question at issue. Although his formulations of AimCE tend to be non‐comparative, he tends to speak of the only constitutive criteria of success he countenances, empirical adequacy and, more problematically, empirical informativeness, as if they are absolute. Of themselves, however, constitutive absolute comparative criteria are powerless to yield, e.g. the judgement that Newton's theory of gravitation is non‐comparatively successful: it is not successful in relation to all theories (nor, even, to all theories formulated at some time or other)! Perhaps, then, van Fraassen is an absolutist who takes this judgement to invoke criteria of success at least some of which are not constitutive (as the formalist proponent of non‐comparative immodest weak AimCE must do). Alternatively, he might recognize that the judgement issues from criteria of success, all of which are constitutive, but be concerned only to have identified the absolute criteria involved.
(44) A theory is ‘fully empirically informative’ with respect to certain phenomena iff it expresses all true propositions that are solely about those phenomena.
(45) In Section 2.2a.
(46) At least from an empiricist point of view, whether one theory is a rival of another turns on the relation between their empirical contents. To a first approximation, a substantial overlap is necessary and sufficient.
(47) I am grateful to Roger Young for organizing, and inviting me to speak at, the conference on Bas's work at which I first presented some of the ideas in this chapter, and to Jim Edwards, Adam Reiger, and Alan Weir, for helpful comments and conversation. Thanks too to Bradley Monton for his excellent editorship, and to Bas for the beautiful Image.