2.1. Driesch's Vitalism

One example of an unacceptable explanation that supports (P1) comes from the scientists who argued for the existence of some sort of vital force in animals from the eighteenth to the early twentieth centuries. There were certain facts of organic development that did not appear to be amenable to explanation in mechanical, physico-chemical terms. For example, Hans Driesch, working at the turn of the twentieth century, discovered that sea urchin embryos that were halved at a very early stage of development could still generate two normally proportioned (though somewhat smaller) adult urchins. He later discovered that other organisms, such as water hydras, had similar abilities to develop into normal adult forms even when the cells of the early forms were perturbed in various ways. In order to explain his observations of the sea urchin and other organisms, he introduced the concept of an entelechy: a non-physico-chemical causal factor that directs the overall development of the organism from a simple zygote into a complex adult that forms an integrated whole. Driesch characterizes an entelechy as “the relation of localities in the absolutely normal case” of the organism (Reference Driesch1929, 91). After presenting his sea urchin findings, Driesch writes that “we are required to accept a new kind of elementary … process, because no machine produced by physical and chemical means can be contrived as the basis for the events in question” (Driesch Reference Driesch1899; quoted in Sander Reference Sander1992, 3). Driesch thinks that no such machine is possible because “a machine … cannot remain itself if you remove parts of it or rearrange its parts at will” (Reference Driesch1929, 104). Put simply, a dishwasher that has been halved cannot still function as a dishwasher, and Driesch claims that the same principle holds for any machine; thus, he concludes, the developing embryo cannot be a machine.

For Driesch, as the first of the above quotations indicates, the entelechy's existence is justified on the grounds of its apparent explanatory power: such an entity guiding development would explain facts that were previously unexplained or considered inexplicable. In “natural science,” Driesch claims, “something new and elemental must always be introduced whenever what is known of other elemental facts is proved to be unable to explain the facts in a new field of investigation” (Reference Driesch1929, 105). Despite the fact that the vitalist hypothesis explained these otherwise unexplained facts, most of Driesch's contemporaries did not accept his hypothesis (Sander Reference Sander1992, Reference Sander1993), and far more current scientists would reject the introduction of entelechies—despite the fact that there are still unexplained developmental facts today. Driesch's critics observed that introducing entelechies into the scientific account of embryonic development did not generate any new testable claims, and Driesch agreed (Jennings Reference Jennings1912, 435). Furthermore, positing a specific entelechy for each species does not unify apparently disparate facts. First, each species requires its own specific entelechy; that is, we must posit as many distinct entelechies as there are species of organisms. Second, the claims ‘Sea urchins can develop normally when dismembered at an early stage of development’ and ‘Hydras can develop normally when dismembered at an early stage of development’ are unified by the inductive generalization that ‘Many (or all) organisms can develop normally if dismembered at an early stage of development’. There is no further unification added by the claim that each organism has an entelechy that regulates its development. This is analogous to the virtus dormativa case, in which the inductive generalization ‘Opium makes everyone sleepy' unifies the two facts that opium makes John sleepy and makes Jane sleepy, but these two facts are not further unified by postulating a virtus dormativa in the opium. Of course, if further assumptions were made about the manner in which an entelechy or a virtus dormativa functions, it is conceivable that new predictions could be generated or distinct sets of phenomena unified; however, without such further assumptions, I submit that scientists would be unwilling to accept explanations involving an entelechy or a virtus dormativa.

2.2. Kepler's Astronomical Model of Nested Platonic Solids

The centerpiece of Kepler's Mysterium Cosmographicum provides another example of an explanation that both is unacceptable and explains only one already accepted fact. In the preface, Kepler writes, “There were three things in particular, about which I persistently sought the reasons why they were such and not otherwise: the number, the size, and the motions of the heavenly spheres” ([1596] Reference Kepler and Duncan1981, 63). That is, Kepler wants to explain why there were exactly six planets (instead of five, or seven, or …) and why the distances between the planets are what they are (instead of more or less). To explain why there are six planets, he appeals to the five perfect (or Platonic) solids: if these five were nested one inside the other and there were globes in between each as well as inside the innermost solid and outside the outermost, there would be one globe for each of the six planets’ (average) orbital paths. And if the solids are ordered properly, the distance between the planets’ orbits approximately matches the distance between these globes. These two explanations are not ones that scientists in Kepler's time were (for the most part) willing to accept (Caspar Reference Caspar and Hellman[1950] 1959, 69–71)—or, for that matter, in our own time. Each explanans-parameter explains exactly one previously accepted fact: correlating the five Platonic solids to six planetary orbs explains nothing more than why there are six planets, and the particular ordering of the nested solids explains nothing but the previously known (approximate) distances between the planets. Neither explanans generates new experimental predictions, and neither unifies apparently disparate realms of previously accepted facts.Footnote ⁴

2.3. ‘Just-So’ Evolutionary Stories

Another type of explanation that scientists do not accept is the ‘just-so story’ about the evolutionary histories of traits observed today. Critics of adaptationism claim that adaptationists’ explanations of current traits are just-so stories (most famously in Gould and Lewontin Reference Gould and Lewontin1978), whereas adaptationists claim that this characterization is an unfair caricature of their work. Whether adaptationists’ work actually resembles just-so stories enough to condemn it is irrelevant for present purposes: both sides to the dispute agree that just-so stories should not be accepted. A just-so evolutionary story has the following form: from the existence of a certain trait T observed in an organism today, infer that a certain set of selective pressures P that would favor selection of T operated on the organism's ancestors (Sterelny and Griffiths Reference Sterelny and Griffiths1999, 243). In the caricature version of adaptationism, the only evidence that the selective pressures P existed is that we observe trait T now. Under those circumstances, selective pressures P constitute an explanation of a single already-accepted claim (that the organism has observed trait T).

Both adaptationists and their critics agree that, in order for a particular evolutionary explanation to be scientifically acceptable, there must be more evidence for the explanans than merely the current trait it was introduced to explain. Gould writes, “Not all sociobiology proceeds in the mode of storytelling for individual cases. It rests on firmer methodological ground when it seeks broad correlations across taxonomic lines, as between reproductive strategy and distribution of resources, for example, or when it can make testable, quantitative predictions” (Reference Gould1978, 531). Current evolutionary psychologists are also accused of telling just-so stories. However, leading evolutionary psychologists, such as Leda Cosmides, hold that their theories are not explanations that explain only one already accepted phenomenon: “the method of evolutionary psychology outlined here … is hypothetico-deductive, rather than speculative. In a speculative approach, one first discovers a psychological mechanism, and then one speculates about what adaptive problem it evolved to solve. The approach advocated here is the reverse. … It is a constrained and predictive approach, rather than a compilation of post hoc explanations for known phenomena” (Reference Cosmides1989, 190). What Cosmides calls the ‘speculative’ method is precisely the kind of explanation I am claiming that scientists reject. She holds that her theories do stand on what Gould called “firmer methodological ground,” since they make new predictions and even reveal “previously unknown cognitive features” (Cosmides and Tooby Reference Cosmides, Tooby, Mehler and Franck1995, 91) that are given a unified treatment (instead of being a ‘post hoc compilation’) within her theory. As Elisabeth Lloyd points out (Reference Lloyd1999, 212), both the adaptationists and their critics agree on what an acceptable explanation looks like in the abstract: it must make new predictions or provide a unified account of apparently different phenomena (such as Gould's “broad correlations across taxonomic lines”). The dispute is, in large part, whether particular adaptationist explanations meet these general methodological standards. Just-so stories, along with vitalism, Kepler's nested-solids model, and virtus dormativa explanations, all illustrate and substantiate (P1): an explanans that explains only a single piece of established data is not considered scientifically acceptable—even if it is the only explanans currently available. As Lipton (Reference Lipton1991) notes, proper abductive reasoning permits inference to the best explanation only when the best is good enough: if our best explanans for some fact fails to meet a minimal threshold for explanatory goodness, abductive reasoners should not accept it. My claim is that the four cases presented in this section fail to meet this minimal standard.

Cases of scientifically unacceptable explanations that fit (P1) could be multiplied. Perhaps Newton's famous refusal to offer an explanation for the existence of gravity could be understood as motivated by something like (P1). A prominent recent example of such an unacceptable explanation may be found in so-called intelligent design (ID) theory. We again observe the same pattern seen in Driesch and the others: ID's proponents point to some widely accepted fact that (purportedly) has not yet been given a satisfactory evolutionary explanation, such as the blood-clotting cascade or the locomotive system of bacterial flagella, and claim that the existence of an intelligent designer who intervened to create the cascade and the motor would explain these previously unexplained facts. I will not here delve into the details of ID or other unacceptable explanations that fit the pattern specified in (P1); instead, I turn to three pressing objections to (P1).

3.1. Examples’ Unacceptability Misdiagnosed

One might grant that all four of the above attempted explanations are scientifically unacceptable but object that I have not properly identified the true reasons for their unacceptability. For example, one could argue that the only reason scientists reject entelechies is that they are nonphysical entities that act on the physical world, and such a form of causation is unintelligible or unscientific; therefore, the standard described in (P1) is irrelevant to scientists’ rejection of entelechies. This was the explicit, professed rationale of Driesch's contemporary Wilhelm Roux for rejecting the concept of entelechy: “since it is presumed to rule over material structure like a sovereign, this entelechy is not acceptable to exact natural science” (Reference Roux1912, 241). Similarly, the objector continues, Kepler's explanation of the number of and distances between the planets is motivated by the assumption that a conscious, beneficent being created the universe.Footnote ⁵ What is fundamentally unacceptable in Kepler's case, the objector alleges, is the theological motivation and grounding of the nested-solids explanation. And virtus dormativa explanations, as mentioned above, could be dismissed on the grounds of being somehow covertly circular. So these cases are problematic because they are unacceptably metaphysical, theological, or circular, not because they qualify under (P1) as unacceptable explanations.

This is a serious objection; I offer two replies. First, in the case of just-so stories, the hypotheses concerning the environments of our Pleistocene ancestors lack the metaphysics or theology found in Driesch and Kepler. Thus, even if the objection as stated above is accepted, we can fall back on the just-so stories example to substantiate (P1). Second, the reason scientists today reject certain theological and metaphysical accounts of phenomena may well be that such accounts generate no new predictive content and fail to unify disparate phenomena genuinely, despite being on the scientific table for 2,500 years.Footnote ⁶ If the posit of entelechies could be used to make novel empirical predictions that could not be made without it, then scientists could conceivably accept it if the predictions were borne out—just as quantum mechanics, with all its counterintuitive strangeness, was accepted several decades ago. Similarly, any theological hypothesis that did generate new predictive content could be (at least arguably) considered scientific.Footnote ⁷ In short, we may interpret theology and metaphysics as failing to meet scientific standards because they lack both predictive and unifying power.

3.2. Successful Ad Hoc Hypotheses

One might object to (P1) by pointing to the phenomenon of what could be called ‘successful ad hoc hypotheses’. When a scientific theory appears to conflict with some piece of data, defenders of the theory often generate an ad hoc hypothesis in order to ‘save’ the theory. The ad hoc hypothesis is introduced solely to ‘explain away’ the anomalous observation. In general, scientists cast a disapproving eye on ad hoc maneuvers; this disapprobation constitutes good evidence for (P1). However, in some cases, ad hoc hypotheses are vindicated and lead to unforeseen discoveries: the hackneyed example of the discovery of Neptune via an apparent violation of Newton's laws illustrates this point vividly. LeVerrier and Adams suggested the existence of a planet beyond Uranus, on the grounds that the existence of such a planet would explain exactly one accepted fact: the observed anomalous behavior of Uranus. Such successful ad hoc hypotheses appear to provide a counterexample to my claim that scientists do not accept explanations that neither make novel predictions nor unify disparate phenomena.

However, when scientists finally accepted the hypothesis that a planet existed beyond Uranus, it was not solely on the grounds that the Uranus data were anomalous: the later, direct telescopic observation of the new planet triggered the acceptance of the hypothesis. Thus, the hypothesis of a planet beyond Uranus is not like the three cases above, for it generated new empirical predictions that were later vindicated and thus does not rest on a single source of evidence but on two: both the apparent anomalies in Uranus's behavior and the telescopic observations of a planet in the predicted place. Other successful ad hoc hypotheses will likely have the same structure: although they may be initially introduced in order to account for a single (anomalous) piece of evidence, they will be accepted only when they generate new, testable predictions that are independently corroborated by experiment, thereby receiving evidential support from a different source than the original one.

3.3. Self-Evidencing Explanations

A final, apparently decisive objection can be drawn from a concept recently discussed by Peter Lipton (Reference Lipton1991, 26), which Hempel calls ‘self-evidencing explanations’. An explanation is “self-evidencing,” Hempel says, “only if, at the time of its presentation, the occurrence of the explanandum event provides the only evidence, or an indispensable part of the only evidence, available in support of some of the explanans-statements” (Reference Hempel1965, 372–373). If such explanations are routinely accepted in science, that would apparently constitute decisive evidence against (P1). Hempel claims that we do find such explanations in science, but “an explanation that is self-evidencing may for that reason rest on a poorly supported explanans and may therefore have no strong claim to empirical soundness” (373). Thus Hempel appears to be agreeing with my primary claim. However, immediately thereafter, he writes that “even this is not inevitable,” which appears to spell trouble for (P1). Hempel gives an example of the circumstances under which a self-evidencing explanation would be accepted: “In the case of the absorption spectrum of a star, for example, the previously accepted background information, including the relevant theories, may indicate that the dark lines observed occur only if the specified elements are present in the star's atmosphere, and then the explanandum, in conjunction with the background information, lends very strong support to the crucial explanatory hypothesis” (373). If self-evidencing explanations are accepted only under these types of circumstances, then my thesis is safe. In this case, the background information includes ‘If the detector registers pattern P, then elements $E_{1},\,\ldots ,\,E_{n}$ must be present in the source’, and the observation we wish to explain is that the detector registers pattern P. Thus the explanans ‘Elements $E_{1},\,\ldots ,\,E_{n}$ are present in the source’ is a deductive consequence of the explanandum and the background assumptions. The entelechies of Driesch, the nested Platonic solids of Kepler, and just-so evolutionary stories are clearly not deductive consequences of their respective explananda and background theories. (We should probably also allow an explanation to count as self-evidencing when the background assumptions include probabilistic claims; in the above example, this would amount to something like ‘In 99% of the cases in which the detector registers pattern P, elements $E_{1},\,\ldots ,\,E_{n}$ are present in the source’. But entelechies etc. fare no better on this probabilistic standard.)

4.1. The No-Miracles Argument

The NMA is, according to many of its prominent defenders, an instance of inference to the best explanation (IBE), also known as abductive inference.Footnote ⁹ The basic form of abductive inference is:

1. (1) q is the best explanation of p

2. (2) p

3. ∴ q

How is the NMA supposed to instantiate this schema? The target explanandum (p above) is, in Psillos's words, “the overall empirical success of science” (Reference Psillos1999, 71), that is, the claim that mature scientific theories make many accurate observable predictions (in alternative formulations, they are ‘empirically adequate’ or ‘instrumentally successful’).Footnote ¹⁰ The proponent of the NMA claims that the empirical success of mature theories is best explained by the truth or approximate truth of such theories (and, in some versions, also by such theories’ theoretical terms referring successfully). The claim that mature scientific theories are approximately true is, for proponents of the NMA, the thesis of scientific realism—or at least an indispensable component of that thesis. Then, because IBE licenses the inference from a fact to the best explanation of that fact, the proponent of the NMA infers scientific realism from the mature sciences’ empirical success. The inference can be cast, following the schema above, as follows:

1. (1) The (approximate) truth of mature scientific theories is the best explanation of their empirical success.

2. (2) Mature scientific theories are empirically successful.

3. ∴ Mature scientific theories are (approximately) true.

This account of the NMA closely follows Psillos (Reference Psillos1999, 71). Let us call the conclusion of this argument ‘scientific realism’.

However, pinning down the meaning of ‘mature’ in the statement of scientific realism can create difficulties.Footnote ¹¹ ‘Mature’ cannot mean ‘empirically successful’ because then the explanandum of the NMA becomes the tautologous ‘Empirically successful theories are empirically successful’, and tautologies do not require explanation. And ‘mature’ presumably cannot merely be shorthand for a list of particular theories current scientists find praiseworthy, for the particle physics of the next century will presumably count as a mature scientific theory too, even if it is as different as quantum mechanics is from its classical counterpart. Perhaps the best gloss of ‘mature scientific theories’ is semiostensive: ‘these theories (pointing to general relativity, evolutionary biology, etc.) and those relevantly similar to them’. Or, if one truly despairs of finding a useful, noncircular gloss of ‘mature’, one could give up on this formulation of realism entirely and search for another that does not crucially rely on the notion of a science's maturity.

Part of the appeal of the NMA, for its proponents, is that it is supposedly part of a thoroughly naturalistic philosophy of science, where ‘naturalism’ is the view that “the epistemology of science should employ no methods other than those used by the scientists themselves” (Psillos Reference Psillos1999, 73). Similarly, Boyd writes, “The epistemology of empirical science is an empirical science” (Reference Boyd1989, 13). On this view, the philosopher who argues for scientific realism via the NMA is using the same methods and epistemological standards as a scientist who uses inference to the best explanation to infer the existence of a new type of fundamental particle. Doppelt puts the point clearly: “Explanationists [= proponents of the NMA] adopt a naturalistic stance in epistemology and claim that their scientific realism is a scientific hypothesis justified by the very sort of abductive inference (IBE) effectively employed by scientists in producing knowledge” (Reference Doppelt2005, 1079). Since science is rife with abductive inference, Putnam claims that scientific realism is “viewed … as part of the only scientific explanation of the success of science, and hence as part of any adequate description of science and its relations to its objects” (Reference Putnam1975, 73; my italics). Boyd articulates a similar view: “the scientific realist's abductive argument” is “the only scientifically plausible explanation for the reliability of a scientific methodology which is so theory-dependent” (Reference Boyd1989, 9; my italics; see also Reference Boyd, Asquith and Giere1980, 622). However, the type of explanation that the NMA uses to explain the empirical success of science is exactly the kind of explanation, discussed in Section 2, that scientists do not accept; that is, realism neither makes new predictions nor unifies previously disparate claims. Thus, Putnam and Boyd are wrong to claim that realism is a “scientific explanation.”

4.2. Realism Generates No New Predictions

Realism generates no new empirical predictions over and above those already made by the claim that mature scientific theories are empirically successful, just as postulating a virtus dormativa does not make any new predictions over and above those already made by the claim that opium makes people sleepy. Quantum electrodynamics makes certain experimental predictions, but adding ‘QED is approximately true (and not merely empirically adequate)’ to the standard formulation of QED will not yield any new predictions. Similarly, adding the thesis of scientific realism to the conjunction of all our mature scientific theories (whichever those may be) does not generate any novel predictions either, over and above the predictions already made by adding the generalization ‘Mature scientific theories are empirically successful’. Doppelt makes this point as well: “explanationists do not use a novel-prediction standard in giving their own naturalistic justification of scientific realism. Psillos and other explanationists clearly assume that the ability of … scientific realism, properly formulated, instantaneously to explain already well-known phenomena—the success of science—can confirm it and make it empirically successful, independently of novel prediction” (Reference Doppelt2005, 1080). It is widely held that epistemic force accrues to particular scientific theories, in part, because they make novel predictions that are borne out. The thesis of scientific realism found in the NMA apparently enjoys no such probative force.

A realist might urge that her position does generate novel predictions:Footnote ¹² specifically, realism explains why novel uses or extensions of existing theories are empirically successful, a fact that is supposedly not otherwise predicted (for a well-developed articulation of this realist view, see Leplin Reference Leplin1997).Footnote ¹³ But this view seems unfair to the opponent of realism. The antirealist can endorse the explanandum of the NMA, ‘Mature sciences are empirically successful’. This sentence obviously does not have the same meaning as ‘Mature sciences are empirically successful in nonnovel applications’. But the realist could make a prediction that the antirealist could not only if the antirealist were committed merely to the latter and not to the former. For ‘Mature scientific theories are empirically successful’, which an antirealist accepts, presumably predicts that novel extensions of a theory, being themselves (parts of) theories, will also be empirically successful. Thus, the realist hypothesis does not make a prediction that is not already made by ‘Mature sciences are empirically successful’.

In response, the realist might say, “Let's modify the NMA. Change the explanandum to ‘Mature scientific theories have been empirically successful in previously encountered cases'. I now take this as my original justification or ‘inductive base' for believing that mature theories are approximately true. Then, adding the realist hypothesis to my sum total of beliefs generates a new prediction over and above the explanandum, namely, novel, future extensions of mature theories will also be empirically successful.” This response is problematic, however. Consider the analogous case being made for virtus dormativa: imagine Molière's doctor says that the original justification for ‘Opium makes people sleepy’ is that it has made a certain set of people sleepy; from this pattern, he infers a virtus dormativa. Then the doctor makes the following prediction, based on his postulate of a virtus dormativa: when further people, not on the original list of opium takers, are exposed to opium, they will also become sleepy. Then, when this prediction is borne out, he counts that as evidence of a dormative virtue. However, most would agree that Molière's doctor does not have evidence for his virtus dormativa hypothesis; rather, this is simply more evidence for the generalization that opium makes people sleepy. The defense of scientific realism currently under consideration appears relevantly similar; thus, when novel predictions are borne out, this should be taken as evidence for the generalization that mature scientific theories are empirically successful (perhaps adding ‘and not only for the phenomena they were originally introduced to explain’). Put otherwise, all these supposedly novel predictions that the realist makes can also be made without appeal to theories’ approximate truth, by instead merely appealing to ‘Mature scientific theories are approximately true’.Footnote ¹⁴

Finally, as referees suggested to me, one might hold that realism makes a prediction about the historical record of scientific development. Realists hold that empirical success is a sign of truth. Supposing that the realists are correct, and theories that enjoy a great deal of empirical success are usually approximately true, what would the historical record of scientific development then look like?Footnote ¹⁵ It is not obvious that this should predict anything about the course of scientific history. However, a realist may suggest that if realism is correct, we would expect to see empirically successful theories replaced by radically different alternatives only very rarely (rare replacements must be allowed, since everyone is a fallibilist). To reach that prediction (‘Most empirically successful theories are not replaced by radically different rivals') as a conclusion from the original premise (‘Most empirically successful theories are approximately true'), we need an additional premise, namely, that approximately true theories (once found) are not replaced by radically different rivals.

Unfortunately, there are a number of problems with the realist's generation of this prediction. First, there are questions of operationalization: how do we count historical theories? And how rare, exactly, must radical replacement be for the historical record to count as evidence for realism? Let us set those questions aside (in part because they apply to anyone who wants to cite the historical record as evidence, such as the antirealist who appeals to the pessimistic induction). Second, either the argument we used to generate the prediction has a premise that is most likely false or the conclusion does not follow (even inductively). If the additional premise we supplied in order to make the prediction is understood to be the exceptionless ‘All approximately true theories (once found) are not replaced by radically different rivals’, then the conclusion follows, but this premise is very likely false: it is highly improbable that humans never give up an approximately true theory once they find it. However, if we weaken the added premise to ‘Most approximately true theories (once found) are not replaced by radically different rivals’ (which might be true), then we no longer have an acceptable inductive argument form, since

1. Most A are B

2. Most B are C

3. ∴ Most A are C

is not a legitimate inductive argument pattern.Footnote ¹⁶ Finally, even if this problem could be overcome, and realism somehow managed to make a definite prediction that massive theoretical upheavals are relatively rare, it is not clear that this is a new prediction, that is, one that is not also generated by other commonly held beliefs about science shared by realist and antirealist alike. In particular, scientific development is recognized to favor, in general, more conservative changes over more radical ones. That is, when a theory must be modified and two possible modifications that would both save the phenomena are available, the modification that yields the smaller theoretical change is chosen, ceteris paribus. This can explain the (purported) relative scarcity of massive upheavals in the historical record without appealing to the more contentious realist claim that empirically successful theories are approximately true. In sum, it is not clear that realism can make a prediction about the historical record, and even if it can, it is not a new prediction, since the prediction can be made from other widely accepted claims about scientific theory choice.

4.3. Realism Fails to Unify Disparate Claims

Even if scientific realism did not generate any new predictions, might it nonetheless be an acceptable explanation, on the grounds that it unifies previously disparate claims, just as Newton's law of universal gravitation unifies patterns in terrestrial free fall, the paths of planets and comets, and tidal behavior? I believe the answer is no: realism, as it occurs in the NMA, is an explanation of only one pattern, namely, the ‘overall empirical success of mature sciences’.Footnote ¹⁷ The proponent of the NMA might object to this last claim as follows: scientific realism unifies the empirical success of particle physics with the empirical success of evolutionary biology, organic chemistry, and so on. Thus, the objector continues, scientific realism does rest on different sources of evidence, so the NMA is not akin to the objectionable kinds of explanations scientists are unwilling to accept. However, this apparent unification is spurious: it is analogous to a vitalist claiming that because sea urchins, water hydras, and other organisms can all develop into more mature forms despite certain perturbations in their early embryonic stages, positing an entelechy for each organism provides a unified account for these regularities. As discussed earlier, the fact that several organisms exhibit this developmental regularity may well provide evidence for the inductive generalization that all organisms exhibit similar regularities. However, the fact that many organisms can survive such perturbations does not provide evidence that there is a specific entelechy in each species responsible for this ability. Again, ‘Opium makes John sleepy' and ‘Opium makes Jane sleepy' are already unified by the fact that opium makes people sleepy; positing a dormative virtue in addition to explain the inductive generalization achieves no further unification. The analogy with the NMA should be clear: the empirical successes of general relativity, organic chemistry, and evolutionary biology provide evidence for the inductive generalization ‘Mature sciences are empirically successful', but any substantive ‘unifying work’ is done at the level of this inductive generalization, which is the explanandum of scientific realism. No further unification is achieved by additionally claiming ‘These mature sciences are (approximately) true’.

The defender of the NMA can make a second, related objection to my argument. The NMA advocate notes that (P1)–(P3) of my argument are compatible with the existence of acceptable abductive inferences in particular scientific disciplines and points out that scientists make proper abductive inferences to the existence of electrons and their properties, the existence of the weak force and its properties, and so on. For present purposes, I am willing to grant this. Suppose, the objector continues, that we simply form a giant conjunction of the conclusions of each of these proper abductive inferences: ‘Electrons exist, have rest mass of $9.1066\times 10^{-28}$ g, have charge of $-1.602\times 10^{-19}$ coulombs, … [whatever other properties our best theory attributes to them], and DNA is a double helix, composed of four types of nucleotides arranged on a sugar and phosphate backbone, where adenine and thymine are always paired and cytosine and guanine are always paired and … [whatever other properties our best theory attributes to DNA] and … [all the rest of the contents of current mature scientific theories]’. Call this sentence the giant conjunction. Now, one might think that the NMA is defensible if ‘Mature sciences are (approximately) true’ means nothing more than the giant conjunction, though it is far from clear that the giant conjunction is what many scientific realists intend their thesis to mean (see the next paragraph). However, this response runs into a thicket of problems involving intertheoretic relationships among the various sciences: the conjuncts of the giant conjunction will include the claims of both quantum mechanics and general relativity, and the compatibility of these two theories is dubious at best (Barrett Reference Barrett2003). Similar sorts of problems appear elsewhere: there is substantive empirical evidence that the concept of a gene varies widely among the subdisciplines of the biological sciences (e.g., Stotz, Griffiths, and Knight Reference Stotz, Griffiths and Knight2004). The upshot of this is that the NMA proponent's objection holds water only if all our mature scientific theories are compatible. I grant that this is certainly a possibility, but there are strong indications (in addition to the two just cited) that our sciences exhibit significant disunity. In response to such strong disunity, a defender of the NMA could conceivably lean on the qualifier ‘approximately’ that appears in the statement of realism and suggest that inconsistent theories might nonetheless be approximately true, though not true simpliciter. If we set aside the problems faced by the notion of approximate truth (Miller Reference Miller1974), perhaps a sufficiently ingenious account of inconsistency in scientific theories can save this proposal. However, it stretches credulity to accept that a contradictory sentence can be even approximately true.Footnote ¹⁸ At least, I would sooner give up scientific realism than accept that a contradiction can approximate the truth.

Finally, for the sake of argument, let us bracket the above worry about the disunity of the sciences and suppose that our sciences (or the portions of our sciences that the NMA defender wishes to endorse) are consistent. What would the ‘giant conjunction’ form of the NMA look like then? I think that there are at least two options: the fact being explained is just a giant conjunction of either all the predictions of the various theories or (as before) the empirical success of mature scientific theories. On option 1, the first premise will take the form ‘General relativity is the best explanation of gravitational lensing, the precession of Mercury's perihelion, and … [all the other phenomena general relativity purports to explain], and the theory of evolution via common descent is the best explanation of homologies, the fossil record, and … , and quantum mechanics is the best explanation of the results of the double-slit experiment, the observed violation of Bell's inequalities, and … , and …’.Footnote ¹⁹ The second premise will consist of descriptions of gravitational lensing etc. and of homologies etc. and of the double-slit experiment etc. and … . The conclusion will be descriptions of quantum mechanics and general relativity and evolution by common descent and … .

Now it is clear that such an argument is a ‘mere conjunction’ of the individual abductive inferences used in the particular sciences, so there is no substantive unification here and, furthermore, no argument for any sort of ‘global’ realism. In the context of the NMA, these individual abductive inferences have been called ‘first-order’ inferences to the best explanation. Certain philosophers who find value in the NMA are concerned only to defend first-order instances of IBE (e.g., Worrall Reference Worrall1989). However, many proponents of the NMA want to push a ‘second-order’ IBE: the success of the first-order inferential practice is best explained by the fact that our scientific practices usually deliver (approximate) truth. Psillos clearly falls into the second camp: “NMA is not just a generalization over scientists’ abductive inferences. … The (first-order) instances of explanatory reasoning involve the claim that it is reasonable to accept that particular theories [e.g., general relativity or quantum mechanics] are relevantly approximately true. … NMA is a kind of meta-abduction. The explanandum of the NMA is a general feature of scientific methodology—its reliability for yielding correct predictions” (Reference Psillos1999, 79). So, with this terminology, this first version of the giant-conjunction NMA is simply a conjunction of all the first-order abductive inferences. The considerations presented in Section 2 concerning the kinds of explanations scientists are unwilling to accept would not affect this sort of NMA. But many realists endorse Psillos's global conception of realism; the giant-conjunction NMA as described above simply does not capture their view, so they cannot appeal to it in response to the argument presented here.

Could there be a giant-conjunction version of the NMA that does do justice to global realism? This is the second option mentioned above. The first premise would be ‘General relativity and quantum mechanics and evolution by natural selection and … best explain the empirical success of science’. The second premise would then be, as in the original formulation of the NMA, ‘Mature sciences are empirically successful’. This certainly better captures the kind of global realism Psillos espouses. However, the first premise is probably false: the general theory of relativity does not explain why scientific theories reliably yield correct predictions; rather, it explains gravitational lensing phenomena and the precession in the perihelion of Mercury. Similarly, quantum mechanics explains the results of the two-slit experiment, not the empirical success of scientific theorizing.

In sum, the giant-conjunction defense of NMA does not withstand scrutiny since it suffers from one or more of the following problems (depending on how the defense is construed). The giant-conjunction defense (i) founders on the disunity of science (since the disjunction would likely be inconsistent), (ii) fails to capture the ‘global’ or second-order character of the realist thesis NMA is often invoked to justify, or, finally, (iii) contains a false premise, since particular sciences, taken in isolation or together, do not explain why scientific methodology so often generates correct predictions. Thus, scientific realism explains the empirical success of science no better than a virtus dormativa explains the soporific effects of opium.