1. What is Scientific Realism?

1.1 Epistemic Achievements versus Epistemic Aims

It is perhaps only a slight exaggeration to say that scientific realism is

characterized differently by every author who discusses it, and this

presents a challenge to anyone hoping to learn what it is. Fortunately,

underlying the many idiosyncratic qualifications and variants of the

position, there exists a common core of ideas, typified by an

epistemically positive attitude towards the outputs of scientific

investigation, regarding both observable and unobservable aspects of the

world. The distinction here between the observable and the unobservable

reflects human sensory capabilities: the observable is that which can,

under favourable conditions, be perceived using the unaided senses (for

example, planets and platypuses); the unobservable is that which cannot

be detected this way (for example, proteins and protons). This is to

privilege vision merely for terminological convenience, and differs from

scientific conceptions of observability, which generally extend to things

that are detectable using instruments (Shapere 1982). The distinction

itself has been problematized (Maxwell 1962, Churchland 1985,

Musgrave 1985, Dicken & Lipton 2006), but if it is problematic, this is

arguably a concern primarily for certain forms of antirealism, which

adopt an epistemically positive attitude only with respect to the

observable. It is not ultimately a concern for scientific realism, which

does not discriminate epistemically between observables and

unobservables per se.

Before considering the nuances of what scientific realism entails, it is

useful to distinguish between two different kinds of definition in this

context. Most commonly, the position is described in terms of the

epistemic achievements constituted by scientific theories (and modelsthis qualification will be taken as given henceforth). On this approach,

scientific realism is a position concerning the actual epistemic status of

theories (or some components thereof), and this is described in a number

of ways. For example, most define scientific realism in terms of the truth

or approximate truth of scientific theories or certain aspects of theories.

Some define it in terms of the successful reference of theoretical terms

to things in the world, both observable and unobservable. (A note about

the literature: theoretical term, prior to the 1980s, was standardly used to denote terms for unobservables, but will be used here to refer to any

scientific term, which is now the more common usage.) Others define

scientific realism not in terms of truth or reference, but in terms of belief

in the ontology of scientific theories. What all of these approaches have

in common is a commitment to the idea that our best theories have a

certain epistemic status: they yield knowledge of aspects of the world,

including unobservable aspects. (For definitions along these lines, see

Smart 1963, Boyd 1983, Devitt 1991, Kukla 1998, Niiniluoto 1999,

Psillos 1999, and Chakravartty 2007a.)

Another way to think about scientific realism is in terms of the epistemic

aims of scientific inquiry (van Fraassen 1980, p. 8, Lyons 2005). That is,

some think of the position in terms of what science aims to do: the

scientific realist holds that science aims to produce true descriptions of

things in the world (or approximately true descriptions, or ones whose

central terms successfully refer, and so on). There is a weak implication

here to the effect that if science aims at truth and scientific practice is at

all successful, the characterization of scientific realism in terms of aim

may then entail some form of characterization in terms of achievement.

But this is not a strict implication, since defining scientific realism in

terms of aiming at truth does not, strictly speaking, suggest anything

about the success of scientific practice in this regard. For this reason,

some take the aspirational characterization of scientific realism to be too

weak (Kitcher 1993, p. 150, Devitt 2005, n. 10, Chakravartty 2007b, p.

197)it is compatible with the sciences never actually achieving, and even the impossibility of their achieving, their aim as conceived on this

view of scientific realism. Most scientific realists commit to something

more in terms of achievement, and this is assumed in what follows.

1.2 The Three Dimensions of Realist Commitment

The description of scientific realism as a positive epistemic attitude

towards theories, including parts putatively concerning the

unobservable, is a kind of shorthand for more precise commitments

(Kukla 1998, ch. 1, Niiniluoto 1999, ch. 1, Psillos 1999, Introduction,

Chakravartty 2007a, ch. 1). Traditionally, realism more generally is

associated with any position that endorses belief in the reality of

something. Thus, one might be a realist about one's perceptions of tables

and chairs (sense datum realism), or about tables and chairs themselves

(external world realism), or about mathematical entities such as numbers

and sets (mathematical realism), and so on. Scientific realism is a

realism about whatever is described by our best scientific theoriesfrom this point on, realism here denotes scientific realism. But what, more precisely, is that? In order to be clear about what realism in the

context of the sciences amounts to, and to differentiate it from some

important antirealist alternatives, it is useful to understand it in terms of

three dimensions: a metaphysical (or ontological) dimension; a semantic

dimension; and an epistemological dimension.

Metaphysically, realism is committed to the mind-independent existence

of the world investigated by the sciences. This idea is best clarified in

contrast with positions that deny it. For instance, it is denied by any

position that falls under the traditional heading of idealism, including some forms of phenomenology, according to which there is no world

external to and thus independent of the mind. This sort of idealism,

though historically important, is rarely encountered in contemporary

philosophy of science, however. More common rejections of mind-

independence stem from neo-Kantian views of the nature of scientific

knowledge, which deny that the world of our experience is mind-

independent, even if (in some cases) these positions accept that the

world in itself does not depend on the existence of minds. The

contention here is that the world investigated by the sciencesas distinct from the world in itself (assuming this to be a coherent distinction)is in some sense dependent on the ideas one brings to scientific investigation, which may include, for example, theoretical

assumptions and perceptual training; this proposal is detailed further in

section 4. It is important to note in this connection that human

convention in scientific taxonomy is compatible with mind-

independence. For example, though Psillos (1999, p. xix) ties realism to

a mind-independent natural-kind structure of the world, Chakravartty (2007a, ch. 6) argues that mind-independent properties are often

conventionally grouped into kinds (see also Boyd 1991 and Humphreys

2004, pp. 2225, 3536).

Semantically, realism is committed to a literal interpretation of scientific

claims about the world. In common parlance, realists take theoretical

statements at face value. According to realism, claims about scientific entities, processes, properties, and relations, whether they be observable

or unobservable, should be construed literally as having truth values,

whether true or false. This semantic commitment contrasts primarily

with those of so-called instrumentalist epistemologies of science, which

interpret descriptions of unobservables simply as instruments for the

prediction of observable phenomena, or for systematizing observation

reports. Traditionally, instrumentalism holds that claims about

unobservable things have no literal meaning at all (though the term is

often used more liberally in connection with some antirealist positions

today). Some antirealists contend that claims involving unobservables

should not be interpreted literally, but as elliptical for corresponding

claims about observables. These positions are described in more detail in

section 4.

Epistemologically, realism is committed to the idea that theoretical

claims (interpreted literally as describing a mind-independent reality)

constitute knowledge of the world. This contrasts with sceptical

positions which, even if they grant the metaphysical and semantic

dimensions of realism, doubt that scientific investigation is

epistemologically powerful enough to yield such knowledge, or, as in

the case of some antirealist positions, insist that it is only powerful

enough to yield knowledge regarding observables. The epistemological

dimension of realism, though shared by realists generally, is sometimes

described more specifically in contrary ways. For example, while many

realists subscribe to the truth (or approximate truth) of theories

understood in terms of some version of the correspondence theory of

truth (as suggested by Fine 1986 and contested by Ellis 1988), some

prefer deflationary accounts of truth (including Giere 1988, p. 82, Devitt

2005, and Leeds 2007). Though most realists marry their position to the

successful reference of theoretical terms, including those for

unobservable entities, processes, properties, and relations (Boyd 1983,

and as described by Laudan 1981), some deny that this is a requirement

(Cruse & Papineau 2002, Papineau 2010). Amidst these differences,

however, a general recipe for realism is widely shared: our best

scientific theories give true or approximately true descriptions of

observable and unobservable aspects of a mind-independent world.

1.3 Qualifications and Variations

The general recipe for realism just described is accurate so far as it goes,

but still falls short of the degree of precision most realists offer. The two

main sources of imprecision here are found in the general recipe itself,

which makes reference to the idea of our best scientific theories and the notion of approximate truth. The motivation for these qualifications is perhaps clear. If one is to defend a positive epistemic attitude

regarding scientific theories, it is rational to do so not merely in

connection with any theory (especially when one considers that, over the

long history of the sciences up to the present, some theories were not or

are not especially successful), but rather with respect to theories that

would appear, prima facie, to merit such a defence, viz. our best theories.

And it is widely held, not least by realists, that even many of our best

scientific theories are likely false, strictly speaking, hence the

importance of the notion that theories may be close to the truth (that is, approximately true) even though they are false. The challenge of

making these qualifications more precise, however, is significant, and

has generated much discussion.

Consider first the issue of how best to identify those theories that realists

should be realists about. A general disclaimer is in order here: realists

are generally fallibilists, holding that realism is appropriate in

connection with our best theories even though they likely cannot be

proven with absolute certainty; some of our best theories could

conceivably turn out to be significantly mistaken, but realists maintain

that, granting this possibility, there are grounds for realism nonetheless.

These grounds are bolstered by restricting the domain of theories

suitable for realist commitment to those that are sufficiently mature and

non-ad hoc (Worrall 1989, pp. 153-154, Psillos 1999, pp. 105108). Maturity may be thought of in terms of the well established nature of the

field in which a theory is developed, or the duration of time a theory has

survived, or its survival in the face of significant testing; and the

condition of being non-ad hoc is intended to guard against theories that

are cooked up (that is, posited merely) in order to account for some known observations in the absence of rigorous testing. On these

construals, however, both the notion of maturity and the notion of being

non-ad hoc are admittedly vague. One strategy for adding precision here

is to attribute these qualities to theories that make successful, novel

predictions. The ability of a theory to do this, it is commonly argued,

marks it as genuinely empirically successful, and the sort of theory to

which realists should be more inclined to commit (Musgrave 1988,

Lipton 1990, Leplin 1997, White 2003, Hitchcock & Sober 2004, Barnes

2008; for a dissenting view, see Harker 2008).

The idea that with the development of the sciences over time, theories

are converging on (moving in the direction of, getting closer to) the truth, is a common theme in realist discussions of theory change (for

example, Hardin & Rosenberg 1982 and Putnam 1982). Talk of

approximate truth is often invoked in this context, and has produced a

significant amount of often highly technical work, conceptualizing the

approximation of truth as something that can be quantified, such that

judgments of relative approximate truth (of one proposition or theory in

comparison to another) can be formalized and given precise definitions.

This work provides one possible means by which to consider the

convergentist claim that theories can be viewed as increasingly

approximately true over time, and this possibility is further considered in

section 3.4.

A final and especially important qualification to the general recipe for

realism described above comes in the form of a number of variations.

These species of generic realism can be viewed as falling into three

families or camps: explanationist realism; entity realism; and structural

realism. There is a shared principle of speciation here, in that all three

approaches are attempts to identify more specifically the component

parts of scientific theories that are most worthy of epistemic

commitment. Explanationism recommends realist commitment with

respect to those parts of our best theoriesregarding (unobservable) entities, processes, laws, etc.that are in some sense indispensible or otherwise important to explaining their empirical successfor instance, components of theories that are crucial in order to derive successful,

novel predictions. Entity realism is the view that under conditions in

which one can demonstrate impressive causal knowledge of a putative

(unobservable) entity, such as knowledge that facilitates the

manipulation of the entity and its use so as to intervene in other

phenomena, one has good reason for realism regarding it. Structural

realism is the view that one should be a realist, not in connection with

descriptions of the natures of things (like unobservable entities and

processes) found in our best theories, but rather with respect to their

structure. All three of these positions adopt a strategy of selectivity, and

this and the positions themselves are considered further in section 2.3.

2. Considerations in Favour of Scientific Realism (and Responses)

2.1 The Miracle Argument

The most powerful intuition motivating realism is an old idea,

commonly referred to in recent discussions as the miracle argument or no-miracles argument, after Putnam's (1975, p. 73) claim that realism is the only philosophy that doesn't make the success of science a miracle. The argument begins with the widely accepted premise that our best theories are extraordinarily successful: they facilitate empirical

predictions, retrodictions, and explanations of the subject matters of

scientific investigation, often marked by astounding accuracy and

intricate causal manipulations of the relevant phenomena. What explains

this success? One explanation, favoured by realists, is that our best

theories are true (or approximately true, or correctly describe a mind-

independent world of entities, properties, laws, structures, or what have

you). Indeed, if these theories were far from the truth, so the argument

goes, the fact that they are so successful would be miraculous. And

given the choice between a straightforward explanation of success and a

miraculous explanation, clearly one should prefer the non-miraculous

explanation, viz. that our best theories are approximately true (etc.). (For

elaborations of the miracle argument, see Brown 1982, Boyd 1989,

Lipton 1994, Psillos 1999, ch. 4, Barnes 2002, Lyons 2003, Busch 2008,

and Frost-Arnold 2010.)

Though intuitively powerful, the miracle argument is contestable in a

number of ways. One sceptical response is to question the very need for

an explanation of the success of science in the first place. For example,

van Fraassen (1980, p. 40; see also Wray 2007, 2010) suggests that

successful theories are analogous to well-adapted organismssince only successful theories (organisms) survive, it is hardly surprising that our

theories are successful, and therefore, there is no demand here for an

explanation of success. It is not entirely clear, however, whether the

evolutionary analogy is sufficient to dissolve the intuition behind the

miracle argument. One might wonder, for instance, why

a particular theory is successful (as opposed to why theories in general

are successful), and the explanation sought may turn on specific features

of the theory itself, including its descriptions of unobservables. Whether

such explanations need be true, though, is a matter of debate. While

most theories of explanation require that the explanans be true,

pragmatic theories of explanation do not (van Fraassen 1980, ch. 5).

More generally, any epistemology of science that does not accept one or

more of the three dimensions of realismcommitment to a mind-independent world, literal semantics, and epistemic access to

unobservableswill thereby present a putative reason for resisting the miracle argument; these positions are considered in section 4.

Some authors contend that the miracle argument itself is an instance of

fallacious reasoning called the base rate fallacy (Howson 2000, ch. 3,

Lipton 2004, pp. 196198, Magnus & Calendar 2004). Consider the following illustration. There is a test for a disease for which the rate of

false negatives (negative results in cases where the disease is present) is

zero, and the rate of false positives (positive results in cases where the

disease is absent) is one in ten (that is, disease-free individuals test

positive 10% of the time). If one tests positive, what are the chances that

one has the disease? It would be a mistake to conclude that, based on the

rate of false positives, the probability is 90%, for the actual probability

depends on some further, crucial information: the base rate of the

disease in the population (the proportion of people having it). The lower

the incidence of the disease at large, the lower the probability that a

positive result signals the presence of the disease. By analogy, using the

success of a scientific theory as an indicator of its approximate truth

(assuming a low rate of false positivescases in which theories far from the truth are nonetheless successful) is arguably, likewise, an instance of

the base rate fallacy. The success of a theory does not by itself suggest

that it is likely approximately true, and since there is no independent

way of knowing the base rate of approximately true theories, the chances

of it being approximately true cannot be assessed. Worrall (2009)

maintains that these contentions are ineffective against the miracle

argument because they depend crucially on a misleading formalization

of it in terms of probabilities.

2.2 Corroboration

One motivation for realism in connection with at least some

unobservables described by scientific theories comes by way of

corroboration. If an unobservable entity or property is putatively capable of being detected by means of a scientific instrument or

experiment, one might think that this could form the basis of a defeasible

argument for realism regarding it. If, however, that same entity or

property is putatively capable of being detected by not just one, but

rather two or more different means of detectionforms of detection that are distinct with respect to the apparatuses they employ and the causal

mechanisms and processes they are described as exploiting in the course

of detectionthis may serve as the basis of a significantly enhanced argument for realism. Hacking (1983, p. 201; see also Hacking 1985, pp.

146147) gives the example of dense bodies in red blood platelets that can be detected using different forms of microscopy. Different

techniques of detection, such as those employed in light microscopy and

transmission electron microscopy, make use of very different sorts of

physical processes, and these operations are described theoretically in

terms of correspondingly different causal mechanisms. (For similar

examples, see Salmon 1984, pp. 217219, and Franklin 1986, pp. 166168, 1990, pp. 103115.)

The argument from corroboration thus runs as follows. The fact that one

and the same thing is apparently revealed by distinct modes of detection

suggests that it would be an extraordinary coincidence if the supposed

target of these revelations did not, in fact, exist. The greater the extent to

which detections can be corroborated by different means, the stronger

the argument for realism in connection with their putative target. The

argument here can be viewed as resting on an intuition similar to that

underlying the miracle argument: realism based on apparent detection

may be only so compelling, but if different, theoretically independent

means of detection produce the same result, suggesting the existence of

one and the same unobservable, then realism provides a good

explanation of the consilient evidence, in contrast with the arguably

miraculous state of affairs in which theoretically independent techniques

produce the same result in the absence of a shared target. The idea that

techniques of (putative) detection are often constructed or calibrated

precisely with the intention of reproducing the outputs of others,

however, may stand against the argument from corroboration.

Additionally, van Fraassen (1985, pp. 297298) argues that scientific explanations of evidential consilience may be accepted without the

explanations themselves being understood as true, which once again

raises questions about the nature of scientific explanation.

2.3 Selective Optimism/Scepticism

In section 1.3, the notion of selectivity was introduced as a general

strategy for maximizing the plausibility of realism, particularly with

respect to scientific unobservables. This strategy is adopted in part to

square realism with the widely accepted view that most if not all of even

our best theories are false, strictly speaking. If, nevertheless, there are

aspects of these theories that are true (or close to the truth) and one is

able to identify these aspects, one might then plausibly cast one's realism

in terms of an epistemically positive attitude towards those aspects of

theories that are most worthy of epistemic commitment. The most

important variants of realism to implement this strategy are

explanationism, entity realism, and structural realism. (For related work

pertaining to the notion of selectivity more generally, see Miller 1987,

chs. 810, Fine 1990, Jones 1991, and Musgrave 1992.)

Explanationists hold that a realist attitude can be justified in connection

with unobservables described by our best theories precisely when

appealing to those unobservables is indispensible or otherwise important

to explaining why these theories are successful. For example, if one

takes successful novel prediction to be a hallmark of theories worthy of

realist commitment generally, then explanationism suggests that, more

specifically, those aspects of the theory that are essential to the

derivation of novel predictions are the parts of the theory most worthy of

realist commitment. In this vein, Kitcher (1993, pp. 140149) draws a distinction between the presuppositional posits or idle parts of theories, and the working posits to which realists should commit. Psillos (1999, chs. 56) argues that realism can be defended by demonstrating that the success of past theories did not depend on their

false components: it is enough to show that the theoretical laws and mechanisms which generated the successes of past theories have been

retained in our current scientific image (p. 108). The immediate challenge to explanationism is to furnish a method with which to

identify precisely those aspects of theories that are required for their

success, in a way that is objective or principled enough to withstand the

charge that realists are merely rationalizing post hoc, identifying the

explanatorily crucial parts of past theories with aspects that have been

retained in our current best theories. (For discussions, see Chang 2003,

Stanford 2003a, 2003b, Elsamahi 2005, McLeish 2005, 2006, Saatsi

2005a, Lyons 2006, and Harker 2010.)

Another version of realism that adopts the strategy of selectivity is entity

realism. On this view, realist commitment is based on the putative ability

to causally manipulate unobservable entities (like electrons or gene

sequences) to a high degreefor example, to such a degree that one is able to intervene in other phenomena so as to bring about certain effects.

The greater the ability to exploit one's apparent causal knowledge of

something so as to bring about (often extraordinarily precise) outcomes,

the greater the warrant for belief (Hacking 1982, 1983, Cartwright 1983,

ch. 5, Giere 1989, ch. 5). Belief in scientific unobservables thus

described is here partnered with a degree of scepticism about scientific

theories more generally, and this raises questions about whether

believing in entities while withholding belief with respect to the theories

that describe them is a coherent or practicable combination (Morrison

1990, Elsamahi 1994, Resnik 1994, Chakravartty 1998, Clarke 2001,

Massimi 2004). Entity realism is especially compatible with and nicely

facilitated by the causal theory of reference associated with Kripke

(1980) and Putnam (1985/1975, ch. 12), according to which one can

successfully refer to an entity despite significant or even radical changes

in theoretical descriptions of its properties; this allows for stability of

epistemic commitment when theories change over time. Whether the

causal theory of reference can be applied successfully in this context,

however, is a matter of dispute (see Hardin & Rosenberg 1983, Laudan

1984, Psillos 1999, ch. 12, and Chakravartty 2007a, pp. 5256).

Structural realism is another view promoting selectivity, but in this case

it is the natures of unobservable entities that are viewed sceptically, with

realism reserved for the structure of the unobservable realm, as

represented by certain relations described by our best theories. All of the

many versions of this position fall into one of two camps: the first

emphasizes an epistemic distinction between notions of structure and

nature; the second emphasizes an ontological thesis. The epistemic view

holds that our best theories likely do not correctly describe the natures of

unobservable entities, but do successfully describe certain relations

between them. The ontic view suggests that the reason realists should

aspire only to knowledge of structure is that the very concept of entities

that stand in relations is metaphysically problematicthere are, in fact, no such things, or if there are such things, they are in some sense

emergent from or dependent on their relations. One challenge facing the

epistemic version is that of articulating a concept of structure that makes

knowledge of it effectively distinct from that of the natures of entities.

The ontological version faces the challenge of clarifying the relevant

notions of emergence and/or dependence. (On epistemic structural

realism, see Worrall 1989, Psillos 1995, 2006, Votsis 2003, and

Morganti 2004; regarding ontic structural realism, see French 1998,

2006, Ladyman 1998, Psillos 2001, 2006, Ladyman & Ross 2007, and

Chakravartty 2007a, ch. 3).