This paper investigates the relationship between naturalized epistemology and the realism-antirealism debate, by analyzing Michael Devitt's interesting defense of realism from naturalized epistemology. My analyses are centered upon his two key claims, i.e., (1) the sceptical problematic should be abandoned because it is unanswerable; and (2) when we take science for granted, we automatically commit ourselves to realism. To the first claim, I propose a revised sceptical problematic to avoid the problem. To evaluate the second claim, I compare it with Arthur Fine's NOA (natural ontological attitude). NOA does not involve such a commitment, and I show that we do not have to accept any more than NOA. The combination of my two conclusions consists of an argument against Devitt's defense of realism.
Recently philosophers of science had tended to adopt some kind of a naturalistic attitude towards science. Some use a naturalistic attitude to defend realism, and some to defend anti- or non-realism. Among the former, Michael Devitt's Realism and Truth (Devitt 1997, henceforth RT) proposes a strong and interesting defense of common-sense realism and scientific realism using naturalized epistemology. The purpose of this paper is to analyze this argument critically, especially by comparing it with Arthur Fine's (1986a, 1986b) "natural ontological attitude", an example of non-realistic naturalism. My conclusion will be that despite Devitt's own claim, his argument fails to show that we have more than the natural ontological attitude. Also, my argument for a revised sceptical problematic supports the view that we should prefer the natural ontological attitude to Devitt's realism.
2. Devitt's argument for realism
2.1 Ontology and epistemology
Devitt's argument for realism is based upon several methodological maxims, among which the most important and characteristic one is Maxim 3: "Settle the realism issue before any epistemic or semantic issue" (RT, 4). There are several arguments to defend this maxim. First, as we will see soon, Devitt argues that if we start from epistemology, we end up with instantaneous solipsism (75). Second, we have much more knowledge in ontology (by physics, biology and so forth) than in epistemology or semantics (232). Moreover, since epistemology and semantics are problems about people and language, they are just a tiny part of the comprehensive theory of the world provided by science (232, 284). Therefore, to start from semantics is to "put the semantic cart before the realist horse" (4). This is one of the most controversial claims in this book, as I will discuss below.
2.2 What is realism?
In chapter 2 of RT, Devitt clarifies what he means by "realism." Basically his realism has two dimensions, i.e., the independence dimension and the existence dimension (RT 14-22).
The independence dimension of realism is that the world and entities exist objectively, externally and independently of us. But what does the phrase "independently of us" mean? According to Devitt, this means that "it is not constituted by our epistemic values, by our capacity to refer to it, by the synthesizing power of the mind, by our imposition of concepts, theories or languages" (RT, 15; emphasis in original). Please note that the definition is compatible with the idea that the world is knowable. Note also that mere objectivity is not enough for independence, because a mental object can be objective too (to make this point clear, Devitt introduces the expression to exist "independently of the mental"; 20).
The existence dimension of realism is that to be a realist one should commit to not only the existence of external world, but also to a "structured set of entities" (RT, 17) in the world. Actually Devitt thinks that this is still weak, and he claims that the realism that is worth fighting for holds that "we are more or less right" in the common-sense and scientific entities we posit (18).
With these considerations, Devitt formulates his "Realism" (with capital R) as follows (RT, 23):
Tokens of most current common-sense and scientific physical types objectively exist independent of the mental.
As a subdivision of this Realism, Devitt introduces the distinction between "Common-Sense Realism" and "Scientific Realism" (25-26). Common-sense Realism is realism about observable entities, and Scientific Realism is about non-observable entities.
Before we move to his defense of this view, we should note one characteristic of this definition of Realism. He does not use the notion of truth to define Realism. This shows a strong contrast with some influential definitions of realism (e.g., van Fraassen 1980, 8). In this sense, in Hacking's terminology, this is "realism about entities," not "realism about theories" (Hacking 1983, 27). This is in accord with maxim 3, because, if he uses the notion of truth in the definition of realism, he should settle the matter of the truth (i.e. semantics) before the realism issue.
2.3 Naturalized epistemology and defense of Realism
The key move in Devitt's argument for Realism is to dismiss the sceptical problematic, starting from naturalistic ontology. First, Devitt formulates the Cartesian sceptical problematic. It insists that "the Realist is justified in his belief only if he can give good reasons for eliminating alternative hypotheses to Realism" (RT, 62). To make this problematic plausible, Devitt adds two qualifications (61-62). First, the sceptics are not requiring absolute certainty; otherwise, the sceptical problematic would be too strict to be an interesting question. Rather sceptics are requiring good reasons to eliminate other alternatives. Second, sceptics themselves are not committing themselves to any belief. They are just exploiting premises behind Realism for reductio ad absurdum. Then, what are counterpossibilities to Realism? Descartes himself offered some in the First Meditation (Descartes 1642); Our senses may totally deceive us; we may dreaming; there may be a Deceitful Demon who makes us believe in the existence of the world. Therefore, according to the sceptical problematic, if we fail to eliminate these counterpossibilities, Realism is not justified.
Devitt distinguishes four different kinds of reply to the challenge. First, there are various kinds of foundationalism (RT, 65-71). Devitt runs through sense-data theory, Lockean representative realism, phenomenalism, and dismisses all of them as failing to answer the sceptics, because none of them can eliminate skeptical counterpossibilities like the Deceitful Demon. Devitt also points out other difficulties for each of foundationalist alternatives. A second group of replies comes from Kant and other constructivists influenced by Kant (71-73). Devitt offers two interpretations of these views, and argues that under the first interpretation Kantianism shares the problems of other foundationalisms, and under the second interpretation its metaphysics becomes too mysterious to believe. A third reply is instantaneous solipsism, which admits nothing other than the belief that the person is now experiencing. This is "a doctrine that is literally incredible" (75), and Devitt illustrates this point with an example of a logician who wrote a letter saying that she was a solipsist (64). A fourth reply is Devitt's own, namely, naturalized epistemology.
Devitt diagnoses the failure of previous replies as coming from their starting point. The sceptical problematic sets the standard so high that we should devise a priori epistemology to answer skepticism. But it is dubious whether a priori epistemology is possible at all, and the failure of other alternatithe above-mentioned attempts suggests that maybe it is impossible. If so, "scepticism is simply uninteresting: it throws the baby out with the bath water" (75). Rather, "we should use the empirical method, not the a priori one, and should set scepticism aside" (RT, 73, emphases in original). The alternative Devitt provides is naturalized epistemology described by Quine (1969). According to Devitt, "naturalized epistemology takes science, and hence its posits pretty much for granted" (RT, 76).
Devitt asserts that scientific theory as it stands is "thoroughly Realist" (RT, 79), at least on observable entities. The language used in science talks not about, say, sense data, but about mind independent objects (74, 79-80). Thus we become Common-Sense Realists just by taking this alternative (though Devitt does not deny the possibility we can reconsider Realism from this position; 76). From this point of view, epistemology is an empirical investigation of our psychological processes. Normative epistemology is also a part of science, because to find a reliable method we should first know the relation between the world and human beings. In short, "we use our view of what is known to arrive at our view of the knowledge process" (79). This is, of course, in accord with his maxim 3.
In short, Devitt dismisses the sceptical problematic because all attempts to answer have failed, and as a result we have no choice other than adopting the naturalistic problematic and being a Realist. In this sense, "Realism is the only theory in town" (73, emphasis in original). He also gives an argument by abduction, especially for Scientific Realism, but this is out of the scope of this paper.
I would like to analyze Devitt's argument along the following lines. First, I evaluate Devitt's argument against the sceptical problematic, i.e., the argument that the sceptical problematic is just uninteresting. Then I compare Devitt's own starting point, "taking science for granted", with Arthur Fine's "natural ontological attitude" (NOA). I show that NOA is sufficient for avoiding the sceptical problematic. I will also argue that, under the assumption of maxim 3, naturalized epistemology cannot lead to Realism.
3.1 Is the sceptical problematic uninteresting?
First of all, we should evaluate the extent to which traditional attempts to answer scepticism have failed. We should recall that Devitt has added a qualification to scepticism; namely, sceptics do not require absolute certainty, but good reasons to believe (RT, 61). This means that it is not enough for sceptics to raise an alternative possibility like Deceitful Demon, but they should also show that these are reasonable doubts. If traditional alternatives fail to eliminate these reasonable doubts, they fail to answer sceptics. The assessment of failure hinges upon the degree of 'reasonableness' we require from sceptics. But curiously enough, in his assessment of traditional alternatives, Devitt never goes through this procedure, and is content with showing just that there are always possible replies from sceptics (e.g. 66, 67).
Even though we admit that all alternatives have failed, Devitt's next move is not warranted either. Devitt claims that Realism is the only theory in town, but this is not the case. To see this, please recall in what sense alternatives have failed. Their major failure has been not being able to answer to sceptics. Now, what Devitt proposes is to set scepticism aside. But this move cancels the failure of traditional alternatives, because they no longer have to answer to sceptics! To put it in a slightly different way: if we accept the sceptical problematic, no viable position (including Realism) is available; if we do not accept the sceptical problematic, almost all positions (including traditional answers) are available. In any case, it is not the case that Realism is the only theory in town. Of course this anarchy is not what Devitt wants. He might reply that because we adopted a naturalistic problematic, other alternatives are eliminated for scientific considerations. But this does not help much. Locke might say that he adopts a Lockean problematic, phenomenalists that they adopted a phenomenalistic problematic, and that from their points of view naturalized epistemology should be eliminated.
How can we avoid this anarchy? I think the answer is in the way Devitt has confined the sceptical problematic. As we saw above, we can play with the degree of 'reasonableness' we require from sceptics. If we admit almost implausible sceptical assumptions, theories should have a high degree of certainty to answer them. The other extreme is to require absolute certainty to the sceptical doubt (I mean, sceptics cannot mention the Deceitful Demon without showing the existence of such a Demon with absolute certainty), and in this case almost all theories are admissible. We can change the required degree of certainty between these two extremes, and apparently if we set the required degree of certainty reasonably high, we can rule out many implausible alternatives. Now, here is the revised sceptical problematic: choose the theory which has the highest degree of certainty (i.e. requires the lowest degree of reasonableness from skeptics) among viable alternatives. Even though I think that naturalized epistemology have a high level of certainty compared with other alternatives, I have no reason to assume that this is the most plausible theory. It is a matter of further analysis. Also please notice that this problematic does not exclude circular justifications, because, for example, if all viable alternatives are circular, we should choose the most certain (circular) argument. Anyway, this problematic is not uninteresting at all. This analysis also shows that just suggesting sceptical possibility does not necessary mean the person is talking about the sceptical problematic in Devitt's original sense. If the person can give a reasonable ground for the sceptical doubt, the possibility is that she is playing my revised version of the sceptical problematic.
3.2 ontological commitment
3.2.1 Fine's NOA
To analyze Devitt's claim that the naturalized epistemology sets our starting point in Realism, let us compare Devitt's argument with Fine's (1986a, 1986b) so called "natural ontological attitude" (NOA).
Fine objects to both realists and antirealists because they are both "inflationists" (Fine 1986b, 166-171). Inflationism is an interpretation of science "in accordance with a set of prior, extra-scientific commitments" (171). There is no ground for such extra-scientific commitments. Realism is an inflationism because of its commitment to the truth of scientific theories. Instrumentalism is inflationism because it introduces the notion of instrumental reliability as a sole criterion of theory choice without having any argument for it (Fine 1986b 166-171).
Then, what if both realists and antirealists add nothing beyond what they can agree upon? This is what Fine calls the "core position" (Fine 1986a, 128). According to Fine, "both realists and antirealists accept the results of scientific investigations as 'true,' on par with more homely truths" (ibid.). More concretely, they agree that "there really are electrons, and that they really carry a unit negative charge and really do have a small mass (of about 9.1*10-28 grams)" (129). What antirealists add to the core position are particular analyses of the concept of truth. Then what do realists add to it? According to Fine, it is "a desk-thumping, foot-stamping shout of 'Reality'" (ibid.). Now, Fine's proposal is to stop adding anything, and take the core position as it is. This is what he calls NOA, the natural ontological attitude (130).
Why does he endorse this position? He call it "homely line" argument (Fine 1986a 126-127). First, we naturally trust our senses. We also have similar confidence in the safeguard systems of scientific investigation. This lead us to trust scientists. So if they tell us that there really are molecules, we must accept that. This means we had better be realists. Fine sums up this argument as follows: "it is possible to accept the evidence of one's senses and to accept, in the same way, the confirmed results of science only for a realist; hence, I should be one." (127, emphasis in original). This acceptance of scientific truth involves the attitude "to take them into one's life as true, with all that implies concerning adjusting one's behavior, practical and theoretical, to accommodate these truths" (ibid.). Therefore, even though Fine use the word 'realism' in this context, his NOA is concerning to adjustment of behavior, not a metaphysical position like Devitt's one.
3.2.2 Comparison of the two views
It is time to examine the ontological commitments of Devitt and Fine. To quote them again, Devitt says that "naturalized epistemology takes science, and hence its posits pretty much for granted" (RT, 76), and Fine says that NOA "accept the results of scientific investigations as 'true,' on par with more homely truths" (Fine 1986a, 128). One important difference between the two attitudes is that while Devitt's commitment make him a Realist, Fine denies that he is commiting to something about the external world. How should we understand this difference?
What Fine made clear was that there are several different ways to take science for granted. Fine's own way is to incorporate science into his life and adjust his behavior. But when a realist like Devitt take science for granted, he requires more. This is obvious from Devitt's argument for abduction (Devitt 1997, 111-113). He claims that a good scientific explanation presupposes Realism. For example, when we explain how a television set works, we need to talk about photons and the photo-electric effect. Replying to Hacking's objection that the reality of photons plays no role in the explanation, Devitt says: "the explanation quantifies over photons.... if the explanation is taken literally and seriously and is right, then there are photons; for the explanation is committed to photons" (Devitt 1997, 112; emphases in original). Therefore we are committed to Realism when we take this explanation literally and seriously. Moreover, Devitt claims, if we do not take the explanation literally and seriously, it does not explain at all (ibid.). From this point of view, Fine's NOA does not take science seriously enough, and by doing so, NOA robs science of its explanatory power.
Devitt's argument presupposes a not-so-implicit assumption: 'taking a sentence literally' means interpreting the sentence by the correspondence of the sentence to the world. But wait a minute. Is this not a semantic claim about the correspondence theory of truth? If so, Devitt is violating his Maxim 3, "Settle the realism issue before any epistemic or semantic issue" (RT, 4). For, as the summary in section 2.3 shows, his key move for the defense of Realism is that taking science for granted entails adopting Realism. Fine tries to break up this relationship, and Devitt's reply to that kind of objection is the above argument for abduction. Thus if it is a semantic argument, Devitt cannot appeal to it if he obeys Maxim 3.
3.2.3 Devitt's argument for ontological commitment
Of course Devitt is aware of this problem. He tries to solve it by revising Quine's famous argument for ontic commitment (Quine 1961; Devitt 1997, 50-53). When is a person ontologically committed to a, or an F, in uttering assertively a sentence S? Devitt proposes two criteria. One is the semantic criterion: "a person is so committed if a, or an F, must exist for S to be true" (Devitt 1997, 51); The other is the non-semantic criterion: "a person is so committed if, in asserting S, he says that a exists or that there exists an F" (ibid.). According to Devitt, the semantic criterion depends on the non-semantic one. For example, the truth condition of the sentence
(1) Lulu is a cat.
is analyzed as follows (ibid.):
(a) (1) is true if and only if there exists something such that 'Lulu' designates it and 'cat' applies to it.
Here, the "something" must exist for the sentence to be true (and this is required by the semantic criterion) only if the use of the word "exists" in the sentence must be understood as an expression of ontological commitment (this is what the non-semantic criterion says). Therefore, for the semantic criterion to work, the non-semantic criterion should also work. According to the non-semantic criterion, sentences like
(2) Cats exist.
simply commit ourselves ontologically to the existence of cats, without any recourse to a semantic theory. But how can a sentence mean something without a semantic theory? Devitt's reply is "if (2) does not establish ontological commitment, then nothing does" (52). For, if we appeal to a metalanguage to give meaning to (2), the metalanguage either has a predicate to allow us to express ontological commitment or not. If it has, there is no reason we cannot use the predicate in our object language. If it does not, then the metalanguage cannot establish ontological commitment for (2) anyway.
Is this argument enough to support Devitt's Realism? I think not. First of all, as Devitt's analysis shows, ontological commitments through sentences like (1) depend on certain specific semantic theories like (a). On the other hand, a vast majority of statements in scientific theories are expressed without the word 'exist' (just think about F=ma or E=mc2 or whatever theoretical statements you like). Therefore to accept these statements does not mean having ontological commitments unless we also accept a semantic theory like (a). Devitt cannot appeal to these statements to defend his Realism.
Nevertheless, we may still use sentences like (2) when we explain the mechanism of a television set. If these sentences are an indispensable part of science, and Devitt's analysis of (2) is right, then maybe Devitt has a point. But I do not think that his argument for the non-semantic criterion is persuasive. As he says, if (2) does not establish ontological commitment, then nothing does. Then the correct answer is maybe "nothing does". For, what is so magical about the word 'exist' that it can somehow connect us to reality without any semantic theory? As Putnam forcefully argues, our intention to use the word in that way cannot do the job where a semantic theory cannot (Putnam 1981, 41-43). Before we can say anything about the relationship between an intention and reality, we need to articulate the intention in a linguistic form, e.g. " I want to use the word 'exist' in such and such way." Of course this sentence does not mean anything without a semantic theory; on the other hand, without such an articulation, intention is too vague to be discussed. Thus, I have no idea how to explain the magical power of the word 'exist', and without such an explanation, we had better give up the idea of establishing ontological commitment by that word. Therefore, following Devitt's lead, nothing establishes ontological commitment.
A conclusion which immediately follows is that Devitt's argument for the defense of Realism summarized in section 2.3 totally fails. Taking science for granted or accepting scientific theories literally does not commit us to any ontological position, because nothing can do that. Of course this conclusion depends on the premise that the magical power of 'exist' is unacceptable without explanation. If we find an explanation, or if we just decide it is acceptable, then Devitt's argument is saved. But still the power of the argument is limited in the sense that many scientific statements do not use the word 'exist'. Moreover, this argument cannot persuade Fine, because Fine can reply that he just does not accept such a magical power (cf. Fine 1986a, 131). Even worse, even if we accept the existence of the magical power, we do not have to accept that the power is in the word 'exist'. For example, why do we not assume that the power is in the sentence "I want to use the word 'exist' here to express ontological commitment"? If the sentence simply commits us to an ontological position without any semantic theory (and I do not think that this is any less plausible than the idea that the sentence 'cats exist' does the trick), the word 'exist' obtains the power to express ontological commitment by virtue of this sentence. What's the difference? The difference is, the sentence "I want to use the word 'exist' here to express ontological commitment" is quite unlikely to occur in any scientific discourse. With this revised version of magical power, we can retain the power to express ontological commitment without automatically committing ourselves to the existence of electrons every time we say "electrons exist".
3.2.4. Can we do without ontological commitment?
The analysis in the previous section suggests that NOA is more plausible than Realism as an interpretation of "taking science for granted". But this conclusion implies that we have no ontological commitment whatsoever (of course we may want to use the revised version of magical power, but this is at least as implausible as Devitt's original version). Is this a viable option? I think so, and Fine's description of NOA provides a way to think about the situation. As I quoted above, NOA involves taking scientific theories into our life as true and adjusting our behavior to accommodate them. By the adjustment of behavior, those who adopt NOA can behave just like a Realist. As far as our everyday life and scientific investigations are concerned, there is no difference between Realism and NOA. In this sense, this option is a viable alternative.
In this paper, I have analyzed two aspects of Devitt's argument for Realism. First, I analyzed his claim that the sceptical problematic is just uninteresting. I proposed a revised version of the sceptical problematic, and argued that this revised problematic is not uninteresting. Second, I analyzed Devitt's argument for ontological commitment by comparing it with Fine's NOA. I found that Devitt's argument hinges upon the magical power of the word 'exist', and that by denying such a power, Fine's NOA becomes a plausible alternative. I also showed that NOA is a viable alternative, though we may have to do without any ontological commitment.
But this does not immediately mean that we should accept NOA. Maybe it is true that when we take science for granted we adopt NOA, but why should we stay there? All Fine shows is that realistic and antirealistic interpretations of scientific discourses are not necessary. This is too weak for "strong NOA", namely believing in no more than the core position upon which realists and antirealists agree (Kukla 1994, 972). But here is one way to make sense of strong NOA. My revised version of the sceptical problematic says that we should choose the most certain alternative among the viable ones. Now, I have argued for the viability of NOA. Since NOA is a weaker position than any other realist or antirealist alternatives, NOA is more certain than any others (as Devitt admits, a weaker position is always easier to defend; RT, 109). Therefore, if we adopt the revised sceptical problematic, we have to prefer NOA over Realism. Hence strong NOA is defended. I think that this combination, NOA and the revised sceptical problematic, provides a good case against realism. 
 Details of these analyses are worth discussing, but because of lack of space I cannot spend time on them. Anyway my major criticisms on Davitt are independent from these details.
 Devitt's arguments from abduction can be found in pp. 73-75, 108-110, 139-142, 147-149 of RT. These arguments depend on his view on ontological commitment, which I criticize below. See section 3.2.2. below.
 Devitt might reply that anti-Realistic traditional alternatives like phenomenalism have other difficulties, so they cannot be saved by this move. Of course phenomenalists have failed to translate physical object language into phenomenal language. But maybe they have other virtues over Realism to compensate for this failure. My point here is that we cannot easily dismiss any alternative without absolute criteria like the sceptical problematic. What we need is detailed comparisons among alternatives.
 Surprisingly enough, Fine uses van Fraassen as a paradigm case of instrumentalism, but of course van Fraassen is not an instrumentalist in the exact meaning of the word. Instrumentalism reinterprets scientific statements as something else, while van Fraassen claims that science should be taken literally. See van Fraassen 1980, 9-13.
 This is a part of his more radical argument that even semantic theories cannot establish such a relationship between words and reality. For my own analyses on Putnam's argument and on Devitt's reply, see Iseda (1995) .
 I am very thankful to Prof. Devitt for his comments on an earlier draft of this paper.
Descartes (1642) Meditations on First Philosophy, translation in E. Anscombe and P. Geach (trans., ed.) Descartes; Philosophical Writings. Prentice Hall (1971).
Devitt, M. (1997) Realism and Truth, second edition with new afterword. Princeton: Princeton University Press.
Fine, A. (1986a) Shaky Game; Einstein, realism and the quantum theory . University of Chicago Press.
-- (1986b) "Unnatural attitudes: Realist and instrumentalist attachment to science" in Mind 95, 149-79.
Iseda, T. (1995) "Is metaphysical realism defeated?" (Japanese) in Kagakutetsugaku (Philosophy of Science) 28, 61-78.
Hacking, I. (1983) Representing and Intervening; Introductory topics in the philosophy of science. Cambridge: Cambridge University Press.
Kukla, A. (1994) 'Scientific realism, scientific practice, and the natural ontological attitude" in British Journal for the Philosophy of Science 45, 955-975.
Putnam, H. (1981) Reason, Truth and History. Cambridge: Cambridge University Press.
Quine, W.V. (1961) " On what there is" in From a Logical Point of View, 1-19. Harvard University Press.
-- (1969) "Epistemology Naturalized" in Ontological Relativity and Other Essays 69-90. New York: Columbia University Press.
Van Fraassen, B.C. (1980) The Scientific Image. Oxford: Clarendon Press.