Return to Human Knowledge and Human Nature contents  

 

12. KNOWLEDGE BY BEST EXPLANATION

 

In this final chapter I shall consider whether the concession of nativism would enable an empiricist to solve the problem of induction, and of justifying non-deductive modes of argument generally.

 

Preliminaries to a Problem

As I noted in the last chapter, if our perceptual beliefs are in fact both true and reliably caused, then they will constitute first-order knowledge of the world. I also argued that the supposition that they are reliably caused by physical objects is overwhelmingly the best explanation of their existence. Now if we could somehow know that inference to the best explanation is a generally reliable process, then we could also know that our perceptual beliefs are largely true. In that case we should have second-order knowledge also. We should be able know that we have knowledge of the physical world, thus avoiding scepticism.

            However, a good deal more is at stake in this chapter than this. For inference to the best explanation is also implicated in the vast superstructure of belief which we erect on the basis of our perceptual beliefs, including many of the beliefs of common-sense, as well as those of science. For example, many of my beliefs about the past presuppose the reliability of memory. I believe that memory is generally reliable, in turn, because the hypothesis that it is so provides the best explanation of the way in which my memories cohere with those of other people, and with other traces of the past, such as photographs and written records. In the same way, many of my beliefs about the future (such as that my house will be where I expect it to be when I return home in the evening) presuppose that I inhabit a generally regular, stable world, in which most events happen in accordance with predictable patterns. I hold this belief, in its turn, because it provides the best available explanation of the regularities of the past. Even my belief that my chair continues to exist when I turn my back on it is justified, if it is, because it provides the best available explanation for the fact that the chair is still there when I turn around again. And, of course, my beliefs concerning the theoretical entities of science, such as electrons, are held because the hypothesis of their existence provides the best explanation of a variety of experimental data.

            All this, too, would be vindicated, if non-deductive modes of argument could be justified. In that case we should not only be able to defeat the sceptic about the physical world. We should also be in a position to defeat the kind of sceptic who, while allowing that we know of the existence of the physical world, and of the objects we are currently perceiving, would deny that we have any knowledge of the past, of the future, of regions of space not currently under our observation, or of such things as electrons, which are too small to be perceived. For in all of these cases our beliefs rely, in one way or another, on inference to the best explanation.

            I am going to suggest that the concession of a certain sort of nativism may enable the empiricist to justify realism about our knowledge of the physical world. However, one obvious problem in attempting to make use of nativism in combatting the sceptic, is that most of the arguments given in support of nativism seem to presuppose that there is a physical world, and that inference to the best explanation is a reliable method for acquiring beliefs about that world. For example, in the final section of chapter 7 we concluded that our concept of best explanation is very likely innate, since it does not appear to be explicitly taught, and since it is hard to see how it could be acquired from experience. In effect, we argued that the best available explanation of our possession of the concept best explanation, is that it is innate. We then immediately face the charge of circularity, if we try to make use of the conclusion of this argument in trying to justify our reliance upon inference to the best explanation. However, I propose to set this problem to one side for the moment. I shall focus initially on the question whether the innateness of our concept of best explanation could in any case help with the problem of justification. I shall return to the charge of circularity in a later section.

            Most of the discussions since Hume, of attempts to justify non-deductive patterns of reasoning, have focussed upon induction - that is, on arguments whose premisses take the form 'Within our observation all As have been Bs' and whose conclusions either have the form 'Therefore all As are Bs' or the form 'Therefore the next A will be a B'. But as we noted in chapter 7, induction itself should properly be understood as an instance of the wider category of inference to the best explanation. Accordingly, in the next section I shall briefly outline the traditional problem of induction, and will explain the main attempts that have been made to solve that problem. In each case I will show how those attempts have analogues in the attempt to justify inference to the best explanation. None of these attempts, as they stand, will prove successful.

 

The Problem of Induction

The first point to notice about inductive arguments is that they are not deductively valid. From the fact that all observed As have, so far, been B, it certainly does not follow that all As are Bs. It is always at least conceivable that the very next A we observe, or indeed all the remaining As, will turn out not to be B. Nor does it follow from our observations that probably all As are Bs, except in those rare cases where the set of As is finite and we have already observed a substantial percentage of them. So we certainly cannot validly deduce that all As are Bs (or even that the next A will probably be B) on the basis of our past observations. For a valid argument is precisely one where it is inconceivable that the premisses should be true while the conclusion is false.

            Suppose we could know that nature is broadly uniform - that the same general laws and principles will obtain in all regions of space and time, and that, as Hume puts it, the future will broadly resemble the past. Then we would know that most regularities in nature are projectible. In which case the insertion of this thesis into inductive arguments, as an extra premiss, would turn them into deductively valid ones. That is, from 'All observed As have been Bs' together with 'Most regularities in nature are projectible' we could validly deduce 'Probably this regularity is projectible - so probably all As are Bs'. The trouble, however, concerns how we are to know that nature is uniform. The only plausible source of our knowledge of this is itself inductive. The only real ground for claiming that nature is mostly regular is that it has mostly been so within our experience in the past. But then the move from 'Nature has mostly been regular in the past' to 'Nature is mostly regular' is an inductive one, in which case our argument will apparently have been moving in circles.

            Inductive arguments are not deductively valid, and the attempt to turn them into deductive arguments through the insertion of an extra premiss appears hopeless. Yet they do seem to stand in need of justification. It certainly is not intuitively obvious that generalising from observed regularities is a reliable way of gaining general knowledge of the world. (And even if it were intuitively obvious, empiricists would still demand a natural explanation of the reliability of such intuitions.) If our task is to know what we know, then plainly we cannot simply take the soundness of induction for granted. On the contrary, if we are to make use of induction in epistemology, we shall have to give reasons for thinking that induction is generally reliable. There have in fact been only three main attempts to justify our reliance upon induction. These are the so-called 'pragmatic', 'analytic' and 'inductive' proposals respectively.¹ Only the last two of these need concern us here, for reasons which I shall now briefly explain.

            The pragmatic proposal for solving the problem of induction consists in some version of the following argument.

 

Nature is either regular or not. If it is regular, then induction will gain general knowledge for us better than any other method. If nature is not regular, then no method will be successful in acquiring general knowledge. So either way, given that we want general knowledge, it is rational for us to employ induction.

 

Notice, however, that such an argument does not permit us to conclude that induction is generally reliable. The claim is not that induction is likely to produce truths, but merely that it is more likely to produce truths than any alternative method we might follow.

            The pragmatic proposal only establishes, at best, that it is reasonable to employ induction, since induction may succeed whereas no other policy will. However, what we need is an argument whose conclusion is that induction (or rather the wider category of inference to the best explanation) may reasonably be regarded as reliable. For what we require is knowledge of its reliability, if we are to use inference to the best explanation to give us knowledge of the physical world. Compare the following: I flip a coin, asking you to call 'Heads' or 'Tails'. I tell you that if you call 'Heads' and guess right I shall give you a prize, whereas if you call 'Tails' and guess right you will get nothing. Plainly you would have good pragmatic reason to call 'Heads', since this is the only strategy which might gain you a reward. But you certainly would not know that the coin will come up Heads, even if it does in fact do so. This is exactly analogous to the proposal that we are justified, on pragmatic grounds, in employing induction.

            The analytic proposal to solve the problem of induction, claims that in order to justify induction we need look no further than our concept of justification itself. On this view, induction forms part of our conception of what justification is, in such a way that 'Induction is justified' is an analytic truth. It is therefore a mistake to try to justify induction in terms of anything else, least of all in terms of deduction. Rather, inductive arguments form a primitive (that is, basic) part of our practice of justifying and seeking justification for beliefs, fully on a par with deductive arguments. Plainly this proposal admits of extension to the wider category of inference to the best explanation, if anything becoming somewhat more plausible in the process.² We need only claim that it is analytic - a necessary component of our concept of justification - that an inference to the best available explanation of some phenomenon is always justified.

            However, the main trouble with this response to the problem of induction is that it is surely possible for us to question the appropriateness of accepted standards of justification. We can allow that inductive practices form part of our concept of justification, and hence allow that inductively grounded beliefs are justified by accepted standards, and yet ask whether those standards are reliable guides to the truth. The deep question is whether inductively grounded beliefs are likely to be true, not whether we call such beliefs 'justified'. In order to see this, imagine a community who employ a counter-induction rule. This entitles them to derive, from a premiss of the form 'All observed As have been Bs', the conclusion 'All the remaining As are not B'. They, too, may respond to the demand that they justify their practice by claiming that the proposition 'Counter-induction is justified' is, for them, analytic. They may claim, with just as much warrant as ourselves, that the practice of counter-induction forms a necessary ingredient in their concept of justification. Yet plainly their practice cannot be justified (in the sense of 'truth-conducive') if ours is. Exactly similar points may be made in connection with the proposed analytic justification of our practice of inferring to the best explanation.

            The inductive proposal to solve the problem of induction argues that, since induction has been remarkably successful in obtaining truths for us in the past, it is, very likely, a generally reliable mode of inference. This argument is, of course, itself inductive, thus immediately giving rise to a charge of circularity. But the proponents of this proposal may reply that the reliability of induction does not actually figure as a premiss of the argument. Rather, induction is here used as a rule, or principle of inference, in the course of arguing for the reliability of such inference. Moreover, we cannot insist that all principles of inference should be transformed into explicit premisses, on pain of a vicious regress.³ Then since the only premiss of the above argument is the past success of induction, there is no formal circularity. In the same way, we may defend the wider category of inference to the best explanation by arguing that the best explanation of our past success in the use of such inference, is that as a mode of argument it is generally reliable. Here, too, inference to the best explanation is used, rather than taken as a premiss.

            The situation may be compared with the use of soundness proofs in deductive logic, where we demonstrate (deductively) that our system of rules can only generate truths from truths. Generally such proofs will employ the very same principles of inference which are in question. For example, consider the rule of &-elimination, which allows us to derive 'A' (or alternatively 'B') from 'A & B'. A proof of the soundness of this rule might proceed as follows.

 

1. If 'A & B' is true, then 'A' is true and 'B' is true. (Definition)

2. Suppose 'A & B' is true.

3. Then 'A' is true and 'B' is true. (From 1 and 2)

4. Then 'A' is true. (From 3)

5. So if 'A & B' is true, then 'A' must be true. (From 2-4)

 

Here the &-elimination rule has itself been used, in the step from line 3 to line 4. Such proofs are generally said to be explicative rather than persuasive, in that they cannot convince someone of the soundness of &-elimination who refuses to employ it at all. Rather, they amount to a use of our principles of reasoning to explain to ourselves how it is that such principles can guarantee truth from truth. In the same way, it may be said, we can explicate the reliability of induction, or of inference to the best explanation, by means of an argument which employs that very principle.

            The main problem for this proposal arises out of our imagined counter-inductive community, mentioned in connection with the analytic proposal above. For notice that we cannot hope to convince the members of this community of the error of their ways, by pointing to their extensive failures in the use of counter-induction in the past. For they will reason counter-inductively, from the premiss that counter-induction has mostly failed in the past, to the conclusion that most future cases of counter-induction will succeed. Similarly, if we imagine a community of people who have the practice of reasoning to the worst explanation of a given phenomenon (either explicitly, or by inverting all of our canons of best explanation - for example, preferring the more complex of two explanations, other things being equal), then they will respond to past failures by concluding that counter-explanation is very probably reliable. For the hypothesis of general reliability provides the worst available explanation of their lack of success in the past. The point is that we do seem to need more than a merely explicative argument in support of induction or inference the best explanation. For otherwise, what reason do I have, beyond mere laziness, for not switching over to the counter-inductive or counter-explanatory practice?

            What emerges, is that attempts to justify either induction or inference to the best explanation have not met with any great degree of success. I shall now consider whether we might fare any better if we were given, as a premiss, the innateness of our concept of best explanation. I shall henceforward address myself to this wider category of non-deductive argument, leaving induction to be taken care of implicitly as a result.

 

The Arguments Re-vamped

One distinctive feature of the nativism defended in chapter 7, is that it places our use of inference to the best explanation firmly within the faculty of reason. For we do seem to possess a conscious concept of best explanation, which serves to guide our evaluations of competing explanations in particular cases. (This contrasts with the view of Hume, who thought that induction resulted from innately given aspects of our faculty of imagination.) What this then means is that inference to the best explanation does not merely happen to belong within our conception of justification, as the analytic proposal maintains. Rather, it is an innately determined constitutive part of the human reasoning faculty itself. It therefore follows that our imagined counter-explanatory community is impossible for us. For while innateness does not immediately imply inevitability - for example, sexual desire is innate, but some may successfully train themselves not to feel it - inferring to the best explanation is surely too basic a principle of our cognition to be alterable by conscious choice or training. Indeed, many of Hume's reasons for the psychological inevitability of inductive reasoning carry over to the case if inference to the best explanation, now construed as part of our faculty of reason.

            However, we still face the main objection raised against the analytic proposal, namely that it cannot settle the question of the connection between accepted standards of justification, and truth. For this, we need to turn to consider the wider analogue of the inductive proposal (concerning the justification of inference to the best explanation). Here, too, the claimed innateness of our concept of best explanation may be of real help. For if the principle of inferring to the best explanation forms a constitutive part of our cognitive apparatus, then of course we cannot but rely upon it in forming our beliefs, whether at the first-order level, or when we come to do epistemology. If inferring to the best explanation is part of what reasoning is, for us, then the fact that it can only be supported by itself need give no particular cause for concern, or for scepticism. Just as our best use of deductive reason supports its own soundness, so too our best use of non-deductive reason supports its own general reliability. For the best explanation of our success in reasoning to the best explanation in the past, is that such inference is generally reliable.⁴ (To see the extent of our past success, reflect again on the way in which inference to the best explanation is implicated in almost every aspect of our practical lives, as well as on its use in science.) Moreover, we cannot now object that counter-explanatory communities could equally well use inference to the worst explanation to support the general reliability of their practice. For since such communities are not genuinely possible for us, the mere fact that they are imaginable need not undermine the justification of our own procedure. At the very least, all sense of arbitrariness in the thought that I have not switched over to a counter-explanatory practice is removed.

            It appears that an empiricist who can be brought to endorse an appropriate form of nativism, concerning our concept of best explanation, may make considerable headway with the problem of justifying non-deductive modes of inference. In particular, such an empiricist may advance versions of the analytic and inductive proposals which are much stronger than their ordinary - non-nativistic - counterparts. Inference to the best explanation can in consequence be seen to form a constitutive part of human reason, and then the best use of that reason leads to the conclusion that it itself is generally reliable. However, the innateness hypothesis also gives rise to an argument for the reliability of inference to the best explanation which is largely independent of these, as I shall now explain.⁵

 

A New Argument from Nativism

My claim is that the only plausible evolutionary explanation, of why inference to the best explanation should be an innately determined aspect of human reason, is that it is generally reliable. To this it might possibly be objected that an innate concept of best explanation, while having no survival-value in itself, could be a by-product of something which does have value for survival. But this suggestion is too nebulous to take seriously. Until some specific proposal is put forward, concerning what this other innate property might be, it will be more reasonable to believe that our concept of best explanation has survival-value in its own right.

            However, a real attack can be mounted on the supposed reliability of at least one of the strands in our concept of best explanation, namely that of simplicity. For it appears easy to explain how a faculty of reason which chooses between theories on grounds of simplicity (other things being equal) could have a value in survival which is unrelated to truth. For theories which are simpler will be easier to operate with and think in terms of, hence making cognitive processing more efficient. I have a number of points to make in response to this suggestion. The first is that, even if one strand in the concept of best explanation bears no relation to truth, this does not begin to undermine the reliability of inference to the best explanation as a whole. Indeed (and this brings me to my second point), it is hard to see what the value of efficient cognitive processing could be, unless it is that we are more likely to be successful in deriving truths from simple theories than from complex ones. But then this value will only manifest itself, in general, when the theory in question is itself true, or at least close to the truth. While it can indeed happen that true conclusions are derived from wholly false premisses, this will only occur by accident.

            But the point I most want to insist on is this. While it is indeed the case that one strand in our notion of simplicity is a matter of notational elegance, or economy of expression, there is another strand in the notion which is arguably more important, and which is best subsumed within the notion of explanatory power. This is the sort of structural simplicity which is at issue when we are told not to postulate entities beyond necessity, and to prefer explanations which employ the fewest number of explanatory factors. For example, if my house has been burgled then I should conclude, other things being equal (that is, in the absence of two or more sets of footprints or fingerprints, or of evidence that burglars most often work in pairs), that the crime has been the work of a single individual. For the hypothesis of a single burglar is simpler than the hypothesis of two, or three, or four. But the virtue of this sort of simplicity is not a matter of ease of expression. It is, rather, that the simpler theory at the same time explains the absence of any evidence of more than one burglar, thus having greater explanatory power than its rivals. And it is hard to see how the survival-value of explanatory power can similarly be explained away in terms unrelated to truth.

            It can further be objected, that while evolution might have been reliable in selecting a concept of best explanation which would generate truths concerning matters of immediate significance for the survival of primitive people, this gives us no reason to think that it will be generally reliable, particularly with regard to theoretical science. One response, is to deny that there is any principled distinction between common-sense and science. For example, suppose it is said to be distinctive of scientific theories that they deal with unobservable entities, such as electrons. It may then be proposed that evolution gives us no reason to trust inference to the best explanation in connection with unobservables. But we only have to reflect to see that this will not do. For the past is just as unobservable as an electron. Yet to be able to reason that the best explanation of fresh paw-marks in the mud is that a tiger has been here very recently, may be crucial in ensuring our survival.

            Another point is that our concept of best explanation had to be fit to serve us in almost any environment, from the equator to the poles. For of course the distinctive fact about human beings as a species, as we noted in discussing the innateness of folk-psychology in chapter 8, is that we are uniquely adaptable, having a distribution over the globe which no other species has. It is hard to see how any concept could subserve this purpose except one which would be generally reliable. For concepts which are reliable only in connection with phenomena on Earth, for example, can be ruled out on the grounds that they might lead one to predict that the sun will not return next day, and hence to fail to make crucial provision for the morrow.

            Of course it must be possible, in principle, to cobble together some sort of artificial concept which would embrace only those aspects of nature having significance for pre-technological survival. And it might then be suggested that evolution could have selected a concept of best explanation which would be reliable in those circumstances only. But to this we can make two separate replies. The first appeals to yet another form of innateness, namely our innate ability to detect genuine resemblances between things. It is plain that we do have such an ability, and that it has survival-value in its own right.⁶ In which case it is unlikely that pre-technological and scientific phenomena should strike us as so similar (in all those respects which matter for explanation and prediction), when really they are not. The second reply is that we have every reason to think that evolution has delivered us a concept of best explanation whose reliability is not limited in the manner suggested. For the best explanation, both of our success in making sense of the world in the past, and of our continuing scientific success, is that our extended use of inference to the best explanation continues to be reliable. (Here the argument merges with the analogue of the inductive proposal, outlined above.)

            We can reply in the same sort of way to a rather different objection. This would be that the survival-value, and hence reliability, of our concept of best explanation in the past provides no guarantee of continued survival-value, or of reliability, in the future. For in fact, the best explanation for the observed regularity of nature in the past, is that nature is governed by laws which operate independently of time and place. Moreover, the best explanation of our success in making sense of the world in the past, is that we have an innate ability to detect those properties which figure in projectible laws. So we have reason to think that those properties of the world which conferred survival-value on our use of inference to the best explanation in the past, will continue to do so. (While there is an obvious circularity in this reply, it is in fact not vicious, but a virtuous coherence, for reasons which will emerge in the next section.)

            A final objection to the argument from the innateness of our concept of best explanation to the conclusion of reliability, is that such a concept only enables us to make a choice from amongst the various theories which have actually been proposed. So to say that it is reliable is just to say that, of the theories suggested, it tends to select the one that is most likely to be true. But then the use of such a concept will only generally lead to the truth, if the correct theory is amongst those antecedantly given. So even if the concept of best explanation is generally reliable, this gives us no reason to think that inference to the best explanation is generally reliable. But the reply to this is easy. It is that we surely have to suppose that our faculty for generating hypotheses would evolve alongside our concept of best explanation. For what would be the survival-value of having a concept of best explanation which is reliable in the above sense, if we did not also tend to generate ranges of hypotheses which include the correct one?

            I conclude that it is very hard indeed to see what possible advantage possession of a concept of best explanation could confer in survival, unless such a concept were generally reliable in obtaining for us truths, around which we could then construct our plans and projects. In which case, not only does the innateness of inference to the best explanation mean that such inference forms a constitutive part of any possible human concept of justification, but it also gives us two reasons for thinking that there is a connection between our standards of justification and truth. (These reasons are, namely, that the best explanation of our past success in the use of best explanation is general reliability, and also that general reliability provides the best explanation of the innateness of our concept of best explanation.) There remains, however, the charge of circularity outlined at the outset of our discussion. For the argument for nativism is itself an inference to the best explanation. Indeed, it is one which presupposes that we have knowledge of a world of physical objects, since we could otherwise hardly be entitled to appeal to evolutionary theory. I shall now turn to the task of rebutting - or at least defusing - this charge.

 

A Modest Coherentism

Undoubtedly the most important role for the claimed innateness of our concept of best explanation, is that it forces on us at least a modest form of coherentism. For recall from chapter 7, that what may make one explanation better than another is, in part, its greater internal consistency and its higher degree of coherence with other received beliefs and theories. In accepting that inference to the best explanation is an innately determined constitutive aspect of human reason, we are therefore accepting that the justification for our beliefs must consist partly in their overall coherence. This may then enable us genuinely to 'boot-strap' our way, not only to a justified belief in the reliability of inference to the best explanation, but also to a justified belief in the physical world as the best available explanation of our experience.

            It is true that there is a kind of circle involved in using inference to the best explanation to argue for the innateness of our concept of best explanation, which is then used in turn in arguing that inference to the best explanation is generally reliable. But this circularity is not a vicious one. Rather, it is just what one would expect, given that the relation of justification consists, in part, in coherence within and between neigbouring theories. The best use of our reason leads us to believe that inference to the best explanation is a constitutive part of our reason, and hence that an aspect of what a justification for a belief must consist in, for us, is a matter of coherence. The best use of our reason then leads us to believe that our reason itself is generally reliable, and that the best explanation of our perceptual beliefs is that perception is a generally reliable guide to the states of the physical world. These beliefs and explanations interlock in a highly satisfactory way, enabling us to say that they are amply justified - indeed, that they constitute genuine knowledge.

            While we have accepted a coherentist conception of justification, this is still a relatively weak form of coherentism. (Alternatively, it may be considered a weakened form of foundationalism.) For we do not have to allow that all coherent networks of true belief constitute knowledge, even provided that there exists no other equally coherent network concerning the same subject-matter. Rather, we can insist that explanatory theories need to be tied down in perception in order to count as justified. So the example given in chapter 5, of a coherent network of beliefs about a character who is in fact fictional, will not count as justified on this view. What is given to us - what must form, as it were, the foundation of our coherent system of belief, if it is to be justified - is that there seems to us to be a world of physical objects in three-dimensional space. These seemings of the world constitute the data for which our explanatory theories are constrained to account.

            It is not just in accepting a coherentist conception of justification that empiricists should now move away from classic foundationalism. They should also allow an element of coherence to infect those very seemings of physical reality which constitute the foundation for the epistemological enterprise (that is, to know what we know). For when it is pointed out that even such seemings presuppose a grasp of their constitutive concepts (that it seems to you that you are sitting on a chair presupposes that you have a grasp of the concept chair), an empiricist need no longer reply (as I did in the last chapter) that grasp of the constitutive concepts must be presupposed in any enquiry. Rather, we can allow that sceptical doubts about our own conceptual abilities are possible, but reply that the best available explanation for the fact that I seem able to make regular judgements and classifications concerning what seems to me to be a largely regular world, is that I do indeed possess the ability to classify things in a regular manner.⁷ So although my seemings of physical reality constitute a foundation in the theory of knowledge, in so far as they are the basic data which need to be explained, we do not have to claim that they are absolutely certain, or that they are wholly independent of any other beliefs.

            We can reply in a similar manner to the sceptical suggestion that the demon may be making me go wrong in my use of inference to the best explanation, leading me mistakenly to prefer the realist hypothesis over the hypothesis of the demon. For in fact the best explanation of the intelligibility and coherence of my current appraisals of inferences to the best explanation, is that such appraisals are generally correct. And again the apparent circularity here is not vicious, but is rather one of virtuous coherence.

            Indeed, it may be worth noting that the present line of defence of realism about knowledge may also extend to realism about truth, which we discussed briefly in chapter 1, and have since been taking for granted. For the best theory of my abilities to identify and classify things (that is, of the abilities which underpin the representative powers of my thought), is that they are realised in categorical bases in the brain, which operate in accordance with laws of nature which are independent of space and time. So there are determinate truths about how those capacities would respond in remote regions of space and time which are in fact inaccessible to me, which seems to be enough for my present thoughts about those places and times to be determinately true or false. Or so, at any rate, I believe.⁸

            I conclude that we are sufficiently justified in believing inference to the best explanation to be a generally reliable method of forming beliefs. This is so, because the best explanation of our possession of a concept of best explanation is that the concept is innate, and because the best explanation of its innateness is that it is generally reliable. It is also because the best explanation of our past success in the use of inference to the best explanation, is that it is generally reliable, and because the best explanation of past regularities in nature, is that nature is law-governed in such a way that past regularities will continue. And it is because the innateness of inference to the best explanation forces us to see that a justification for a belief may consist in the fact that it hangs together with other beliefs in just this sort of interlocking way. We may therefore conclude that we do indeed know that we have knowledge of the physical world, since the hypothesis of reliable perception provides easily the best available explanation of our perceptual beliefs.

 

An Argument for Open Minds

It is important to be clear about just what the argument above has shown - in particular, that it provides us with no basis for claiming to have defeated someone who is already a convinced sceptic. For even if inference to the best explanation is unavoidable by virtue of its innateness, this does not mean that we necessarily have to believe that the conclusions of such inferences are likely to be true. For example, the analogue of Hume's position on induction remains possible. Someone may concede that inference to the best explanation is unavoidable for practical purposes, while denying that such inference can be justified. And while they may be prepared to act as if the conclusions of such arguments were true - again for practical purposes - they may refuse to allow that they are likely to be true.⁹ Such a person will happily follow the whole argument of this and the preceeding chapter, accepting the conclusions for practical purposes only. So their sceptical position is just as intact at the end as it was at the beginning.

            While my argument cannot rationally convince the convinced sceptic, what I do claim is that it should convince all who approach the issue with an open mind. All that is required, is that my reader should be prepared to use an inference to the best explanation and believe, at least tentatively, that the conclusion is true. This need not beg any questions in favour of the reliability of inference to the best explanation, since at this stage it can be left as an open question whether or not inference to the best explanation is really reliable. The reader, like myself, may approach the matter perfectly prepared to be convinced that inference to the best explanation is not reliable, or to conclude that no verdict can be delivered either way. But in fact, as it turns out, the best use of our faculty of inference to the best explanation leads us to the conclusion that such inference is generally reliable.

            That the sceptic cannot be defeated does not mean that the sceptic wins. It should come as no surprise that there are epistemological standpoints which are immune to rational persuasion. Consider, for example, someone who is a convinced paranoid. Of course nothing that we say or do can rationally require them to give up their belief that all are conspiring against them. For anything that we say can immediately be incorporated into the paranoid hypothesis, as being just one more twist in the plot. Yet this need not mean - plainly - that we are unjustified in believing that there is no plot. What we have to recognise, is that the task of epistemology is more modest that we might initially have thought. The task cannot be to produce reasons for belief which will be acceptable to anyone, no matter what their starting point. It can only be to produce reasons acceptable to those who have not yet made up their minds. So while I cannot claim to have defeated the sceptic, I can claim to have sufficiently justified an opposing realism about our knowledge of the physical world.

 

On to Conclusion