What Finite Minds Can Know: Epistemic Limitations and the Existence of Infinite Mathematical Objects Abstract The classical British empiricists were finitists. They argued that we lack an adequate idea of any infinite object on the grounds that we cannot perceive an infinite object. The empiricist approach lingers in contemporary constructivist philosophy of mathematics, such as the views of A.W. Moore (1989), (1990), and (2007). Mainstream contemporary mathematics is undeniably infinitist: one needs infinite sets to do real analysis and set theory. This suggests that the main philosophical motivations for finitism are flawed. We reconstruct what we take to be the main argument for finitism: the argument from epistemic limitations (‘EPLIM’). We identify several key epistemological assumptions—verificationism, idealism, and the finitude of perceptual experience—needed to make the argument valid. We indicate why we hold that all these assumptions are false. We consider in detail some arguments against idealism about mathematical objects. We conclude by developing a realist response to EPLIM and showing how epistemic limitations are fully compatible with the reality of infinite mathematical objects. 1. Introduction Several branches of mathematics—such as number theory, geometry, topology, real analysis, and set theory—express truths that can only be satisfied by infinite domains. From a realist perspective, mathematical knowledge thereby requires a vast ontology. How can this vastness be reconciled with the apparently limited epistemic capacities of human agents? The human mind has finite computational and perceptual resources. The infinite objects of mathematics threaten to outstrip the mind’s ability to know them. This problem, dubbed ‘the problem of remoteness’ by Lavine [1994], is essentially an epistemological problem.1 Like most epistemological problems, the problem is the result of bad epistemology, an artefact of the way the problem is formulated. Or so we shall argue. Mathematical knowledge provides a special test case for empiricism. Roughly speaking, empiricists hold that the source and basis of all meaningful concepts is sensory experience. Mathematics appears to provide knowledge of infinite objects (or at least of the concept of infinite objects). The concept of infinite object does not seem to be derivable purely on the basis of sensory experience. All that sensory experience presents is a finite object (a finite line segment, a finite collection of objects) together with the open ended possibility of repeating operations (such as dividing and counting) indefinitely. Thus the limits of our experience together with the infinite character of some mathematical knowledge put severe pressure on empiricist epistemology. 2 Moore [1990] suggests that the existence of the infinite is ultimately empiricism’s undoing: There can be no doubt that empiricism was one of the great philosophical movements, of deep and lasting significance: but this is partly because of lessons we can learn from its ultimate failure. It is a very important feature of the infinite that it helped signal that failure (Moore [1990]: 83). Before we can evaluate Moore’s insightful comment, some terminological preliminaries are in order. We can distinguish between three general varieties of finitism: metaphysical, semantic, and epistemological.2 The tripartite division between kinds of finitism is as follows: (i) Metaphysical finitism—the belief that there are no infinite objects. (ia) Mathematical finitism—the belief that there are no infinite sets or infinite numbers. (ib) Physical finitism—the belief that the universe (or space, time, is finite in size, or contains finitely many members. (ii) Semantic finitism—the belief that talk about the infinite is non-sense. (iii) Epistemological finitism—the belief that only finitary reasoning is epistemically secure; reasoning involving an infinite number of steps or the use of quantifiers ranging over infinite domains is suspect. With regard to (ia) mathematical finitism, we can further distinguish between strict and liberal varieties. Strict mathematical finitists believe that there are upper finite bounds to numbers and other mathematical entities.3 Liberal mathematical finitists think that there may be a potential infinity of numbers (or other mathematical entities), but deny that there is an actual completed infinity of numbers (or other mathematical entities). In mainstream, classical contemporary mathematics, strict mathematical finitism is assumed to be false. Infinitism has won the day, as indicated by the widespread assumption of the Axiom of Infinity.4 Moore’s comment suggests an argument from the truth of infinitism to the falsity of empiricism. This inference raises several philosophical questions: (iv) Does empiricism compel finitism? (v) Are there any good reasons to be finitist? How about epistemological considerations? (vi) Does the practice of classical mathematics provide an argument against finitism? Old fashioned empiricism is not a live option. However, several positions in the contemporary philosophy of mathematics (e.g Dummett [1966], Moore [1989], Wright [1982], [1985]) share empiricism’s view that mathematical knowledge must be grounded in the appropriate experience of mathematical objects. Moore [1989] argues that we cannot make sense of the idea of experiencing an infinite object and therefore that statements about actually infinite objects are in fact unintelligible. However, Moore allows that we can make sense of statements about potential infinites. Moore thus argues for liberal semantic finitism. The classic example of a liberal finitist position is Gauss’s interpretation of the limit concept. Gauss stated: I protest against the use of infinite magnitude as something completed, which in mathematics is never permissible. Infinity is merely a facon de parler, the real meaning being a limit which 3 certain ratios approach indefinitely near, while others are permitted to increase without restriction (Gauss [1831]:216) Liberal finitism appears to present the best of all worlds. On the one hand, it allows for the use of the infinite in the calculus and elsewhere. On the other hand, the actual infinite—which is epistemologically suspect—is banned from mathematics. However, even liberal finitism is in tension with the mathematical practice of quantifying over infinite sets. Using Quinean methodology, such quantification carries an ontological commitment to completed, infinite objects.5 We argue that finitism (both strict and liberal mathematical varieties) is not justified by traditional epistemological considerations. Section §2 of this essay offers a reconstruction of the argument from human epistemic limitations used by finitists. (This argument is referred to as ‘EPLIM’). Section 2.1 shows that Hume and Berkeley embraced a version of the argument from epistemic limitations. Section 2.2 excavates some of the bridging principles needed to make the argument valid. Section 2.3 attempts to refute one such bridging principle, the alleged datum (D) that human beings cannot encounter an infinite object. Section §3 argues against bridging principle (I) which depends on a consequence of idealism. Section §4 presents the realist response to the EPLIM argument. We conclude in section §5 that the vastness of mathematical ontology coheres well with due acknowledgement of the mind’s epistemic limitations. To be sure, there may be better arguments for finitism than the argument from epistemic limitations presented here. Nonetheless, as we show, the argument has been tremendously influential. Finitists from Hume to such ‘Wittgensteinian’ finitists as A.W. Moore have embraced some version of EPLIM. If we are right, then the single most influential philosophical path to finitism has been closed off. Infinitists can have their cake and eat it too: they can maintain the existence of the actual infinite in mathematics and respect human epistemic limitations. 2. The argument from epistemic limitations. 2.1 Reconstruction of the argument What is striking about the classical empiricists’s arguments for finitism is the way they move from consideration of the mind’s capacities to a conclusion about the nature and existence of mathematical objects. This kind of inference (from mind to world) lends the argument an idealist character. The idealism is explicit in Berkeley’s argument for finitism: Every particular finite extension which may possibly be an object of our thought is an idea existing only in the mind, and consequently each part thereof must be perceived. If therefore, I cannot perceive innumerable parts in any finite extension that I consider, it is certain that they are not contained in it… (Berkeley [1710], section 124: 77-8). Berkeley’s conclusion is that any finite extension does not, indeed cannot, contain infinitely many parts. Berkeley is a metaphysical finitist and atomist. Only extreme idealism (such as Berkeley’s) would legitimate the leap from the fact of our inability to perceive an infinity of parts to the conclusion that such parts do not exist in spatial extension. 4 Like Berkeley’s argument, the main thread of Hume’s argument for finitism moves from an epistemological premise to a metaphysical conclusion. Hume’s epistemological premise is that the human mind has finite powers of discrimination and can only form an idea of an interval with a finite number of parts. Hume’s metaphysical conclusion is that space and time themselves are only made up of a finite number of parts (‘minimal perceptibles’). As commonly noted, the inference from epistemological premise to metaphysical conclusion appears to depend on a kind of phenomenalism. Phenomenalism is itself a species of idealism.6 Only if space and time themselves were the product of human sense-impressions would it be legitimate to draw inferences about them based on the character of human sense-impressions and ideas.7 Idealists hold that the nature of objects (space, time, and reality itself) is itself constrained by the interpretative activity of the human mind. This reversal of priorities is the point of the ‘Copernican revolution’ in philosophy that Kant introduced, whereby ‘we try the hypothesis that objects conform to the structure of our knowledge’ (Kant ([1781]/[1911]:Bxvi-xvii, 22). Consequently, idealists should expect that they can infer the structure and properties of objects from their knowledge of the structure of human concepts.8 Idealism interacts strongly with the belief that the human mind has finite capacities, with striking repercussions for the philosophy of mathematics. Suppose that one takes the finitude and limitations of the human mind very seriously. Like a computer, the human mind has a limited ‘storage capacity’, a limited time during which it can function and apply operations, and can only take in and process a limited amount of ‘data’. If the objects of mathematics are, in some sense, (as idealists hold) constructions of the human mind, then these artefacts will necessarily conform to the limitations of that mind.9 It should be obvious where this line of thought goes: a finite mind can produce (it is said) at most a finite output, and so the objects of mathematics—the ‘output’—must be finite (and finitely many) as well.10 We can generalize the argument and arrive at (the schema of) an argument for mathematical finitism called ‘the argument from epistemic limitations’ (‘EPLIM’ for short): (1) Human minds, being finite, are limited in their capacity to comprehend and represent objects to representing finite objects. (2) The objects of mathematics must be comprehensible and knowable (by the human mind). (3) Therefore, there are no bona fide infinite mathematical objects. The conclusion (3) is a statement of metaphysical finitism, since it concerns mathematical objects. Premise (1) is a statement of a certain kind of epistemic finitism, as it concerns the human mind and its epistemic capacities. Premise (2) is intended to serve as a bridge between the two positions. Now, as it stands, the argument is incomplete. Supplementary premises are needed to derive the metaphysical conclusion from the epistemological premises. Neither premise (1) nor premise (2) is innocuous. Both premises stand in need of disambiguation and could be false. In particular, we claim that (1) is false if interpreted properly.11 More surprisingly, (2) is highly contentious, and from a realist 5 perspective, false. As we will show, mathematical realists should tolerate the possibility of a certain amount of ignorance of the properties of mathematical objects as a consequence of their position. To be sure, few philosophers nowadays want to be strict finitists as recommended by (3). Most would prefer to be liberal finitists. EPLIM can be recast as an argument for liberal finitism. Premise (1) of the original argument can be relaxed to allow that the human mind can comprehend and represent potential infinities. The revised argument then runs: (EPLIM-revised) (1-R) The human mind can comprehend and represent only potential infinities, not actual infinities. (2-R) The objects of mathematics must be comprehensible and knowable by the human mind. (3-R) The objects of mathematics are at most potentially infinite, not actually infinite. The revised argument, with its weaker conclusion, is still suspect. Like the original version of the argument, the revised argument tries to derive a metaphysical conclusion from epistemological premises. 12 From a realist perspective, such a move is illegitimate. Unless there are a priori reasons to think that objects must reflect the structure and limitations of the minds that represent them, there can be no reason to conclude anything specific about the nature of objects from our possibly inadequate conceptions of them. It’s pretty clear that as it stands EPLIM is incomplete. It does not reflect the pivotal assumptions of idealism and epistemic finitude central to the empiricists’ arguments. The conclusion (3) simply does not follow directly from the premises without the addition of a couple of bridge principles. Many empiricists would carry these additional principles as part of their baggage and so would tend to read them into EPLIM anyway. However, we argue that EPLIM does not survive scrutiny of these bridge principles. Once these assumptions are made explicit, realists can argue that we have good reason to reject them (and the conclusion of EPLIM). 2.2 The Bridging Principles (V), (D), and (I) One bridge principle must tie concepts to sense-impressions or experience. After all it is the fact (if it is one) that we cannot experience there being something infinite that leads empiricists to say that we lack a positive conception of the infinite.13 However, the lack of an adequate concept of X does not license the inference that X does not exist. To turn the foregoing observations into an argument against the actual infinite, we need to conclude that our lack of an adequate concept of the infinite somehow translates into a lack of ontological reality. Such a move is verificationist. Here we agree with the suggestion of Benardete (1966) that empiricist and idealist objections to the actual infinite are based on a dubious verificationism.14 So our first bridge principle will be a version of the verification principle of meaning (‘verificationism’ for short): (V) A concept is meaningless and illegitimate unless it is a concept of an object of possible experience. 6 In any case, the addition of (V) does not suffice to enable us to draw a conclusion about mathematical objects based on premises about concepts in mathematics. Another bridge principle is needed, linking concepts and objects in such a way that concepts a priori impose their structure (and constraints) on the corresponding objects. The simplest way to guarantee this feat is by assuming a consequence of idealism: (I) What concepts we have determine what mathematical objects there are. To derive the conclusion, we must add the bridge principles and the alleged datum that (D) We never experience anything actually infinite. The additional premises (D), (I), and (V) function as ways of sharpening up the original premises of EPLIM so as to yield the finitist conclusion. (D), (I), and (V) are all highly contentious. However, they are all necessary to derive the ontologically finitist conclusion. For example, if we accept only (V) and (D), then all that follows is that we have no legitimate concept of the (actual) infinite (in mathematics). It does not follow that there can be no infinite mathematical objects. Acceptance of (V) and (I) with denial of (D), however, also suffices to scupper the conclusion. If one thought that, pace (D), one could have an experience of an infinite object, then it would follow (even given (V)) that one could have a legitimate concept of an infinite object and (given (I)) that such an infinite object exists. Of course denial of (D) is out of keeping with the original spirit of EPLIM. In particular, the denial of (D) contradicts premise (1) of the original EPLIM argument. In full detail, with the additional premises inserted, the final revised version of the argument runs: (1*) Our legitimate concepts of (basic) mathematical objects must be derived from our experience, or at least possible experience. [Verificationism (V)] (2*) We have no possible experience of anything truly infinite. [Alleged datum (D)] (3*) We can have no legitimate concept of an infinite mathematical object. [from 1,2] (4*) Our concepts determine what mathematical objects there are. [Idealism (I)] (5*) Therefore, there can be no legitimate infinite mathematical objects. [from 4,5] a. The argument as it now stands is valid. Premises (1*), (2*) and (4*) – Verificationism, the Datum, and Idealism, respectively— are all highly contentious, and ultimately, we think, should be rejected as false. As these premises form the crucial bridging principles needed to derive the conclusion, we can conclude that the argument fails. In this paper, we shall focus our discussion on reasons to reject (D) and (I). We think (V) is not really a live option.15 We cannot refute verificationism here. However, a few brief remarks 7 may remind readers why it is objectionable. The verification principle of meaning required for any meaningful statement S, that we have would have to know, in principle, how we could verify S. Means of verification might include observation as well as mathematical proof. The verification principle condemns great swathes of mathematics to the realm of meaningless. For example, prior to its proof, Fermat’s Last Theorem was, by verificationist lights, not meaningful. Although we knew in general what was needed to prove Fermat’s Last Theorem and we apparently understood the theorem, we do not know the specific proof. Indeed, we did not even know that it could be proven. It seems far more likely that the verification principle of meaning is too acidic rather than that mathematicians do not possess the understanding that they take themselves to have.16 2.3 Rejecting Datum (D): That we cannot experience the infinite Let us now consider datum (D) (premise (2*))— that we cannot and could not in principle experience anything infinite.17 Although prima facie plausible, (D) is in fact not conclusively indicated by any of the considerations typically put forward in favour of it. The locus for discussion of (D) has been the literature on the possibility of completing ‘supertasks’ (infinite tasks).18 It is logically possible for someone to sub-divide a line according to the series 1+1/2+1/4+1/8 +…etc by performing the subdivisions at increasingly faster rates, taking one minute for the first step, a half for the second, a quarter for the third, and so on. Perhaps the laws of physics would have to be different in order for someone to accomplish this feat, but there is no logical incoherence in supposing this possibility to obtain. Moore (1989) argues in a Wittgensteinian vein that although the idea of completion of the supertask is not logically incoherent, we do not really understand what it means to say that a supertask has been completed. Moore’s reasoning is one of the best articulations of the case for finitism based on epistemological (and semantic) considerations. We therefore examine and criticize his argument in depth. Here is a representative passage from Moore’s argument: But might there not just be, say, infinitely many stars—or some (natural) process whereby a particle, as a result of oscillating through successively shorter distances, managed to complete infinitely many oscillations in a finite time? Perhaps. But again, this has to be understood in terms of the infinite possibilities that might be written into what we could encounter in experience. Let us suppose that space and time themselves provide endless possibilities of movement and reorientation. Then we can imagine empirical evidence to suggest that, if we were to travel farther and farther away from some point, or if we were to delve deeper and deeper into some small region of space, we should always be able to find specifiable phenomena (stars, the traces of the particle’s movement, or whatever). But this is not to envisage an ‘infinite reality’. 8 It may seem obscure what more could possibly be required. But more is being presupposed when someone talks of performing infinite many tasks , or more specifically, surveying all the natural numbers, in a finite time; they are sanctioning the possibility of a kind of direct encounter with the infinite. We have yet to be presented with a situation, real or imaginary, where it would be appropriate to talk in these terms….It is an abuse of grammar to describe anything as being, essentially the outcome of an infinite process, as it is to describe any process as being infinitely old [italics ours]. (Moore, [1989], in D. Jacquette, ed. [2002]: 317). Moore’s case for finitism consists of two steps. First, Moore suggests that we do not understand the claim, because we cannot in principle have the requisite experience corresponding to the claim. Second, Moore tacitly takes demonstration of a lack of understanding of a claim to suffice for showing the claim to lack truth. We suggest that both of these steps are unwarranted. With regard to the first step, why should the fact that we cannot clearly delineate a situation, an experience that would count as ‘a direct encounter of the infinite’ count against our understanding statements involving infinity, unless we hold to a verificationist-empiricist of meaning? The position of the classical rationalists in the debate over understanding infinity was that the idea of infinity is not an idea of experience or imagination, but an idea of reason, intellect, and understanding.19 Such ideas do not need to have a complete (or accessible) empirical instantiation in order to be understood. Indeed, we want to suggest that it is a mistake to understand knowledge of a mathematical truth (such as ‘The set of natural numbers N is infinite’) on the model of perceptual acquaintance with an object. Rigid insistence on the model of acquaintance is one of the empiricist’s great mistakes (and appears again in Moore (1989)). Moore alleges that we cannot experience an infinite array of stars. But is that claim true? We can experience a portion of an infinite array of stars. This portion is compatible with the array being finite or infinite. Perceptual experience itself cannot present an array as infinite, merely as finite and not exhausted. Moore argues that we do not understand statements involving actual infinities on the grounds that we lack any corresponding (perceptual) experience. This amounts to verificationism: the claims in question cannot be understood if we cannot have possible experiences that would verify them. Moore might be right that we cannot have an experience of an actual infinite, but this point is irrelevant. Only on a crude empiricist-verificationist approach must we be able to have an experience to ratify the meaningfulness of every statement. The meaningfulness (and our capacity to understand) statements about actual infinities might be ratified merely by the coherence of talk about such infinities in ZFC set theory, for example. No sensory experience is required; this part of mathematics is a priori discipline. Nonetheless, there might be experiences of finite, but indefinitely large arrays, for example, that are relevant to explaining how we acquire the concept of infinity. However, on our view experience does not directly bequeath meaning to infinitary mathematical statements. Some infinitary portions of mathematics may acquire their meaning indirectly, through inferential relations with finitary portions of mathematics.20 Let’s now consider the second step of Moore’s argument. Moore moves 9 from our lack of understanding of a statement to the statement’s lack of truth. Moore’s argument—like Wittgenstein’s observations on this topic—is entirely about what statements we do and do not understand, and whether the concepts involved are well-formed, and the statements thus ‘grammatical’ in Wittgenstein’s philosophical sense.21 Statements that are completely ungrammatical in the ordinary sense of being syntactically ill-formed cannot express truths and may be gibberish. Wittgensteinians want to say that a whole host of claims that pertain to metaphysics, including claims about infinity and the claims of sceptics, are gibberish.22 But the problem is these statements are not obviously ungrammatical and gibberish. The whole procedure of using grammar to decide what is comprehensible and hence what statements could be true is an idealist procedure that begs the question against realism from the start. Why should the fact that we lack a completely adequate understanding of statements about supertasks in empirical terms thereby mean that such supertasks cannot be accomplished? We should not conflate statements, and our understanding of them, with the reality that those statements purport to describe. Moreover, given human fallibility and epistemic limitations, it is worrisome to make the human imagination the arbiter of metaphysics. No defence of this excessively restrictive verificationism about concepts is offered. The Wittgensteinian line, if it is not to collapse into sheer verificationism, must maintain an agnostic position on the completion of supertasks. 3. Against (I): The problem with idealism. So far we have discussed (V) and (D) as background assumptions of EPLIM. Now we consider (I) (Idealism). We have seen that EPLIM is not valid without a hidden idealist premise.23 On this view, mathematical objects are viewed as constructions or ideas of the human mind, and have no existence independent of the mind. In this way it becomes immediately explicable why mathematical objects should conform to the limitations of the human mind. Each artificial object bears the mark (and imperfections) of its maker. One of the imperfections of the human mind is its inability to survey sequentially an infinite series. Consequently on the idealist view the numbers must suffer a similar imperfection: they must simply give out at the point from which surveying and construction is not possible.24 These considerations may readily be taken modus tollens as an argument for realism: there are infinite mathematical objects (such as the class of natural numbers), where these objects exceed the mind’s capacity to sequentially survey them, so such mathematical objects must exist independently of the mind. We are going to present two strategies for refuting idealism about mathematical objects. The first strategy provides a wedge against idealism by arguing that there are real items that are not conceivable in one sense (and hence lack existence as ideas). The upshot of this discussion is that we should not insist on knowledge by acquaintance of all mathematical objects. Insofar as premise (2) of the epistemic limitations argument is a request for knowledge by acquaintance, it must be rejected. The second strategy exploits a different tack, arguing that the incompleteness of mathematical knowledge is something that should be expected if realism is true. This style of argument for realism (and against idealism) is quite indirect and less conclusive. Nonetheless, it does seem that idealism naturally predicts that there should be few (if any) limits to our knowledge in mathematics, a prediction at odds with mathematical experience. Our wedge against idealism follows a strategy recommended by Thomas 10 Nagel in The View from Nowhere (1986). Idealists generally equate conceivability of an item with its real existence.25 Idealists reason in accordance with principle (Id): If X is not conceivable by us, then this fact entails that X is not real and does not exist. In general, however, (Id) is not valid except for ideas, whose esse est concipi. It is easy enough to generate counter-examples to (Id), both mathematical and otherwise in content: Example 1:The platypus was prima facie not conceivable by European naturalists prior to its discovery by them, but was actual and existent all along. Example 2: The transfinite numbers were not positively conceivable to Locke and Leibniz, but if numbers are real, they are real and yet were not always conceivable. Example 3: The infinite decimal expansion of is still not entirely conceivable to finite minds in full detail (they cannot imagine it and survey it), but (according to realists), it enjoys full mathematical existence; there is, for example, a fact of the matter as to whether the n’th digit is 7, for all natural numbers n. These examples are problematic. The first two examples are not obviously true if the distinction between ideal conceivability and prima facie conceivability is drawn.26 The third example begs the question against realism. In example 1, it might be true that the platypus was always ideally conceivable, even if the European naturalists failed in practice to conceive of it. Likewise, in example 2, the transfinite numbers were ideally conceivable even if Locke and Leibniz were not in a historical position to become aware of transfinite number theory. Surely had such able minds as Locke and Leibniz lived in the 20th century they would have been able to appreciate the truths of transfinite number theory. However, appeals to ideal conceivability are problematic. It is unclear what such ideal conceivability amounts to. We have no non-circular definition of ideal conceivability. Ideal conceivability is supposed to allow us to abstract away from ‘contingent limitations’. But it is not clear where we should stop in our abstraction. Apparently everything is conceivable except that which is contradictory!27 Example (3) would also be rejected by some idealists as a tendentious example. Idealists would say that the lack of full conceivability in detail of π is an artefact of the realist way of describing and interpreting the situation. Some constructivists (intuitionists) hold that only so much of the number exists as can be computed or specified by a construction.28 The statement that ‘there is a fact of the matter as to whether the nth digit is a 7’ is an application of the law of the excluded middle, which intuitionists reject.29 Though example (3) is philosophically tendentious, it represents a mainstream view in the mathematical community. Such counter-examples are not conclusive. So another tack is needed. Idealism implies a kind of arrogance about the scope of human understanding: idealism implies that if x is real, then x will be (ideally) conceivable by the human mind. To refute this principle, we need to find an x such that x is both 11 real and yet not conceivable by us. Of course it is a tricky business to specify something that we cannot conceive.30 Is not to specify x to conceive x in a certain manner? One way round this impasse is to argue by analogy. In general we can see that there is most likely such an x for us, as there exist features of reality that, although evident to us, cannot be conceived by minds less powerful than our own. Nagel (1986) suggests that Gödel’s theorem is not understandable by someone with a permanent mental age of a nine year old, but still true.31 Similarly, but by analogy with more powerful minds, we cannot survey the infinitely many digits in π’s decimal expansion, but a more powerful artificial or divine intelligence may be able to do so. Moreover, despite the fact that we cannot clearly and distinctly perceive the whole expansion of π, it does not follow that π does not have a determinate identity. 32 At least it does not follow unless we have independent reasons to assume idealism. Idealists will again appeal to the ideal conceivability of items in claiming that the conceivability of an object is necessary (and sufficient) for its reality. However, apart from the problems with ideal conceivability already noted, this move sets the standards too high. Less than ideal conceivability should not impugn the reality of an item. In fact we think there are many objects of mathematics that are inscrutable and real. A prime example is the notion of a real number. We have to allow that there are random real numbers. These real numbers correspond to random infinite sequences of digits, which comprise their decimal expansions. Random real numbers cannot be defined by rules laid down a priori and cannot be surveyed in their entirety. Their legitimacy is due to reasoning from inference to the best explanation. We know that the real number line is uncountable by Cantor’s diagonal argument. We also know that we can only compute countably many real numbers. So we infer that there are real numbers that cannot be so computed and defined. These real numbers are the `dark matter’ of analysis. They make our theory of real numbers come out right, even though we lack an individual acquaintance with each of these real numbers.33 Thus there must be mathematical objects that are not known by acquaintance. Premise (2) of the epistemic limitations argument does not elaborate on the kind of knowledge required of the objects of mathematical knowledge. When empiricists conclude that we have no positive idea of the infinite, they presumably do so on the grounds that we lack acquaintance with the infinite. This demand for knowledge by acquaintance is similarly present in Moore’s assumption that to make sense of statements about infinite tasks, we require ‘a kind of direct encounter with the infinite’. And it is dubious, or at least highly contentious, that we do have knowledge by acquaintance of each mathematical object; for example, it is hard to believe in recognisably distinct acts of acquaintance with a 1000 sided polygon and an 1001 sided polygon. However, this situation does not offer support for premise (1) of the epistemic limitations argument, or for the finitist conclusion. For it may well be that we have other ways of understanding infinitary mathematical statements that do not rely on acquaintance.34 The realist has an advantage in allowing that our knowledge of mathematical objects can be partial and incomplete at times.35 We do indeed have epistemic limitations that prevent us from, for example, entirely perceptually surveying an infinite decimal sequence. But such epistemic limitations do not in themselves dictate that we should therefore assume that no such sequence exists independently of our surveying and constructing it. 12 Intuitionists explicitly take such a stance: they hold that objects do not exist until they are explicitly constructed, or at least until a recipe could be given for their construction. The realist motto is: ‘completeness of objects, incomplete knowledge’. The intuitionist (and idealist) tactic is the reverse: ‘complete knowledge, incomplete objects’. Our claim is that realism best explains incompleteness phenomena (both of the informal sort I mentioned and the formal sort), and various forms of idealism (intuitionism, constructivism) do not. However, we recognise that this remark is not conclusive, since idealists will simply work with a different ontology (a much truncated one) that does not exhibit the relevant incompleteness phenomena. Nonetheless, classical mathematics does have phenomena (eg random real numbers) which are unknown in the restricted sense of not being objects of acquaintance (or particular construction). These phenomena can be studied using indirect methods (knowledge by description, inference to the best explanation etc.). Such incompleteness (or better, partial ignorance) is something that we should expect from a realist point of view. Realism precisely allows for facts that transcend our ability to verify (and in some cases even conceive of) them. Thus realism about mathematics suggests that premise (2) of EPLIM is false: it is not true that all the objects of mathematics are knowable and comprehensible to the human mind. The respect in which some mathematical objects are not knowable is that they are not knowable by acquaintance (‘surveyable by perception’). Idealists should not expect such partial ignorance of all the facts about mathematical objects. In fact, in order to eliminate such ignorance, the intuitionists have to tear down classical mathematics to eliminate it. This observation about the philosophy of mathematics is actually a special case of a general point. Nagel [1986] has observed that scepticism about our knowledge of the world is to be expected (even encouraged and tolerated) on any robust realist approach to the world. As Nagel puts it: ‘Realism makes scepticism intelligible.’36 The same holds true, mutatis mutandis, for partial ignorance (incomplete knowledge) and realism about mathematical objects. On the idealist view, any partial ignorance that we display about mathematical objects must be illusory: the objects themselves have fewer properties and less detail than we thought. On the realist view, our partial ignorance is readily explained by the richness of mathematical reality, combined with our epistemic limitations.37 4. The Realist Response. In this section, we consider various broad realist responses to EPLIM. The previous sections were mainly critical of the argument for finitism on the grounds that it makes assumptions (eg idealism, verificationism, the alleged datum) that cannot be defended. This section goes on the offensive, sketching the realist alternatives that enable one to confidently reject the premises of the epistemic limitations argument. As it turns out, none of the realist alternatives depend on the detailed reconstruction of EPLIM (from section 2.2) The realist response is best articulated against the original, simple two-step version of EPLIM. As suggested at the outset, both premises (1) and (2) of the original argument alone can and should be rejected by realists. The denial of these premises results in interesting and plausibly true views that are not often considered in the philosophy of mathematics. The first realist alternative results from denying premise (1) of the 13 epistemic limitations argument, and seeing where that denial leads. We call this solution ‘Cantor’s way’, since it accords with Cantor’s insistence that human beings can represent infinite objects (such as the transfinite numbers and infinite sets). In the Grundlagen (1883), he wrote: What I assert and believe to have proved in this work as well as in my earliest writings is this: that following the finite there is a Transfinitum…that is, there is an unbounded stepladder of determinate modes which in their nature are not finite but infinite, but which just like the finite, can be determined by well-defined and distinguishable numbers. I am convinced that the domain of definable quantities is not exhausted by the finite quantities, and the bounds of our knowledge may accordingly be extended without violence to our nature. ….In particular I contend the human understanding has an unbounded capacity for the stepwise formation of classes of integers. 38 Cantor’s point is consistent with the observation that we cannot survey, sequentially, by a quasi-perceptual act, all of the infinitely many members of the natural numbers. Our perceptual epistemic limitations are not denied. But our powers of representation and comprehension go well beyond what we can perceive. In other words, we could see Cantor’s affirmation of infinitary knowledge as depending on a rejection of the empiricist perceptual model for mathematical knowledge. Cantor stands squarely in the rationalist tradition on which emphasis is placed on an intellectual understanding of the infinite, rather than a sensory perceptual experience of it. What differentiates Cantor’s views from the rationalists is that he has the mathematical theory to back up his assertions. Cantor provides a practical demonstration of the representability of such infinities. In this connection, it is instructive to review briefly how Cantor introduces the transfinite numbers, which he regards as actual infinities. The natural numbers are producing by performing the successor operation (+1) on an initial element 0, leading to the series N={0,1,2,3,4……..}. To get the idea of the infinite number of natural numbers, another operation is needed: taking the limit. is the first ordinal number not reached by the successor operation. To reach , one thinks of it as the first number to succeed all the natural numbers, and simultaneously as the set consisting of the natural numbers in their canonical order. The size of can be gauged on a relative scale: it is 0, the first transfinite cardinal number. To get the next transfinite cardinal number 1, another operation is needed: forming the power set (set of all possible subsets). Thus we take all different subsets of N to derive the power-set P(N)= {Ø, {0}, {1}, {0,1}, {0,1,2}, {1,2,3}, {1,2,3,4}, ….}. For each element n of N we can either put it in a subset or not, so P(N)=20. Cantor further proves that 2n>n as a general rule. So the power-set operation can enable us to keep forming infinite sets larger than any transfinite number. These few operations enable formation of Cantor’s original universe of transfinite cardinal numbers. 39 Let us consider the philosophical character of Cantor’s principles for introducing infinite numbers. Mayberry [1977] perversely calls Cantor’s position ‘finitism’. Mayberry intends to emphasize Cantor’s assumption that we can treat and manipulate transfinite numbers in the same way as finite 14 numbers.40 His operations (successor, limit, power set) take us from finite numbers to infinite numbers. Obviously, these operations would not be acceptable to empiricists: they take us far beyond reflection on sensory experience. Yet, these operations are part of a coherent theory, and we do seem to form positive ideas of infinities through these operations. A purely negative idea of the infinite-- all that is available according to empiricists such as Locke-is simply the idea that such infinity is ‘not finite’. But our knowledge of infinities in set theory appears to go well beyond such restricted negative claims. Cantor’s way of dismissing EPLIM, then, is simply to reject premise (1). In this realist response we identify the different kinds of epistemic limitations that we have, and then note that as regards mere abstract representation and conception (as opposed to perception) we are not as limited as finitists think. From the fact (however ill defined and understood it is) that the human mind is finite, it does not follow that the human cannot have ideas of something infinite. However, another kind of realist response is possible. This response depends on disambiguating premise (2) of the original epistemic limitations argument. Premise (2) states that the objects of mathematics must be ‘comprehensible and knowable’. Two interpretative questions arise. First, to whom must the objects of mathematics be knowable? Second, in what way must be the objects of mathematics be knowable? Let us call the response that emphasizes the different kinds of minds that might be said to know mathematics ‘the Augustinian way’. Although it is natural to assume that what is at stake is comprehension by an ordinary, nonidealized human mind with finite capacities, there seems no principled reason why this need be the case. Perhaps there is some unlimited divine or artificial mind that can indeed survey infinite collections perceptually. Such a mind could have an ‘intellectual intuition’ of all the natural numbers. Here the suggestion is not to abandon the perceptual model of mathematical knowledge but to extend the faculty of perception to include a non-empirical ‘intellectual intuition’. Augustine suggests that God has such an infinite synoptic vision of infinite collections that are too large for human minds to consecutively represent. Augustine writes: Although the infinite series of numbers cannot be numbered [by us], this infinity of numbers is not outside the comprehension of him ‘whose understanding cannot be numbered’. And so, if what is comprehended in knowledge is bounded within the embrace of that knowledge, and thus is finite, it must follow that every infinity is, in a way that we cannot express, made finite to God, because it cannot be beyond the embrace of his knowledge. [Augustine, quoted in Hallett [1984]: 36] 41 Augustine suggests that the omniscience of God’s mind would require that there can be infinite collections that are comprehended as complete and actual in the same way that finite numbers are.42 If one were to accept Augustine’s view, with all of its ontological baggage, then the epistemic limitations argument would straightaway fall down. The argument would be invalid by equivocation. Premise (1) concerns human epistemic limitations. Premise (2)—for Augustinians—concerns divine knowledge. The conclusion concerns mathematical objects. Apart from the problems already identified concerning the need to bridge the gap between epistemological claims in the premises and a 15 metaphysical conclusion, it is clear that if the argument is interpreted in Augustine’s way, then the conclusion (3) cannot be drawn at all. Human epistemic limitations are irrelevant to the ultimate content of mathematics. The Augustinian position, however, is not such a happy resting place for realists. Augustinian realism is, in the end, a form of idealism, whereby real existence requires conceivability by an ideal mind. The essence of realism is to allow objects to exist independently of any mind, human or divine. We saw earlier that premise (2) of the epistemic limitations argument is ambiguous in yet another way: it is unclear in what way the objects of mathematics are comprehensible and knowable. We have suggested that the empiricist insistence that all mathematical objects be knowable by acquaintance is misplaced. In classical mathematics, there are more real numbers than we can name, specify, or calculate. In addition, each real number can be represented by an infinite sequence of digits in decimal expansion: we cannot even imagine surveying all these digits. However, our inability to know the real numbers in this way does not exclude our knowledge of them through other means. We can now envisage an interpretation of the argument on which (1) and (2) are true, but from which no finitist conclusion follows. According to this interpretation, premise (1) asserts that human beings cannot perceptually survey and represent infinite series. Premise (2) asserts that the objects of mathematics have to be knowable in some general sense, but does not require knowledge by acquaintance. Thus with a properly broad epistemology the existence of infinite mathematical objects sits happily alongside human epistemic limitations. There is yet another way to revise premise (2). As hinted earlier (section 3) realists may wish to deny that the objects of mathematics are knowable in the formal sense of being such that they can be characterized by a complete axiomatic system. There is a connection between the infinity of mathematical objects and formal incompleteness results. A detailed discussion is beyond the scope of this essay. However, one should note the observation by Torkel Franzén: From a philosophical point of view, it is highly significant that extensions of set theory by axioms asserting the existence of very large infinite sets have logical consequences in the realm of arithmetic that are not provable in the theory that they extend.43 Franzén’s observation suggests that epistemic limitations (at least as regards what can be proven in a given axiomatic system) should be seen as an inevitable result of trying to grapple with a deeply infinite, inexhaustibly rich ontology. In other words, after Gödel’s results, we might see our epistemic limitations as a reflection of an infinite ontology, not as a reason to deny the existence of such an infinite ontology. 5. Conclusion The argument against actual infinities in mathematics from human epistemic limitations is a bad argument, yet it has been historically influential in various guises. The premises of the argument were endorsed by the classical empiricist philosophers, such as Hume and Berkeley. The premises have also informed contemporary work sympathetic to finitism such as Moore [1989]. We have shown that although EPLIM is not valid in its initial form, it can 16 be made logically valid through the addition of controversial supplementary assumptions such as (V) (verificationism), (I) (idealism), and (D) (the datum that we do not experience anything infinite). These additional assumptions motivate the finitism that we find in empiricism. We have suggested that EPLIM fails to be sound because of the inclusion of these false premises. In particular, we suggested that (I) involves the fallacy of mistaking our epistemic limitations for constraints on reality. We further argued that even (D) is not obviously true, as there is nothing logically incoherent in supposing that we might experience an infinite object (cf. Oppy [2007]).44 We have seen that realists have a number of resources in responding to EPLIM. First, they can and should reject the idealist supposition and the alleged datum that were found to be hidden premises in the argument. Second, they should disambiguate both major premises (1) and (2) of the argument. They can, following the rationalist approach (endorsed by Cantor), suggest that we have positive knowledge of infinities in mathematics where this knowledge is not knowledge by acquaintance. In summation, we have found ample grounds to agree with Moore’s remark that the concept of the infinite signalled the failure of classical empiricism as a philosophical movement. This does not yet demonstrate that more contemporary empiricisms may not fare better. Contemporary empiricism, such as Quine’s approach to mathematics and Maddy’s physicalism, allows for a richer epistemology than the classical empiricist approach. Of pivotal importance is the allowance for modes of knowledge that are indirect (inference to the best explanation, knowledge by description, coherence as legitimation).45 In retrospect, we should not be surprised by the failure of an attempt to constrain metaphysics by using an overly restrictive empiricist epistemology. Far from being a cause to restrict the metaphysics underlying mathematics, epistemic limitations should be taken as a reflection of the richness of that metaphysics.46 School of Mathematics and Statistics, UNSW, Sydney Australia 17 In recent discussion ‘the problem of remoteness’ has been overshadowed by a more ancient and basic problem: the problem of giving an account (preferably causal) of the mind’s knowledge of abstract objects (see Benacerraf (1973]). Yet for physicalists (such as Maddy (1990)) about mathematics the problem of remoteness is more important than Benacerraf’s problem. If basic mathematical knowledge concerns concrete objects, there is still the problem that some mathematical truths require the existence of infinitely many such objects. 2 We find a similar set of distinctions in M. Giaquinto, The Search for Certainty (Oxford: Oxford University Press, 2002). 3 The school of ‘strict finitism’ or ‘ultra-finitism’ is represented by Van Dantzig, ‘Is is 101010 a natural number?’, Dialectica, vol 10, 1955. One of the originators is A. S. Yessenin-Volpin, ‘The ultra-intuitionistic criticism and the antitraditional program for foundations of mathematics’, in Intuitionism and Proof Theory, A. Kino, J. Myhill, and R. E. Vesley, eds., North-Holland, 1970, pp. 1–45. 1 The Axiom of Infinity is a basic assumption of standard Zermelo-Frankel (ZF) set theory. The Axiom of Infinity guarantees that there is at least one infinite set. In symbols: Here if the empty set belongs to the inductive set I, and so does x, then the successor set of x also belongs. 4 We cannot address the opposition to Quine’s method here, though we note that his criterion of ontological commitment (‘To be is to be the value of a (bound) variable’) has come under increasing pressure from nominalists (see Azzouni (2004)). 6 Part of what legitimates the inference, in Hume’s eyes, from idea to reality is his infamous ‘copy principle’. According to the ‘copy principle’, every adequate idea corresponds either to a sense-impression or can be decomposed into simple ideas that do so correspond. Given the premise that the human mind has a finite capacity to perceive, remember, and assemble sense-impressions, it follows that human beings cannot construct or experience an infinite object. By Humean empiricist lights, then, it follows that a finite mind simply cannot form an adequate idea of an infinite object. This is exactly what Hume insists: 'Tis universally allowed, that the capacity of the mind is limited, and can never attain a full and adequate conception of infinity: And tho' it were not allow'd, twou'd be sufficiently evident from the plainest observation and experience' (Hume (1740] Treatise I.II.1) Consequently, even if an infinite object could exist, it could not exist as an object of human knowledge in mathematics. Mathematical knowledge must consist in knowledge of adequate ideas. So there is a swift empiricist argument from our lack of an appropriate experiential basis for our idea of the infinite to the conclusion that mathematics cannot treat of infinite objects. 7Cf. D. Hume, Treatise, Book I Part II section 1. 8 Realists of course reject this kind of inference, on the grounds that the objects known are independent from the knowing subject. 9Proponents of strong AI may deny this claim on the grounds that human beings can create computers that have capacities (such as the capacity to beat chess champions) that most humans themselves lack. The computer metaphor is not 5 18 essential to the philosophical point. 10 Jacquette (2001] appears to endorse this line of reasoning in his defence of Hume’s finitism as a relevant position in contemporary philosophy of mathematics. 11 Even Kant seems to have been aware that assumption (1) is equivocally true at best, true in one sense and false in another. For this reason, Kant rejects conclusion (3) and maintains that the mind can represent space, for example, as being potentially infinitely divisible. For the suggestion that Kant’s prohibition on representing infinity by concepts resulted from his use of monadic logic, see M. Friedman, Kant and the Exact Sciences (Cambridge, MA: Harvard University Press, 1992). Furthermore, Kant’s straightforward acceptance of Euclidean geometry as giving us knowledge of the structure of space commits him towards viewing space as infinite (both in extension and division). So Kant, at least as regards the realm of appearances, is no strict finitist. 12In contemporary terms, the issue is that premises (1) and (2) involve epistemic operators and thus have an opaque context. In the conclusion (3), the epistemic operators are ‘peeled away’ to derive existential conclusions about the objects themselves. But this kind of move is suspect and related to the sin (according to Quine and Kaplan) of ‘quantifying into opaque contexts’. However, we do not think the Quinean diagnosis of what’s wrong with the argument is especially helpful, in treating metaphysical arguments as though they were logical mistakes. 13 Locke (1689] is typical: ‘So that what lies beyond our positive idea towards infinity lies in obscurity, and has the undeterminate confusion of a negative idea; wherein I know I neither do nor can comprehend all I would, it being too large for a finite and narrow capacity….’ (Locke, (1689]/(1955) I.xvii., 151-152). 14 J. Benardete, Infinity (Oxford: Clarendon Press, 1966), chapter II. 15 Some argue that remnants of verificationism are found in Dummett’s philosophy. For the charge that Dummett’s anti-realism is just verificationism dressed up, see M. Devitt, Realism and Truth (Oxford: Blackwell, 1984); E. Craig, The Mind of God and the Works of Man (Oxford: Clarendon Press, 1997): “anti-realism is a very natural successor to positivism” (p.286). For a defence of Dummett against the charge, see C. Wright, ‘Scientific Realism, Observation and Verificationism’, in his Realism, Meaning, and Truth (Blackwell 1986, second edition, 1993). For further argument that Dummett’s semantics is unacceptably verificationist, see W. Alston, A Realist Conception of Truth (Ithaca: Cornell University Press, 1995), chapter 4. 16 In any case, verificationism displays an arrogance about the human ability to know reality that does not seem warranted. Craig (1997] singles out this passage from the positivist Schlick (1936] as particularly awful: …no meaningful problem can be insoluble in principle…This is one of the most characteristic results of our empiricism. It means there are no limits to human knowledge. The boundaries which must be acknowledged are of an empirical nature and therefore never ultimate…there is no unfathomable mystery in the world (Schlick (1936) 156). Without such arrogance about our ability to decide the truth of propositions, verificationism has the consequence (unintended by its originators) that we understand and know almost next to nothing, including in science. It is not clear what the ground of such confidence is, but one guess is that it rests on 19 the phenomenalist and idealist view that objects are constructions of sense-data undertaken by the human mind. If so, then verificationism’s plausibility seems intimately linked with constructivism (Craig (1997) 286). 17 As we have seen, the empiricists endorsed this claim, and together with their empiricist epistemology, it underpinned their finitism. Kant, as we might expect, compromised between empiricism and rationalism on this topic. Kant asserted that the infinite was not an object of perception while holding that it could be intellectually cognized (Kant (1790)/(1952): 99-105). 18For a review of this literature, see G. Oppy, Philosophical Perspectives on Infinity (Cambridge: Cambridge University Press, 2006). We agree with Oppy’s claim that those sympathetic to finitism, such as Crispin Wright, have overstated what can be concluded from the literature on supertasks. 19 Descartes’s Reply to Gassendi, Objections and Replies, ed. J. Cottingham, D. Murdoch, and R. Sturgeon, The Philosophical Writings of Descartes, vol. II, (Cambridge: Cambridge University Press, 1985), 252. 20This kind of picture is developed in detail by S. Lavine, Understanding the Infinite, (Cambridge, Massachusetts: Harvard University Press, 1994). Lavine views his program as Hilbertian. 21On grammar in Wittgenstein, see P.M.S. Hacker, Insight and Illusion (Bristol: Thoemmes Press, 1989), 193-206. 22For Wittgenstein’s views on infinity, see his Philosophical Remarks (Oxford: Basil Blackwell, 1975), including the Appendix 1: ‘The Concept of Infinity in Mathematics’, 304-314. To be sure, there may be different meanings that attach to the term ‘non-sense’ in Wittgenstein’s corpus. We have used the term ‘gibberish’ to single out the prejorative meaning of ‘non-sense’, on which there is really no hidden transcendent meaning to the non-sense statement. There is also a non-pejorative, mystical kind of ‘non-sense’ in Wittgenstein’s view, which is connected with what can be shown, rather than said. For the mystical kind of non-sense, see Wittgenstein’s Tractatus logico-philosophicus, proposition 6.522 (London: Routledge, Kegan and Paul, 1922), p.187. However, in the Philosophical Remarks, there are several passages (e.g. Appendix I, par. 128) where Wittgenstein seems to be saying that talk of the actual infinite is non-sense in the pejorative sense. For more on Wittgenstein’s views on infinity, see A.W.Moore, ‘Wittgenstein on Infinity’, forthcoming in Marie McGinn, ed., The Oxford Companion to Wittgenstein. 23 There seems to be more sympathy for idealism about mathematical objects than for physical objects. One reason is that mathematical objects are commonly supposed to be abstract, causally inert universals devoid of secondary qualities such as smell, taste, and colour. So for those who find a full-blooded Platonism about such objects unattractive, the thesis that mathematical objects are created by human mental activity is a natural alternative. As Dummett explains the motivation of intuitionism, While, to a platonist, a mathematical theory relates to some external realms of abstract objects, to an intuitionist it relates to our own mental operations: mathematical objects themselves are mental constructions, that is, objects of thought not merely in the sense that they are thought about, but in the sense that, for them, esse est concipi.23 Another reason for finding idealism attractive is the emphasis on mathematics as a human activity and a failure to distinguish between the human 20 activity and its real subject-matter. However, as we shall see, as the general reasoning underlying idealism is fallacious, the arguments against idealism apply equally to mathematical objects and other objects. 24 One does in fact find this incredible thesis in B. Rotman, Ad Infinitum...the Ghost in Turing’s Machine, Stanford University Press 1993. 25Nagel characterises idealism in these terms, as ‘the position that what there is must possibly be conceivable by us, or possibly something for which we could have evidence.’ (Nagel 1986: 93). His summary fits Berkeley’s views as suggested by his remarks in Three Dialogues between Hylas and Philonous. Berkeley has Philonous make Hylas concede that it is contradictory to speak of (and conceive) of objects and qualities existing without the mind, hence without conception. (See also Principles of Human Knowledge, section 23). To move to conceivability rather than perception in ‘esse est percipi’ is, if anything, to make the position more plausible. 26 Chalmers (2002] distinguishes between prima facie and ideal conceivability in a bid to defend the principle linking conceivability and possibility (CP principle: if x is conceivable, x is possible). 27 Notoriously, some philosophical positions—such as Graham Priest’s dialetheism (Priest (1987])---would take issue with this conventional view. 28 See Christopher Ormell, ‘The Continuum: Russell’s Moment of Candour’, Philosophy (2006), 81: 659-668. For a realist rejoinder, see A. Newstead and J. Franklin, ‘On the Reality of the Continuum’, Philosophy 83 (2008), 117-27. 29 L.E.J. Brouwer, ‘Does Every Real Numbers have a Decimal Expansion?’, in P. Mancosu, From Brouwer to Hilbert: The Debate on the Foundations of Mathematics in the 1920s, (Oxford: Oxford University Press, 1998), 28-35. 30See Graham Priest’s reconstruction of Berkeley’s argument against ‘existence unconceived’, in Beyond the Limits of Thought (Cambridge: Cambridge University Press, 1995), pp.65-77. 31T. Nagel, The View from Nowhere (Oxford: Clarendon Press, 1986), p.95. 32 (Indeed to give up this principle would be to give up the comparability of the real numbers, a tenet of classical analysis. Of course, that is just what Brouwer and radical intuitionists do. But the consequences are pretty destructive of conventional, mainstream mathematics. ) 33A. Newstead and J. Franklin, ‘On the Reality of the Continuum’, Philosophy, January 2008. 34One obvious consequence of this approach would be that knowledge of the infinite in mathematics is not epistemologically basic. But this seems entirely acceptable: perhaps our knowledge of the infinite is something we are less committed to than our knowledge of finitary arithmetical truths. 35 By ‘incomplete knowledge’ we have in mind something more informal than Gödel’s famous incompleteness results. We simply mean incomplete knowledge in the sense that not all of the object’s characteristics are known to us in our current state. 36Nagel, The View from Nowhere, p.90. 37 To be sure, there may be other reasons to be finitist besides the epistemological considerations. Marion (1998) suggests Wittgenstein maintained finitism, but rejected the argument from epistemic limitations. 38Cantor, Grundlagen, in W. Ewald, (ed.) From Kant to Hilbert, II (Oxford: Clarendon Press, 1996), 891. 39For details, see M. Potter, Set Theory and its Philosophy (Oxford: Oxford 21 University Press, 2004), 117. 40 Mayberry introduces this startling, counter-intuitive characterisation of Cantor. It is counter-intuitive, of course, because Cantor is the defender of infinities par excellence and no finitist in the usual sense. However, Mayberry’s use of the term is meant to suggest that methodologically and epistemologically, Cantor treats finite and infinite numbers as on a par. See J. Mayberry, ‘On a consistency problem for set theory’, British Journal for the Philosophy of Science, 1977, vol.28: 1-34, 137-170. 41Augustine, City of God, Book XII, chapter 19, as quoted in M. Hallett (1984) op. cit, at 36 - denied by Wittgenstein, however, who thinks God should not cheat by going beyond human mathematics: Remarks on the Foundations of Mathematics VII-41 (3rd ed, Oxford: Blackwell, 1978), at 408. 42 The argument was revived by Constantin Gutberlet, Das Unendlich metaphysisch und mathematisch betrachtet (Frankfurt am Mainz, 1878). 43 T. Franzén, Gödel’s Theorem: An Incomplete Guide to its Use and Abuse (Wellesley, Massachusetts: AK Peters 2005), 152. 44 Graham Oppy, Philosophical Perspectives on Infinity, (Cambridge: Cambridge University Press, 2006). 45 For a physicalist approach to mathematics, see P. Maddy, Realism in Mathematics (Oxford: Clarendon Press, 1990). For a modified Quinean realism, see P. Maddy, Naturalism in Mathematics, (Oxford: Clarendon Press, 1998). On inference to the best explanation in mathematics, see C. Jenkins, ‘Knowledge of Arithmetic’, British Journal for the Philosophy of Science 56 (4), December 2005, 727-47. For a moderate Aristotelian empiricism, D. Gillies, ‘An Empiricist Philosophy of Mathematics and its Implications for the History of Mathematics’, The Growth of Mathematical Knowledge, E. Grosholz and H. Breger (eds.) (Boston: Kluwer, 2000), 46-51. 46 Work on this paper was carried out as part of a project funded by the Australian Research Council (ARC) for 2007-2010, Discovery Project no. DP0769997. References Azzouni, J. (2004) Metaphysical Myths, Mathematical Practice, Cambridge: Cambridge University. Berkeley, G. (1982) A Treatise Concerning the Principles of Human Knowledge, (K. Winkler, Ed.). Indianapolis: Hackett. (First published 1710) Bell, J.L. (2005) The Continuous and Infinitesimal in Mathematics and Philosophy. London: Polimetrica. Bernadete, J. (1966) Infinity. Oxford: Clarendon Press. Cantor, G. (1966) Grundlagen, in W. Ewald, (Trans. and Ed.) From Kant to Hilbert, vol. II, Oxford: Clarendon Press. Chalmers, D. (2002) ‘Does Conceivability Entail Possibility?’, in T. Gendler & J. Hawthorne, eds., Conceivability and Possibility, Oxford: Oxford University Press, pp.145-200. Craig, E. (1987) The Mind of God and the Works of Man. Oxford: Clarendon Press. Dummett, M.A.E. (1977) Elements of Intuitionism. Oxford: Clarendon Press. Franzén, T. (2005) Gödel’s Theorem: An Incomplete Guide to its Use and 22 Abuse. Wellesley, Massachusetts: A.K. Peters. Franklin, J. (1994) Achievements and Fallacies in Hume’s Account of Infinite Divisibility. Hume Studies, 20 , 85-101. Franklin, J. (2006) Artifice and the natural world: mathematics, logic, and technology. In K. Haakonssen, (Ed.). Cambridge History of Eighteenth Century Philosophy, Cambridge: Cambridge University Press. Frasca-Spada, M. (1998) Space and Self in Hume’s ‘Treatise’. Cambridge: Cambridge University Press. Gauss, Carl Friedrich. (1831). Brief an Schumacher (1831); Werke 8, 216. Gillies, D. (2000) ‘An Empiricist Philosophy of Mathematics and its Implications for the History of Mathematics’, in E. Grosholz, & H. Breger (Eds.), (pp.46-51): The Growth of Mathematical Knowledge. Boston: Kluwer. Hallett, M. (1984) Cantorian Set Theory and Limitation of Size. Oxford: Oxford University Press. Hume, D. (1740/1978) A Treatise of Human Nature. (P. Nidditch, ed.): Oxford: Clarendon Press. (First published 1740) Jacquette, D. (2000) David Hume’s Critique of Infinity. Leiden: E.J. Brill. Kant, I. (1781/1911) Critique of Pure Reason. (N. Kemp-Smith, Trans.) New York: Macmillan. (First published 1781) Kant, I. (1790]/(1952) Critique of Pure Judgement. (J. Meredith, Trans.) Oxford: Clarendon Press. (First Published 1790) Lavine, S. (1994) Understanding the Infinite. Cambridge: Harvard University Press. Locke, J. (1689)/(1955). An Essay Concerning Human Understanding, London: Prometheus. (First published 1689.) Mancosu, P. & E. Vailati. (1991) ‘Torricelli’s Infinitely Long Solid and Its Philosophical Reception in the Seventeenth Century’, Isis 82, 50-70. Moore, A.W. (1989) ‘A Problem for Intuitionism: The Apparent Possibility of Performing Infinitely Many Tasks in a Finite Time’, Proceedings of the Aristotelian Society 90 , 17-34. Reprinted in D. Jacquette, (Ed.), (2002). Philosophy of Mathematics: An Anthology . Oxford: Blackwell, 312-321. Moore, A.W. (1990) The Infinite. London: Routledge. Nagel, T. (1986) The View from Nowhere. Oxford: Clarendon Press. Okasha, S. (2001) ‘Verificationism, Realism, and Scepticism’, Erkenntnis 55.3, 371-385. Oppy, G. (2006) Philosophical Perspectives on Infinity, Cambridge: Cambridge University Press. Pérez-Ramos, A. (1988) Francis Bacon and the ‘Maker’s Knowledge’ Tradition. Oxford: Clarendon Press. Priest, G. (1995) Beyond the Limits of Thought. Cambridge: Cambridge University Press. Strawson, P. F. (1966) Bounds of Sense. London: Methuen. Wright, C. (1982] ‘Strict Finitism’, Synthese 52 no.1: 203-282. Wright, C. (1986]. ‘Scientific Realism, Observation and Verificationism’, in his Realism, Meaning, and Truth, Oxford: Blackwell, second edition 1993. 23