Topics in Epistemology Sven Rosenkranz Introduction One of the central questions of philosophy is to what extent we can be confident that we can properly identify our relation to the reality of which we are only a small part. There are two aspects to the question. On the one hand, we may ask how our subjective experience can be conceived to be part of the objective reality which we confront, and thereby inquire whether we are able to arrive at a coherent picture of the latter that includes both the painter and the painter’s point of view. This question lies at the heart of the philosophy of mind. On the other hand, we may ask what it takes for us to accurately represent how things objectively are, how we can assure ourselves that we succeed in our attempts to accurately represent them, and how far our ability to do so extends. This is the province of epistemology. We can distinguish three questions that epistemology is concerned to answer: What is the nature of knowledge? What are the sources of knowledge? What is the extent of our knowledge? In this course, we will chiefly be concerned with the third of these questions, which we will discuss in the context of the recent debate about scepticism. 1. The nature of knowledge Knowledge implies belief: if I know that p, then I take it to be the case that p (i.e. I take reality to be such that it is not misrepresented by saying that p). To take it to be the case that p is to believe that reality is as one thereby takes it to be, and so to believe that p. So we get: (i) Kxt[p] Bxt[p], where ‘x’ ranges overs subjects and ‘t’ ranges over times. Although belief is necessary for knowledge, it is not sufficient for knowledge. One may (and frequently does) hold beliefs that do not correspond to how things really are, i.e. one may (and frequently does) believe that p, for some proposition p, although it is not the case that p. By contrast, one cannot know that p unless it is the case that p. So we get: (ii) Kxt[p] p. We may put this by saying that knowledge implies truth. 1 Does the combination of these two necessary conditions suffice for knowledge? In other words, is true belief sufficient for knowledge? No. In order for a belief to count as knowledge it is not enough that the belief be true. It is further necessary that the belief be formed in appropriate ways – that it is, in some sense, responsive to the truth. For instance, if you come to believe that the person you just met is honest with you by consulting the stars, then even if that person is indeed honest with you, you do not know on that basis that she is. And if you come to believe that you have been selected by the recruitment committee because you know you have been lucky in the past, then even if you have in fact been selected, you still do not know on this basis that this is so. In some pertinent sense, your beliefs in these cases are no better than mere guesswork. It is a matter of debate what it takes to form a true belief ‘in appropriate ways’ in order for that belief to constitute knowledge. But we may at least say this much: (iii) Kxt[p] at t x has come to believe that p by successful application of a truthconducive method. The term ‘method’ is used in a broad sense to cover any kind of cognitive procedure, or routine, whose implementation can coherently be thought of as a means to find out whether p. A method is truth-conducive only if its successful application leads to the formation of beliefs that are more likely to be true than false. One way to flesh out the latter thought is by saying that a method is truth-conducive only if it is reliable. Note that successful application of a reliable method does not guarantee the truth of the belief to whose formation it leads. The reliability of a method is not a matter of its always leading to true beliefs, nor of its leading to a true belief at all times at which it is correctly classified as reliable. Rather, its reliability requires it to be such as to yield mostly true beliefs. Consequently, its (present) reliability is a matter of how it works in a whole range of cases (present, past and future). However, just as a meter reliably indicates a given quantity only insofar as it operates under conditions that don’t impair its function, a method will be reliable only relative to the prevailing of favourable environmental conditions. For instance, looking at a middle-sized object’s surface is a reliable means to find out which colour this object has only in circumstances in which the lighting conditions are normal. Similarly, checking on whether the water in the pot is boiling is a reliable means to find out whether the water’s temperature is around 100 degrees Celsius only if one is close to sea-level. So when we here speak of successful application of a method we mean to secure that two rather distinct kinds of conditions are met: (a) the subject is in the right state of mind (is lucid, sober, attentive etc.), performs well, properly responds to the outcome and, in general, does all she is expected to do in order to correctly apply the method (b) the prevailing environmental conditions are favourable, i.e. of the kind relative to which alone the method can count as reliable. 2 As just said, the consequent of (iii) does not guarantee the consequent of (ii): a subject may successully apply an otherwise reliable method, so that the environmental conditions are favourable and the subject does all she is required to do in order to rationally arrive at a belief on that basis, and yet the subject’s belief may be false. Thus, for example, checking how y looks is a pretty reliable means to find out whether y is a duck, but for all that, objects on occasion are not what they look and if confronted with a fake duck on a pond using that means will deliver a false belief. Some philosophers have imposed further conditions on knowledge that are meant to close this gap. Here are two such further conditions: Sensitivity: (iv) Kxt[p] the method m by which x has come to believe that p at t is such that if p were false and x were to apply m, x would not come to believe that p on that basis. The italicised clause is a subjunctive conditional. The truth-conditions for subjunctive conditionals can be understood as follows: A B The nearest A-worlds are B-worlds. Whether a world is nearby or remote is somewhat vague. It depends on the extent of change required in order for that world to be actual (or correspondingly, on how similar that world is to the actual world). In any case, if A holds in the actual world, then ‘A B’ implies that B holds in the actual world, because then the actual world is amongst the nearest A-worlds. It follows that the subjunctive conditional ‘A B’ entails the material conditional ‘A B’. Requirement (iv) is called ‘sensitivity’ because it gives expression to the idea that the method employed must be sensitive to whether p holds. Safety: (v) Kxt[p] the method by which x has come to believe that p at t is such that if x were to believe that p on the basis of m, p would be true. Requirement (v) is called ‘safety’ because it embodies the idea that the method employed does not easily lead one astray. A method of belief formation may be safe and yet fail to be sensitive. For example, I may come to believe, truly as we might suppose, that Mireia is not a political radical by gathering the kind of evidence one usually has for such claims (observations of her behaviour, testimony etc.); and yet, if Mireia was a political radical, she would be wise enough to conceal this, given her circumstances, and so I would still come to believe, albeit falsely, that she is not a political radical on that basis. So my method isn’t sensitive. But given that Mireia is in fact not a political radical, but a sober and moderate thinker whose sobriety and moderateness is deeply anchored in her upbringing and education, worlds in which she is a political radical are rather remote, at least they are more remote than the nearest possible worlds in which I come to hold the same 3 belief on the basis of the same kind of evidence. So my method, though insensitive, is safe all the same. Each of (iv) and (v) entails that if at t x comes to believe that p on the basis of successfully applying method m, x’s belief is true. That this is so can be shown as follows, where ‘Bxtm[p]’ abbreviates ‘At t, x believes that p on the basis of applying m’ and ‘Axtm’ abbreviates ‘At t, x has applied m’. 1 2 2 1,2 5 5 1,2,5 2,5 (1) (2) (3) (4) (5) (6) (7) (8) ¬p Bxtm[p] Axtm ¬p & Axtm (¬p & Axtm) ¬Bxtm[p] (¬p & Axtm) ¬Bxtm[p] ¬Bxtm[p] p assumption assumption from 2 from 1 and 3 Sensitivity from 5 from 4 and 6 from 2 and 7 1 2 2 1,2 (1) (2) (3) (4) Bxtm[p] Bxtm[p] p Bxtm[p] p p assumption Safety from 2 from 1 and 3. Typically, those who think that (iv) imposes a legitimate constraint on knowledge do not say the same about (v), and those who think that (v) imposes a legitimate constraint on knowledge do not say the same about (iv). This is an ongoing debate. Are the conditions mentioned so far jointly sufficient for knowledge? There is the concern that they are not. For note that even if a subject has come to hold a belief by successful application of a reliable method which is safe (or sensitive), the subject need not be aware of, let alone possess evidence for, these features. Thus, for all that has so far been said, the subject may not be aware of, or have evidence for, any of the following: the method used is reliable the environmental conditions under which this method is applied are favourable if p were false and one were to apply the method, one wouldn’t come to believe that p if one were to believe that p on the basis of applying the method, p would be true. These conditions are at least as external to the subject’s mind as the state of affairs to which ‘p’ relates. And doesn’t this just show that more is required of knowledge than that these conditions are satisfied? After all, the principal reason why we strive for knowledge, and are not content with mere true belief, is that we wish to get some reassurance that our representations of reality are accurate. Carefully following through a cognitive procedure by applying a given method in a lucid, sober and attentive way, and forming beliefs in response to the outcome of this application, does not yet guarantee that the method is as a matter of fact reliable and safe (or sensitive) and that the environment does as a matter of fact cooperate. Yet, if these are the external conditions distinctive of knowledge, how can 4 the desired reassurance be obtained by their simple fulfilment, while we so far have done nothing to ascertain that they are fulfilled? Those who take these concerns very seriously, and to disclose that we so far haven’t specified sufficient conditions for knowledge, are the epistemological internalists. By contrast, epistemological externalists rest content with the conditions specified thus far. The dispute between internalists and externalists is sometimes said to turn on the status of the so-called KK-principle: (KK) Kxt[p] Kxt[Kxt[p]], where ‘p’ ranges over non-epistemic states of affairs. Note that without some such restriction of the range of ‘p’, the KK-principle would already be psychologically implausible, because then the K-operator would be said to iterate indefinitely – just plug in the consequent into the antecedent’s ‘p’-slot. Since knowledge implies belief, any knowing subject would have to be credited with infinitely many beliefs. Other principles of iterativity are not subject to this constraint, because they use epistemic notions that do not imply belief, e.g.: Kxt[p] x is in a position to know that Kxt[p] Kxt[p] it is possible for x to know that Kxt[p]. If the KK-principle holds, then it will not do that the environmental conditions are as a matter of fact favourable and that the method used is de facto reliable (and safe or sensitive): in order for a subject to have knowledge it is furthermore required that the subject knows that this is so. However, although they typically deny the KK-principle, externalists qua externalists need have no qualms with any of this, and the internalists’ worries cannot thereby be allayed, if to have such second-order knowledge is again a mere matter of forming beliefs by applying a de facto reliable (and safe or sensitive) method under de facto favourable conditions. A better way to draw the line is to introduce a decidedly internalist notion of evidence. In order to do so, let us first reflect on what is involved in applying a method and to form a belief in response to the outcome of this application. Consider the method of using one’s senses to tell which colour a given middle-sized object has. Looking in the direction of the object and focusing on its surface, we are led to have a certain experience as of this object’s being, say, red. We form the belief that the object is indeed red, and do so in response to this experience. If the object had had a different colour, say, blue, we would have been led to have a qualitatively distinct experience, viz. as of its being blue, and would consequently have formed the belief that it is blue (and not red). Similarly, if we measure a certain quantity by means of a meter, e.g. the pressure in a cooker, we first take note of the meter reading and then form the corresponding belief about the pressure inside. If the meter reading had been different, so would have been our belief. 5 In both cases, we gain evidence for our respective beliefs (experiences in the first case, meter readings in the second). And now the important point is that we would not come to acquire knowledge in this way if we did not take note of this evidence. And even if the evidence itself is external to the subject’s mind (as in the case of the meter readings), taking note of this evidence, possessing it, is not (but is as internal to the subject’s mind as the colour experience). We may now formulate a principle about which internalists and externalists are likely to disagree: (*) There is no necessary condition for knowledge for whose obtaining the knowing subject possesses no evidence. Internalists will accept (*), while externalists will reject it. For the externalist, (*) sets the bar too high, because for the externalist, whether a given piece of information can really count as evidence for p will inter alia depend on its pedigree (its origin or the way it was obtained); and it is unlikely that any evidence will ever ‘wear its pedigree on its sleeves’. Rather, the right kind of pedigree that makes for evidence is the pedigree of information which is made available by successful application of a reliable method under favourable conditions; and we cannot tell whether a given piece of information has this feature just by taking it in. Yet, the idea that knowing subjects possess evidence, evidence for the right pedigree of that evidence, evidence for the right pedigree of the latter and so on and so forth, is, if these pieces of evidence are distinct, just as psychologically implausible as the unrestricted version of the KK-principle that allows knowledge to iterate indefinitely. Let us follow Crispin Wright and call pieces of evidence, on the externalist construal of ‘evidence’ according to which a piece of information counts as evidence only if it has the right pedigree, ‘warrants’. The notion of a warrant is partly internalist in that possessing evidence is a state of mind; and it is partly externalist because having the right pedigree requires that condition be met that relate to matters outside the subject’s mind. Since, as we have noted, the consequent of (iii) does not entail the consequent of (ii), one may possess a warrant for a belief, although this belief is false. We may use the notion of warrant to define an internalist notion of evidence which we may call ‘reason’: x has a reason for p x is in a state subjectively indistinguishable from possessing a warrant for p. To possess a warrant for p is to have a reason for p, since possessing a warrant for p is indistinguishable from itself. But the other direction does not hold, because reasons may lack the right kind of pedigree. From a subjective perspective, the notion of reason is more central than that of warrant, precisely because the subject cannot, just by processing the information she has, distinguish between being in a state in which she possesses a warrant for p and being in a state that is merely subjectively indistinguishable from a state of the first kind. 6 Is (*) too demanding if ‘evidence’ is understood in terms of reasons? Ignoring the sensitivity and safety requirements for the moment, we must check what it would take for a subject to possess reasons for the following external constraints: the method used is reliable the environmental conditions under which this method is applied are favourable. So far we haven’t distinguished between all-things-considered evidence (warrants, reasons) and prima facie evidence (warrants, reasons). Prima facie evidence for p can be overruled by countervailing evidence but still counts as prima facie evidence for p once so overruled, and it will be all-things-considered evidence for p at a given time, if no such countervailing evidence comes to light at that time. Note, though, that even allthings-considered evidence can be overruled in the sense that there may be all-thingsconsidered evidence for p at one time, but no such evidence at another time. We may think of prima facie evidence for p as pieces of information that may or may not add up to all-things-considered evidence for p at a given time depending on the other information available at that time. Rationally formed beliefs respect all-thingsconsidered evidence (rather than merely prima facie evidence). In light of this distinction, our previous description of what application of a method involves needs to be refined. For example, inspecting an object’s surface in order to tell which colour it has will lead us to enjoy a certain experience as of that surface’s being a certain colour, but whether we form the belief that it has that colour will also depend on the other evidence we have at the time. Thus, if we also have evidence to the effect that there are artificial sources of coloured light illuminating the object’s surface, we may well refrain from forming the relevant belief. Of course, that further evidence may not have the right pedigree. But even if we have no direct knowledge of the pedigree of the evidence we have accumulated, we can tell whether it coheres. Coherent evidence accumulated over long periods of time by using different techniques or sense-modalities may put us in a position to draw an inference to the best explanation concerning its origins. As the case may be, a world in which p is the case and the method we apply is reliable and is applied under favourable circumstances may be the best bet in order to account for the overall evidence we have so far accumulated. It is in this kind of way that an internalist may accept (iii) as a constraint on knowledge and yet comply with (*), on her preferred construal of ‘evidence’ in terms of reasons. 2. The sources of knowledge When it comes to the second epistemological question, there are two principal distinctions we need to draw, that between a posteriori and a priori knowledge and that between inferential and non-inferential knowledge. A posteriori knowledge is knowledge based on experience, whereas a priori knowledge is not so based. There may be a priori knowledge that p even if we have to have had some experience in order to acquire the concepts involved in the proposition that p. The terms ‘a posteriori’ and ‘a priori’ do not relate to the genesis of our 7 knowledge, but to its rational grounds (that which we would cite in order to justify our corresponding beliefs). A paradigmatic example of a posteriori knowledge is knowledge by perception (outer sense). It is our visual, auditory or tactile experience that is the source of our perceptual knowledge. A paradigmatic example of a priori knowledge is knowledge by mathematical proof. Here the source of our knowledge is not our outer sense, but reasoned reflection. Of course, we may also prove certain conclusions based on premises furnished by observation. But if the proof of these conclusions really depends on these premises, then the knowledge of them which we thereby acquire is not a priori but a posteriori, even if the proof rules we use can be recognised to be valid on purely a priori grounds. Inferential knowledge is based on inference, whereas non-inferential knowledge is not so based. A paradigmatic example of non-inferential knowledge is again knowledge by perception. Thus, when we see what the meter reads, we do not infer what it reads, even if we go on to infer from the meter reading what the pressure inside the pressure cooker is. A paradigmatic example of inferential knowledge is again knowledge by mathematical proof. However, although mathematical knowledge by proof is a paradigmatic case of both inferential and a priori knowledge and knowledge by perception is a paradigmatic case of both non-inferential and a posteriori knowledge, a posteriori knowledge may be inferential and a priori knowledge may be non-inferential. For instance, given our understanding of the concepts involved, we may come to know a priori that any president presides over something, without drawing any inference in the process. Similarly, we may come to know a posteriori that sometimes partnerships aren’t harmonious by inferring it from the known premise that Sun and Jin quarrel. Also, we cannot determine by the content of a piece of knowledge alone what its source must be: knowledge that the sum of 2 apples and another 3 apples is 5 apples may be arrived at a priori by use of basic arithmetic but it may also be arrived at a posteriori by counting apples. Similarly, one may know non-inferentially that someone is next door (by hearing them quarrel), but may equally know this indirectly, by inference from certain clues. Non-inferential knowledge may be divided into knowledge by outer sense (perceptual knowledge) and knowledge by inner sense. Knowledge by inner sense may be further divided into knowledge by introspection (of our mental lives), knowledge by understanding (of conceptual truths) and knowledge by rational intuition. The latter is arguably a category under which nothing falls, but some philosophers have insisted that it covers our knowledge of some parts of mathematics. Inferential knowledge can be further divided into knowledge by deduction and knowledge by induction, corresponding to the divide between deductive and inductive inference. While deductive inferences necessarily preserve truth, and so help to discern the logical consequences of what we know, inductive inferences are ampliative in that the truth of their conclusions go beyond what the truth of their premises logically guarantees. 8 Examples of deductive inferences are: No human has scales The creature from the black lagoon has scales -----------------------------------------------------------The creature from the black lagoon is not human This is a pen Pens exist outside our minds ------------------------------------------Something exists outside our minds Examples of inductive inferences are: All observed amphibians lay eggs -----------------------------------------All amphibians lay eggs The sun has always risen in the past --------------------------------------------The sun will rise tomorrow A special case of inductive inferences are abductive inferences (inferences to the best explanation). Examples are: Otterbourne was killed with a gun Bourget was seen at the murder site Bourget had a motive Bourget’s finger prints were on the gun ------------------------------------------------Otterbourne was killed by Bourget It appears to me as if there was a pen lying in front of me I feel lucid The lighting seems normal ----------------------------------------------------There is a pen lying in front of me As said, unlike deductive inferences, inductive inferences are not necessarily truthpreserving, i.e. there are possible circumstances in which their premises are true and their conclusion is nonetheless false. This immediately raises the question of why they should nevertheless be considered good inferences on which we can rationally rely when forming our beliefs. One may think that inferences of this kind – e.g. the inference concerning tomorrow’s sunrise – have proved reliable in the past and so should be trusted now. However, this kind of reasoning does itself use an inductive inference, viz. the one that proceeds from the observation that in the past the inference did not lead us astray to the conclusion that it will not lead us astray now – which is no better or worse than to infer from the fact that the sun has always risen in the past that it will rise tomorrow. Accordingly, the worry that inductive inferences are in bad standing has not been allayed. This is the famous problem of induction, first clearly stated by David Hume. Deductive inferences are not beset by this problem and so there are no corresponding qualms about whether it is rational to rely on them in forming our beliefs. Still, it needs to be explained in what sense they are at the service of inferential knowledge; and here philosophers typically invoke the following principle of closure: (Closure) (Kxt[p] & (p ⊢ q) & at t x has competently deduced q from p) Kxt[q], 9 where ‘A ⊢ B’ says that A entails B. We may assume that in order to competently deduce q from p, the subject must know at that time that p entails q. But this knowledge is in any case not enough in order to guarantee that if the subject knows that p, she thereby comes to know that q: she must further perform the deductive inference in question and thereby arrive at the belief that q. As we shall see later, Closure, although widely accepted, is not uncontroversial. However, here I just want to highlight the contrast between Closure and another, closely related idea, viz. that deductive inferences are at the service of acquiring evidence for the conclusion. In the literature, this idea goes by the name ‘warrant transmission’, because the evidence, if any, that a competent reasoner acquires for the conclusion will be the evidence she has for the premises which accordingly transmits across the (known) entailment. Closure may hold, and yet a deductive inference may, at least on occasion, fail to prescribe a route by which evidence may be acquired for the conclusion. Here is an example: Suppose my evidence for the claim A caravan of camels passes by some miles ahead is that it visually appears to me as if a caravan of camels passes by some miles ahead. Suppose that I then reason as follows: A caravan of camels passes by some miles ahead If a caravan of camels passes by some miles ahead, then what I see is no mirage --------------------------------------------------------------------------------------------------What I see is no mirage. The inference is deductively valid, and we can safely assume that the second premise is well-established, given my circumstances. But do I thereby acquire evidence that what I see is no mirage, evidence that is transmitted from the premises to the conclusion? Hardly. For, it seems that unless I could already discount beforehand that I am the victim of an illusion, I cannot take the evidence I have for the first premise to be sufficient to license my endorsement of that premise. Yet, if I could already discount beforehand that I am the victim of an illusion, then I had evidence for the conclusion already before I drew the inference, evidence that is in no way strengthened by the evidence I have for the first premise. Does this example likewise show that Closure fails? No. For if I really do know the first premise on the basis of my visual evidence, so that I am entitled to regard that evidence as sufficient to license my endorsement of it, then I must have independent evidence that puts me in a position to know the conclusion. What the example does show, however, is that the validity of Closure, though necessary, is not sufficient to explain how we can attain knowledge by deductive means. 10 3. The extent of our knowledge The question of how far our knowledge extends is not intended as a question about the extent of the knowledge we have so far acquired. Rather, it asks about the extent to which we can have knowledge. Anti-Realists hold that every truth is knowable, apart from those truths that record our ignorance, e.g., as the case may be, ‘The number of leaves on the Sycamore tree in my garden is even but no one will ever know’. Setting aside such truths (which are already unknowable for logical reasons), Anti-Realism is a fairly optimistic thesis about the reach of our epistemic powers which the Realist denies or at least considers unjustified. Realism seems a reasonable stance. However, denial of Anti-Realism, or its rejection as unjustified, can be understood to rest on rather different grounds. This becomes more obvious once we distinguish between kinds of claims that we endeavour to have knowledge of. Examples of such kinds are: : claims about the colour, shape and location of middle-sized everyday objects claims about the nature of middle-sized objects (whether they are cats, pens or rocks) claims about the microphysical properties of microphysical objects claims about one’s own mental life claims about the mental lives of others claims about lawlike connections between events claims about the past claims about the future Denial of Anti-Realism is compatible with the idea that for each such kind, we can know some claims of that kind. Similarly, rejecting Anti-Realism as unjustified is compatible with the concession that we are justified in thinking that we know some claims of each such kind. Thus, the position we commonly associate with the label ‘realism’ holds that although we can and do know claims of each of these kinds, and are justified in thinking that we can and do know this, we cannot know, or at least do not know that we can know, all truths of these kinds. Scepticism, by contrast, questions our capacity to know any claim of a given kind. Thus, scepticism about the past challenges our confidence that we are able to know anything at all about the past. Scepticism about other minds challenges our confidence that we are able to know anything at all about the mental lives of others. The radical sceptic challenges our confidence that we are able to know anything at all about the external world, not even that there is an external world! In this course, we will be concerned with this radical form of scepticism. The sceptical challenge may take the shape of a flat-out denial that we can have the relevant knowledge (first-order scepticism) or, more guardedly, the denial that we have any justification at all to think that we can have that knowledge (second-order scepticism). Sceptical challenges are developed by means of argument. Keith DeRose gives a general characterisation of first-order sceptical arguments, which I will reproduce here: 11 Let O be an everyday claim about our environment which we ordinarily take ourselves to know, e.g. that one has hands that there is a pen lying in front of one that one’s mother feels sad that one had breakfast this morning. For dialectical reasons, O should be thought of as a best case, i.e. as a claim of a kind (about one’s physical environment, about other minds, about the past) such that if one does not know that claim one does not know any claim of that kind. Let H be some kind of sceptical hypothesis, either to the effect that some cognitively disabling conditions obtains, e.g. that one is dreaming that one is an envatted brain that one is manipulated by an evil demon or to the effect that reality is fundamentally different from what we take it to be, e.g. that there is no external world that are no other minds that reality came into existence 5 minutes ago replete with traces of an apparent past. DeRose then takes the template for sceptical arguments to be this: (1) (2) ¬Kxt[¬H] ¬Kxt[¬H] ¬Kxt[O] Therefore: (3) ¬Kxt[O]. Premise (2) is equivalent to: Kxt[O] Kxt[¬H]. Accordingly, if O is known to entail ¬H, then premise (2) will be derivable from Closure, provided only that x can be assumed to competently perform the relevant deduction. Arguably, all of the following entailments hold: I have hands ⊢ There is an external world My mother feels sad ⊢ There are other minds I had breakfast this morning ⊢ Reality didn’t come into existence 5 minutes ago. If this is known to be so and x competently deduces the right-hand sides from the lefthand sides, then Closure allows us to conclude that if x fails to know the right-hand sides, x likewise fails to know the left-hand sides. 12 Given how O was chosen, x will then fail to know any ordinary claim about her physical environment, any ordinary claim about the minds of others and any ordinary claim about the past. Note that Closure alone will not necessarily furnish us with premise (2) if H is a sceptical hypothesis about a cognitively disabling condition. Thus, the following entailments plausible do not hold: I have hands ⊢ I am not dreaming/being manipulated My mother feels sad ⊢ I am not dreaming/being manipulated I had breakfast this morning ⊢ I am not dreaming/being manipulated. The case of envatment is rather different. Thus, if I am an envatted brain then I don’t have hands, and if I was always an envatted brain, then I won’t have a mother and most definitely didn’t have breakfast this morning. In light of such sceptical arguments, one has in principle three options: one may either succumb to scepticism and accept (3), or one may deny (1) or may deny (2). As we shall see, contextualism offers a more subtle line of response. It remains to be seen whether any of these responses is satisfactory and if so which is more satisfactory than the others. 13