Suppose, for the sake of the argument, that the White House is occupied by a bully president, whose blunt lies not only ever increase in number but also in preposterousness. Or suppose that, to subvert the call for political or legislative action, “merchants of doubt” call into doubt well-established scientific facts such as climate change or evolution. Or suppose that almost every dramatic, politically sensitive event triggers a proliferation of conspiracy theories. Or suppose that journalists and the media continuously get bad press because they are accused of producing fake news. Or still, suppose that although in theory, communication technology allows for maximum freedom of expression of opinion in an ideal marketplace of ideas, in practice, the real marketplace of ideas blocks rather than facilitates the free and open exchange of views.
Our handful of examples illustrates that one needs little imagination to conceive of real-world situations in which finding out what is true, and what isn’t, is of pivotal, societal importance. Can philosophers be of any help in this quest for truth? Philosophers have always been mesmerized by truth. Part of their interest in truth is focused not on trying to find out what is true and what isn’t, but rather on the end-products of such inquiries, i.e., knowledge, to know what is true and what isn’t. In this context, philosophers have developed and defended different theories of knowledge, i.e., rival views on the nature, value, possibility, structure, sources, and kinds of knowledge.
The subdomain of philosophy that is concerned with the theory of knowledge is called epistemology, from the Greek word episteme, knowledge. It’s one of the two pillars of theoretical philosophy, the other being metaphysics (see chapter on ontology). Other subdomains in theoretical philosophy, like, e.g., philosophy of science (see chapter on philosophy of science) or philosophy of mind (see chapter on philosophy of mind), raise epistemological and metaphysical questions in more specific contexts like science and the mind. One might even argue that in practical philosophy as well, the most fundamental philosophical questions are, deep down, epistemological and metaphysical questions. For example, when we evaluate someone’s actions morally, their intentions often play an important role, but can we ever know someone’s true intentions? And what about moral responsibility? Is such a thing even possible when free will and causal determinism seem to be incompatible with one another? In this chapter, we will consider some big questions in epistemology. In Sections 2 and 3, we will discuss answers to the following questions: What is knowledge and is knowledge possible at all? Subsequently, in Sections 4 and 5, we shall look at opposing accounts of justification. Reliabilism, the account of justification discussed in Section 5, shifts our attention away from accounts of what knowledge is to the methods by which we acquire it and that most often result in true beliefs. In the last section, we shall suggest that if epistemology aspires to be of some fundamental help in solving pressing real-world problems, like the ones mentioned above, it should focus more on the context of inquiry by unraveling and pinpointing “heuristics,” or problem-solving strategies, and by aligning these novel findings with the earlier results from traditional epistemology.
One of the central problems in epistemology is the analysis of knowledge, i.e., the question “What is knowledge?” According to the traditional analysis, which has already been considered in Plato’s Theaetetus, the epistemic subject S knows the proposition that p if and only if (1) S believes that p, (2) p is true or, in other words, it is a fact that p, and (3) S is justified in believing that p. For instance, you know that you are reading this chapter because you experience reading this chapter. That experience, in turn, justifies your belief that you are reading this chapter, and your belief is also true: you are in fact reading this chapter. This traditional analysis has also been called the tripartite or the JTB-analysis, i.e., knowledge amounts to Justified True Belief.
In the traditional analysis, the function of the justification condition is to rule out epistemic luck, i.e., cases in which it’s only accidental, coincidental, or fortuitous that the belief of the epistemic subject happens to be true. Consider, e.g., the fictional, lying president hinted at in the introduction (any resemblance to actual presidents is of course purely coincidental). Suppose that out of 1000 of his statements, 999 are lies and only one is actually true. Suppose, furthermore, that an obstinate, gullible adherent believes every claim the president is making, and so, not only believes the 999 falsehoods, but also the president’s rare, true claim. Doesn’t it seem counterintuitive to conceive of that accidentally true belief as knowledge? According to the JTB analysis, it does, and the reason is that the gullible adherent lacks proper justification.
However, is the mere justification condition enough to rule out all conceivable cases of epistemic luck? In his famous three-page paper “Is Justified True Belief Knowledge?”, Gettier (1963) questions whether the three conditions of the traditional analysis are jointly sufficient for knowledge with two quite specific counterexamples. Here we’ll interpret an earlier counterexample by Russell (1948) along the same lines: Russell’s stopped clock. Suppose that a man looks at an analogue clock that has in fact stopped, though he thinks it is working, and looks at it exactly at the moment when it tells the correct time. All three conditions of the traditional analysis of knowledge have been met: the man acquires a justified true belief as to the time of day. However, it’s counterintuitive to conceive of the man’s justified true belief as a genuine case of knowledge. In other words, there are cases of epistemic luck that are not ruled out by the justification condition as such.
Attempted solutions to this Gettier problem try to explain why subjects in Gettier cases do not know that p, despite having a justified true belief that p. Some add a fourth condition to the JTB-analysis, i.e., an X-factor that marks the difference between genuine cases of knowledge and Gettier cases. One attempted solution is the “no false lemmas analysis”: on top of justified true belief, knowledge requires that the belief may not be inferred from any falsehood. The no false lemmas approach solves Russell’s clock: the man doesn’t know the time of the day because, while the clock has stopped, he erroneously assumes it is still going. Another attempted solution is the “no defeaters analysis”: on top of justified true belief, knowledge requires that there may not be any undermining evidence unavailable to the epistemic subject. In the case of Russell’s clock, there is such a defeater: if the man knows the clock has stopped, he wouldn’t acquire his justified true belief as to the time of the day.
Another strategy to solve the Gettier problem seeks to replace the justification condition. For instance, Goldman (1967) suggests that instead of S having justification for her belief that p, the causal condition must be met; i.e., the belief that p must be caused by the fact that p. Goldman’s causal analysis solves Russell’s clock: although the man’s justified true belief is caused by the time indicated by the clock, the indicated time is not caused by the actual time of day, but by the fact that the clock has stopped. Replacing the justification condition is also the strategy that Dretske (1971), Goldman (1976), and Nozick (1981) adopt, proposing the sensitivity condition instead, which requires that S would not believe that p if p were false. Truth sensitivity solves Russell’s clock as well because if the actual time of day differed from the time indicated by the clock, the man would still believe the indicated time to be right.
Although all the alternative analyses mentioned above can successfully deal with Gettier cases like Russell’s clock, none of them have proven immune to counterexamples up until now. According to Williamson (2000), this predicament results from attempting to add an objective condition, i.e., the truth condition, to the analysis of knowledge, which is rather entirely subjective. Another trend in epistemology pointing in a similar subjectivist direction is standpoint epistemology. For example, Sandra Harding (1991) draws our attention to the social group to which the epistemic subject belongs, and whether that group is marginalized.
A second central problem in epistemology is whether knowledge is attainable in the first place. Meeting the belief condition is not hard at all, but meeting the truth condition may be impossible. Can you know that it’s true that the external world and other minds exist? It’s conceivable and thereby at least logically possible that you are the only mind that really exists, and that other minds and the external world are merely your confabulations. Can we really exclude such a skeptical alternative? Can we know for sure that it doesn’t obtain? And if we cannot rule out such skeptical alternatives, are there any basic things we can establish as certainties, like that we have hands or are reading this chapter? In medieval times, skeptical arguments were offered to discuss the notions of knowledge and certainty (Adriaenssen, 2013). But it is Descartes who is credited with having defined the problem of skepticism with skeptical alternatives. In his meditations, starting from the well-known experience of waking up from a dream and realizing it was all a dream and the subsequent question of how we know that we are not dreaming the whole time, Descartes asks us to take one step further and imagine that we are systematically misled by an evil demon. Whereas the thought that we might be dreaming threatens only the existence and our knowledge of the external world, such an evil demon would undercut even something as sure as mathematical knowledge. The logical possibility of the dream and the evil demon constitutes skeptical alternatives, i.e., alternative explanations of our experiences that endanger whatever we think we know. While we need to rule them out, we can never do so completely. Recall, e.g., the sensitivity condition: you cannot know that you are not systematically misled by an evil demon because if you were, you would still believe that there was no evil demon.
Let’s develop a similar but more recent example. Suppose that you are the epistemic subject S and that the proposition p that you know is that you have hands. Enters the skeptical alternative. It’s conceivable and thereby at least logically possible that you are not a person of flesh and bones, but rather a Brain In a Vat (or BIV) in which all the experiences that you have are artificially produced by electric stimulation. Of course, you do not really believe that you are a BIV, but you cannot exclude the possibility either. So far so good; it’s possible that you know that you have hands and that you do not know that you are not a BIV at the same time. However, the skeptical challenge arises from the conditional assumption that “if you know that you have hands, then you know that you are not a BIV.” The underlying idea is that knowledge is factive: you can only know that you have hands if it’s a fact or if it’s true that you have hands. In other words, you can only know that you have hands if it’s true that you are not a BIV, but rather a person of flesh and bones (and hands). So, if you really know that you have hands, then it is implied that you know, or at least are in a position to know, that you are not a BIV.
The argument of the skeptic is that being able to know that you are not a BIV is required for you to know that you have hands, and since you cannot know that you are not a BIV, you cannot know that you have hands either. According to the skeptic, this point generalizes: whoever the epistemic subject S is, whatever the belief p is, and whatever the skeptical alternative is, S cannot know that p because, to know that p, S should be able to exclude all skeptical alternatives to p, but S cannot exclude any of its skeptical alternatives.
In the face of the skeptic’s arguments, we can defend talk about knowledge by drawing attention to the role and importance of context. Normally, we know that we have two hands, the earth is round, and 2+2 equals 4. However, when skeptical alternatives are made salient to us, we no longer know those things. Before you started to read this Section 3, you were not even aware of the possibility of your being a BIV, and so you knew that you have hands in that context. However, by introducing in this section the skeptical alternative that you might be a BIV, the context has changed. Now, in this new context, the skeptical alternative that you might be a BIV has become salient. Now you realize that you cannot exclude the possibility of the skeptical alternative, and as a result, you no longer know that you have hands in this context.
Kindly note that changing the context, making mere logical possibilities salient to cast doubt, is exactly what the “merchants of doubt,” to whom we alluded at the opening of this chapter, typically do to subvert the call for political or legislative action. For instance, on behalf of tobacco companies, they argue that, despite appearances, it’s conceivable and thereby possible that there is no relation between smoking and lung cancer. Or they fabricate so-called scientific controversies to make the logical possibility salient that, since the industrial revolution, there has been no causal relation between human intervention and climate change (Oreskes & Conway, 1981/2010).
We can also defend ourselves against the skeptic’s argument by appealing to context in a slightly different way. We justify our claims all the time by adding reasons for these claims. To adopt a famous example from Austin (1946), when someone claims that a bird is a goldfinch, she might refer to its red head as typical of such a goldfinch. In response, someone else may wonder whether the bird is not a woodpecker instead of a goldfinch, pointing out that woodpeckers too have red heads. This is what we call an appeal to a relevant alternative. In the context of a bird in your garden, it makes sense to discuss whether it is a woodpecker or a goldfinch, but it would be really strange if someone wondered whether it was not a mechanical or stuffed bird, because such alternatives are not relevant in the context. So, to make a knowledge claim, an epistemic subject should only be able to exclude all relevant alternatives, and which alternatives are relevant is determined by the context. However, there are hardly any contexts in which skeptical alternatives are relevant. As a result, skeptical alternatives are hardly ever legitimately challenge knowledge claims.
We might think that we can defend ourselves against wholesale skepticism and the conclusion that we really know nothing by arguing that the alternatives articulated by the skeptics are just very unlikely, that we do not have to take them seriously. There are several explanations for the things we experience, and some are better than others. It is on the basis of the quality of the explanations that we can discriminate between what to believe or not. Pyrrhonian skeptics, though, like Pyrrho, Sextus Empiricus, and Montaigne, have argued that there is no independent, unbiased, rational way to identify the best among competing criteria. Competing explanations simply have their own pros and cons. It is impossible to weigh these arguments properly, let alone settle for an independent, unbiased, rational conclusion. So, basically, Pyrrhonian skeptics are skeptical about the possibility of justification, and one of their main arguments to that effect is known as Agrippa’s trilemma.
Beliefs can be justified by further beliefs. In Austin’s example, the birdwatcher’s belief that the bird in the garden is a goldfinch is justified by her belief that goldfinches have red heads and that the bird at issue has the relevant characteristic. But what justifies these further beliefs? Still further beliefs in a justification chain? This can’t go on ad infinitum, can it? According to Agrippa’s trilemma, justification of beliefs by other beliefs is impossible altogether because there are only three options, and none of them is worth wanting: either
The justification chain terminates in basic beliefs that do not need further justification because they are, in a sense, self-justifying, like axioms in mathematics, or
The justification chain is circular, or
The justification chain does go on ad infinitum.
To avoid Pyrrhonian skepticism about justification, one might accept one of the horns of the trilemma. The first option is to accept the axiomatic horn. Throughout the history of ideas, by far most philosophers have adopted this approach, which is called foundationalism: like a pyramid, human knowledge in general and science in particular rest on a solid foundation of absolutely certain, infallible, basic beliefs. Consider perceptual knowledge. Of course, we may err; our perceptual judgments may be illusory or hallucinatory. However, we cannot be mistaken about having the pure sensory experiences that we do have. Our interpretations may be wrong, but we are infallible in having experienced what Descartes called sensations, Locke ideas, Hume impressions, Mill phenomena, and Russell sense-data. Beliefs that are ultimately justified by such basic beliefs are safe and sound.
There are many objections to classical foundationalism, but the main source of concern is the gap between, on the one hand, pure, uninterpreted sensory experiences and, on the other hand, minimally interpreted perceptual beliefs. Pure, uninterpreted sensory experiences may be infallible, but they are inferentially sterile; because they are experiences, and not yet propositional beliefs, nothing can be inferred from them, and hence they cannot provide foundational justification. Looking at a banana, I have a visual experience, which doesn’t justify anything as such. Only when sensory experiences are minimally interpreted, i.e., if they are turned into propositional beliefs, can they provide justification. So, my visual experience may trigger me to form the belief that there is a curved yellow object in front of me, which in turn may provide justification for a further belief that the object in front in me is a banana. But the trouble with minimally interpreted perceptual beliefs is that they are no longer infallible – the object may in actual fact be pale green and straight – and hence, they cannot be the basic beliefs that terminate the justification regress either.
A second option is to accept the circular horn of Agrippa’s trilemma. This results in the position we call “coherentism.” According to coherentists, justification is holistic and not transferred from one belief to another. Beliefs hang together in more or less coherent “webs of beliefs.” Whether or not a particular belief is justified is determined by whether and to what extent it fits or coheres with the other beliefs in the web of beliefs of the epistemic subject. The minimal conception of coherence is logical consistency, but most coherentists believe that explanatory connections contribute significantly to the overall coherence of beliefs and webs of beliefs. Intuitively, four completely unrelated beliefs are less coherent than, for instance, a doctor’s belief that a single diagnosis explains three disparate symptoms; the diagnosis pulls the symptoms together.
There are many objections to coherentism as well, but the main source of concern is the isolation problem: coherent webs of beliefs can become completely detached from reality. Think, for instance, of conspiracy theories; those often make up internally coherent wholes, in which beliefs hook up neatly. However, from the “outside,” people see that such theories are way off-base.
A third and final possibility is to accept the regressive horn of Agrippa’s trilemma: the justification regress is not vicious, but virtuous. Beliefs can be justified by an infinite chain of justification. In response to the objection that the human mind is finite and limited, Klein (1999) argues that the infinitist is not claiming that the epistemic subject actually needs to have an infinite chain of reasons in mind. What is required is that it must be possible in principle, but not in practice, to provide an infinite chain of reasons. Moreover, Atkinson & Peijnenburg (2017) prove that the longer a justification chain becomes, the more the returns of further justification diminish, so that ultimately, no further justification is called for.
Up until now, we have considered the justification of beliefs by other beliefs. An underlying assumption has been that, at least in principle, the epistemic subject has access to the justification of a belief: the foundationalist can trace back or reduce her belief to a foundational layer of basic beliefs ultimately supporting it, the coherentist is or can become aware that her belief hangs together with other beliefs constituting a coherent web of beliefs, and so on. What these approaches to justification share is internalism: factors internal to the epistemic subject determine whether her belief is justified. However, triggered by the Gettier problem (see Section 2), philosophers have also developed externalist approaches to justification. According to these approaches, only factors external to the epistemic subject determine whether her belief is justified. According to externalism, then, the epistemic subject needs no access to the justification of a belief.
Reliabilism is such an externalist theory of justification. To grasp the basic notion of reliability, we can think of the debate about the media and “fake news.” Journals or networks that often publish news that turns out to be false are unreliable. Readers, listeners, or viewers should not rely on them, but rather base their beliefs on information obtained from reliable outlets: journals and networks that almost never publish false information. So, in this context, being reliable is a function of being truth-conducive; the degree of reliability of the media depends on the ratio of the true to false beliefs they tend to produce. Goldman (1979) argues that what is true for networks and journals is also true of ourselves: some of the ways in which we acquire beliefs are clearly less reliable than others. Guesswork, hasty generalizations, and wishful thinking are clearly unreliable because, more often than not, they produce false beliefs. In contrast, belief-forming processes like standard perceptual processes and good reasoning are reliable: they tend to produce more true than false beliefs. Hence, reliabilism proposes that a belief is justified if and only if it has been formed by a reliable process of belief-formation, and where the reliability of the process is defined in terms of its tendency to form beliefs that are true rather than false.
Unfortunately, reliabilism can be challenged by the Brain-In-a-Vat version of the evil demon scenario discussed above. The reason for that is that we need to be sure about the existence of the external world and our (perceptual) relation to it before we can say anything about the reliability of our perceptions. Suppose that in some scenario, your counterpart, i.e., your local representative in some possible world, has exactly the same perceptual experiences as you have, and forms exactly the same perceptual beliefs as you do. However, there is one all- important difference: in contrast to you, your counterpart is systematically misled by an evil demon and is in actual fact a Brain In a Vat. As a result, almost all your counterpart’s perceptual beliefs are false, and we must qualify the perceptual processes by which they are formed as highly unreliable. However, the input is the same: you and your counterpart have the same perceptual experiences. And the output is the same as well: you and your counterpart form the same perceptual beliefs. So, if your perceptual beliefs are justified, your counterpart’s perceptual beliefs should be justified as well.
There are several other problems with reliabilism that we will not discuss here. However, one thing to note is that a major advantage of reliabilism is that, unlike internalism, it allows for what Sosa calls animal knowledge. A significant part of human knowledge consists of first-order, reliably produced true or apt belief. Your first-order true belief that you are now reading this chapter is reliably formed based on perception (and perhaps also a reliable bit of introspection). However, according to Sosa’s own hybrid account, human knowledge is not restricted to animal knowledge: you can upgrade your apt belief to what Sosa calls reflective knowledge by reflecting on how your apt belief fits coherently with other beliefs you have, like, for instance, your belief that you have bought this book in that bookshop, your belief that you could hardly wait to start reading this chapter, and so on.
We started this chapter with a handful of pressing and complex real-world problems, all related to the epistemic quest of finding out what is and isn’t true. Subsequently, we discussed big questions in epistemology – the analysis of knowledge, the possibility of knowledge, and the nature of justification – and to that end, we made abundant use of imaginary examples and counterexamples. At face value, we are now facing the applicability gap, i.e., the gap between, on the one hand, deep but narrow academic puzzles, and, on the other hand, complex real-world problems. Now, in what sense and in what way can philosophy in general, and epistemology in particular, help to bridge that applicability gap? What modest role can philosophy and epistemology play in addressing and solving the real-world problems of our time? The first key to reaping the harvest of more than 25 centuries of philosophy, for the purpose of solving pressing real-world problems, is “context.” Philosophical questions often are decontextualized questions, and that is why, ever since Plato, stipulated, imaginary cases and scenarios are the ideal tools for philosophical investigations. The downside is that there never seem to be definite answers to decontextualized questions. Philosophers prefer to put more and more possible solutions on the table, and to go on and on with considering and weighing their pros and cons. In a contextual approach, however, one tries to enrich the definition of a problem by bringing in more details and context in such a way that, at the end of day, a solution can be arrived at, i.e., the best solution for the problem at hand in its context.
In addressing real-world problems, philosophical, conceptual questions often and easily pop up. For instance, to become clear about what one knows at some point in time, one needs a conception of knowledge. Or to take another example, to assess to what extent a theory is justified, one needs a conception of justification. As we can learn from Sections 1, 3, and 4, without context, the analysis of concepts like “knowledge” or “justification” or still, for that matter, “coherence,” is an endless game. There simply are no one-size-fits-all solutions to philosophical, conceptual problems because in different contexts, different conceptions are appropriate. Only when we are clear about the context of the real- world problem that we want to solve, our goals, and so on, can we start defining a concept in a fruitful way. To that end, the exploration of possible conceptions during 25 centuries of philosophy has not been in vain. On the contrary, it allows the real-world problem-solvers to significantly speed up defining the concept at hand for the purpose at hand.
From our discussion of skepticism about knowledge in Section 3 and of skepticism about justification in Section 4, we can learn a second important lesson in how epistemology could and should make itself more useful in addressing and solving real-world problems. Epistemology should reorientate itself and focus more on inquiry, how to obtain knowledge, and how to find out what is and what isn’t true, than on analyzing end-products. In the real world, epistemic subjects in general, and problem-solvers in particular, are always in the midst of inquiry: some things they can safely assume and act on, whereas other things they need to find out or investigate further. In the midst of inquiry, there is little use for analyses of idealized, unattainable goals. By contrast, a profound understanding of the structure of inquiry, i.e., the structure of problem-solving itself, would be most useful. For instance, it would suggest to problem-solvers what the best next step is in solving the complex, real-world problem they are addressing.
Let me illustrate this point by adopting an analogy from Schurz and developing it a little bit. Suppose that someone in a hurry asks an epistemologist for the right way to the railway station. A traditional epistemologist will respond: “Among all possible ways that start where you are right now and end at the railway station, find the shortest one and choose it.” With that answer, the epistemologist does not risk anything. It is a true answer, but also completely useless. It bounces back the question that the person asking it is interested in, acting as if it is an answer. Epistemologists can do so much better, but not by providing substantive answers – after all, it is their profession to think and reflect on what knowledge is, not provide answers to all possible questions people might have. What they can do is provide what we can call a heuristic answer, i.e., an answer that sheds light on how to find the answer to the substantive question. The epistemologist can shed light on the quickest way to find that answer, or point out that one should first get one’s goals clear, and so on.
These two lessons fit together nicely. As we discussed toward the end of Section 3, “contexts” or problem-solving situations not only allow for fruitful conceptions but also determine which alternatives are relevant. In their turn, the relevant alternatives determine the further steps to be taken and the further questions to be asked in the process of solving the real-world problem. As a matter of principle, all voices in the marketplace of ideas deserve to be heard, but the epistemologist can learn and teach how to single out the most relevant ones, step by step.
Adriaenssen, H. T. (2013). Knowledge and the veil of representations [PhD Dissertation]. University of Groningen.
Atkinson, D., & Peijnenburg, J. (2017). Fading foundations: Probability and the regress problem. Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-319-58295-5
Austin, J. (1946). Other minds. Proceedings of the Aristotelean Society, Suppl. Vol. 20, 148–187.
Dretske, F. (1971). Conclusive reasons. Australasian Journal of Philosophy, 49(1), 1–22. https://doi.org/10.1080/00048407112341001
Gettier, E. (1963). Is justified true belief knowledge? Analysis, 23(6), 121–123. https://doi.org/10.1093/analys/23.6.121
Goldman, A. (1967). A causal theory of knowing. The Journal of Philosophy, 64(12), 357. https://doi.org/10.2307/2024268
Goldman, A. (1976). Discrimination and perceptual knowledge. The Journal of Philosophy, 73(20), 771. https://doi.org/10.2307/2025679
Goldman, A. (1979). What is justified belief? In G. S. Papas (Ed.), Justification and knowledge (pp. 1–23). Springer Netherlands. https://doi.org/10.1007/978-94-009-9493-5_1
Harding, S. (1991). Whose science? Whose knowledge? In Whose science? Whose knowledge? Thinking from women’s lives. Cornell University Press. https://doi.org/10.7591/9781501712951
Klein, P. (1999). Human knowledge and the infinite regress of reasons. Nous, 33(s13), 297–325. https://doi.org/10.1111/0029-4624.33.s13.14
Nozick, R. (1981). Philosophical explanations. Harvard University Press.
Oreskes, N., & Conway, E. (2010). Merchants of doubt. Bloomsbury. (Original work published 1981)
Russell, B. (1948). Human knowledge: Its scope and limits. George Allen & Unwin. https://doi.org/10.4324/9780203875353
Sosa, E. (2007). A virtue epistemology: Apt belief and reflective knowledge. Oxford University Press.
Williamson, T. (2000). Knowledge and its limits. Oxford University Press.