20.01.03 · philosophy / epistemology

Epistemological skepticism and responses: Cartesian doubt, contextualism, externalism

stub3 tiersLean: nonepending prereqs

Anchor (Master): Descartes, R. — Meditations on First Philosophy (1641)

Epistemological skepticism is the view that we lack knowledge — or that we cannot justify the claim that we have it. The skeptical tradition is as old as philosophy itself, and the attempts to answer it define the shape of modern epistemology. This unit reconstructs the strongest skeptical arguments: Cartesian radical doubt (the dream argument, the evil demon, the brain in a vat), Humean skepticism about induction and causation, Pyrrhonian suspension of judgment, and Agrippa's regress. It then traces the principle of epistemic closure that gives the skeptic their leverage, and examines the major responses — Moorean common sense, relevant alternatives, tracking and safety conditions, contextualism, externalism, Putnam's semantic disproof of the brain-in-a-vat scenario, and transcendental arguments. The unit closes with extensions to skepticism about other minds, Goodman's new riddle of induction, moral skepticism, religious epistemology, and the evolutionary debunking of belief.

Intuition Beginner

Can you know anything for certain? René Descartes (1596-1650) tried to doubt everything he could, to find what, if anything, was certain. He imagined he might be dreaming, or that an evil demon was deceiving him about everything. His conclusion — "I think, therefore I am" (cogito ergo sum) — was meant to be bedrock: even if a demon is deceiving him, he must exist in order to be deceived.

Skepticism is the view that we cannot know what we think we know. Different versions challenge different things: that our senses deceive us, that we cannot know the external world exists, that induction is ungrounded (Hume), that we cannot know other minds. Each version shares a structure. Take a belief you thought was secure, show you could hold it even if it were false, and conclude you do not really know it.

Philosophers have responded in several ways. G. E. Moore argued that "here is a hand" is more certain than any philosophical premise that would deny it. Contextualists argue that "knows" shifts meaning with context — in ordinary talk you know, in a philosophy seminar you do not. Externalists argue that knowing need not require proving you know: a reliable process can deliver knowledge the subject cannot vindicate from the inside.

Visual Beginner

The diagram reads left-to-right as pressure and answer. The central band marks the hinge: every response must say something about closure, the principle that links ordinary knowledge to the denial of skeptical scenarios.

Skeptical challenge	Core claim	Principal response
Dream / evil demon (Descartes)	You cannot rule out dreaming or demon-deception	Cogito; Moorean common sense
Brain in a vat	Your experiences match an envatted brain's	Relevant alternatives; tracking; Putnam
Humean induction	Past regularity does not guarantee the future	Pragmatic vindication; naturalism
Agrippa's regress	Every justification needs another, endlessly	Foundationalism; coherentism; reliabilism
Other minds	You never experience another's consciousness	Analogical argument; Wittgensteinian criteria

Worked example Beginner

Descartes' method of doubt

Descartes' Meditations on First Philosophy (1641) opens with a deliberate experiment. Descartes resolves to treat every belief that can be doubted as if it were false, to see whether anything survives. The method is not ordinary doubt — it is radical and methodical.

Stage one attacks the senses. Eyes and ears deceive us — distant towers look small, sticks bend in water. If the senses have erred once, they cannot be fully trusted. Stage two is the dream argument. In dreams we are certain we see tables and skies, yet none of it is real. There is no internal mark that distinguishes dreaming from waking.

Stage three is the evil demon. Descartes imagines an omnipotent deceiver bent on misleading him — even about mathematics. Perhaps two and three do not make five; perhaps the demon confuses him at every step. Under this hypothesis, no belief is safe — not perception, not memory, not arithmetic.

What survives? Descartes' answer is the cogito. Even if the demon deceives him about everything, there must be a him to be deceived. To doubt is to think; to think is to exist. "I think, therefore I am" is the bedrock — one truth no demon can dislodge. The open question is whether the bedrock is wide enough to rebuild knowledge upon.

Check your understanding Beginner

Formal definition Intermediate+

A skeptical argument proceeds from a hypothesis under which the subject's beliefs are false, through a linking principle, to the conclusion that ordinary empirical knowledge is impossible. The linking principle is epistemic closure under known entailment.

Let $K_{S} p$ denote " $S$ knows that $p$ ." Closure says: if $S$ knows $p$ , and $S$ knows that $p$ entails $q$ , then $S$ knows $q$ .

K_{S} (p) \land K_{S} (p \to q) \to K_{S} (q)

The principle is deeply intuitive. If you know you have hands, and you know that having hands entails you are not a handless brain in a vat (BIV), then you seem to know you are not a handless BIV. Closure is what lets the skeptic move from "you cannot rule out the BIV scenario" to "you do not know ordinary things."

A skeptical hypothesis $H$ is a scenario constructed to be experientially indistinguishable from ordinary life yet under which the subject's ordinary beliefs are false. Let $p$ be an ordinary proposition ("I have hands") and $H$ a skeptical hypothesis ("I am a handless BIV"). The skeptical argument is:

P1. $\neg K_{S} (\neg H)$ — you do not know that $H$ is false.

P2. $\neg K_{S} (\neg H) \to \neg K_{S} (p)$ — if you do not know $\neg H$ , then (by closure, since $p \to \neg H$ ) you do not know $p$ .

C. $\neg K_{S} (p)$ — you do not know $p$ .

The argument is valid. Every response locates the flaw in a premise or in closure itself. Three formal conditions characterise the leading externalist responses:

Sensitivity (Nozick): $S$ 's belief that $p$ is sensitive iff, were $p$ false, $S$ would not believe $p$ : $\neg p □ \to \neg B_{S} (p)$ . A handless BIV would still believe it has hands, so the belief "I am not a BIV" fails sensitivity.
Safety (Sosa): $S$ 's belief that $p$ is safe iff, in all nearby possible worlds where $S$ believes $p$ on the same basis, $p$ is true. Safety is compatible with closure; whether it defeats skepticism depends on whether the BIV world counts as "nearby."
Relevant alternatives (Dretske): $S$ knows $p$ iff $S$ can rule out the relevant alternatives to $p$ . Relevance is context-dependent. Combined with a denial of closure: $S$ can know $p$ without knowing every entailed $q$ .

Key argument: the skeptical challenge and responses Intermediate+

Varieties of skepticism

Pyrrhonian skepticism — Pyrrho of Elis, later systematised by Sextus Empiricus in the Outlines of Pyrrhonism — recommends epoché, the suspension of judgment. The Pyrrhonian seeks, for every argument, an equal and opposite counter-argument; finding equipollence (isostheneia), they suspend judgment and thereby attain ataraxia, tranquility. The Pyrrhonian does not assert that knowledge is impossible (that would be a dogmatic claim of the sort they refuse); they simply decline to assent, living by appearances without committing to their truth ^{[Sextus Empiricus]}.

Academic skepticism — Arcesilaus and Carneades, heads of Plato's Academy in the third and second centuries BCE — took the stronger line that knowledge is unattainable. The Academics allowed that some beliefs are more reasonable than others, and that one can act on the persuasive (pithanon) without claiming knowledge. This moderate position was a precursor of probabilism.

Cartesian skepticism is methodological and local. Descartes uses doubt as a tool: by withholding assent from every belief that could be false, he seeks an indubitable remainder. The dream argument targets perception; the evil demon targets even mathematics. The cogito resists both. The broader project of rebuilding knowledge depends on God's veracity — a dependence that later philosophers (Gassendi, Hobbes, and above all Hume and Kant) judged a weak link ^{[Descartes 1641]}.

Humean skepticism targets induction and causation. Hume's argument runs: every inductive inference presupposes the uniformity of nature — that the future will resemble the past. This presupposition cannot be grounded deductively (the contrary is logically possible) or inductively (that would presuppose what it tries to prove). Causal inference is therefore not rationally warranted but rests on habit or custom — the psychological effect of repeated conjunction. Responses include Reichenbach's pragmatic vindication (induction is the best available policy even if not rationally necessitated), Strawson's naturalistic dissolution (the demand for a justification of induction is itself misconceived — we cannot help inducting), and contemporary Bayesian naturalism ^{[Hume 1748]}.

The skeptical argument reconstructed

The modern skeptical argument — what Barry Stroud calls the "skeptical challenge" — has a precise structure ^{[Stroud 1984]}:

P1 (underdetermination). You cannot know that the skeptical hypothesis $H$ is false. The BIV scenario is constructed so that your experiences would be identical whether or not $H$ obtains; no inspection can distinguish the two.

P2 (closure). If you cannot know $\neg H$ , and you know that $p$ entails $\neg H$ , then you cannot know $p$ .

C. You do not know $p$ (you do not know you have hands, that there is an external world, etc.).

The argument is valid. The dispute is over which premise — or which background principle — to reject. Five responses dominate the literature.

Moorean common sense

G. E. Moore's "Proof of an External World" (1939) offers a modus tollens against the skeptic ^{[Moore 1939]}. Moore holds up his hand and says "Here is a hand; here is another." The proof: here is a hand; if there is a hand, there is an external world; therefore there is an external world. Moore concedes that the proof is not dialectically effective against a convinced skeptic, but insists it is nonetheless a perfectly good proof — its premises are more certain than any philosophical principle that would undermine them. Where the skeptic reasons "you don't know you're not a BIV, so you don't know you have hands," Moore inverts the modus ponens into a modus tollens: "I know I have hands; I know that having hands entails I'm not a handless BIV; so I know I'm not a handless BIV."

The dispute becomes a question of epistemic priority. The skeptic treats the closure of knowledge under known entailment as a constraint that propagates inward from the hardest case (the BIV). Moore treats common-sense certainty as a constraint that propagates outward. There is no neutral court of appeal — each side insists its starting points are more secure.

Relevant alternatives

Dretske's relevant alternatives theory holds that to know $p$ , $S$ must be able to rule out the relevant alternatives to $p$ , not every logically possible alternative ^{[Dretske 1970]}. In the canonical zebra case: at the zoo, seeing a striped equine animal in a pen labelled "zebra," you know it is a zebra. You need not rule out that it is a cleverly disguised mule, because that alternative is not relevant in ordinary zoo contexts. The BIV scenario is similarly irrelevant in everyday contexts.

Dretske combines this with a restricted closure. You can know "this is a zebra" without knowing "this is not a cleverly disguised mule," even though the first entails the second. Closure fails because the alternatives relevant to evaluating a knowledge claim are not closed under entailment: the mule alternative becomes relevant only when explicitly raised.

Tracking and sensitivity

Nozick's tracking theory (1981) analyses knowledge in terms of counterfactual dependence ^{[Nozick 1981]}. $S$ knows $p$ iff:

$p$ is true.
$S$ believes $p$ .
Sensitivity: if $p$ were false, $S$ would not believe $p$ — $\neg p □ \to \neg B_{S} (p)$ .
Adherence: if $p$ were true (in relevant nearby worlds), $S$ would believe $p$ .

The BIV hypothesis breaks sensitivity for the belief "I am not a BIV": if you were a BIV, you would still believe you were not (the vat would stimulate you identically). So you do not know you are not a BIV. But you do know you have hands — because if you did not have hands, the normal perceptual process would not deliver the belief. Tracking preserves ordinary knowledge at the cost of denying closure: you can know $p$ without knowing an entailed $\neg H$ .

Safety

Sosa's safety condition — $S$ 's belief that $p$ is safe iff, in all nearby worlds where $S$ believes $p$ on the same basis, $p$ is true — handles Gettier cases and the fake-barns scenario without abandoning closure ^{[Sosa 1999]}. A belief is unsafe if it could easily have been false. The fake-barns subject's belief ("there is a barn") is true in the actual world but unsafe, because in nearby worlds the subject looks at a facade and forms a false belief. Unlike sensitivity, safety is preserved under known entailment: if $p$ could not easily have been false, and $p \to q$ , then $q$ could not easily have been false either. The skeptical question becomes whether the BIV world is "nearby" in the relevant sense — a question on which safe-belief theorists divide.

Contextualism

DeRose and Lewis defend contextualism: the truth-conditions of " $S$ knows $p$ " vary with the standards operative in the conversational context ^{[DeRose 1995; Lewis 1996]}. In ordinary contexts (low standards, everyday alternatives), you know you have hands. When a skeptic raises the BIV possibility, the standards shift: "you don't know" becomes true — but only relative to that elevated context. Lewis's version imposes rules of relevance: a possibility that is not attended to may be properly ignored, but the moment it is raised (the skeptic's move), it must be ruled out — and it cannot be.

DeRose's bank cases sharpen the point. In a low-stakes variant, a depositor asks whether the bank will be open Saturday and we say "she knows." In a high-stakes variant (a large cheque that must clear Monday), the same evidence no longer suffices, and we say "she doesn't know." The contextualist says the standards for "knows" shift with the practical context of the attributor, not the subject — a move that dissolves the skeptical paradox by making the skeptic and the ordinary attributor both right, relative to their respective contexts.

Exercises Intermediate+

Exercise 2 (medium, argument reconstruction).

Reconstruct Hume's problem of induction in premise-conclusion form, and identify precisely where the skeptic's leverage comes from.

Hint

Hume's argument turns on the principle of the uniformity of nature. How could that principle itself be justified?

Answer

P1. Every inductive inference presupposes the uniformity of nature (UN): that the future will resemble the past, that unobserved cases will resemble observed ones.

P2. UN cannot be justified deductively, because its falsity is logically consistent — a world in which regularities suddenly reverse is conceivable.

P3. UN cannot be justified inductively without circularity, because any inductive justification of induction already presupposes UN.

C. UN cannot be rationally justified; induction rests on habit, not reason.

The skeptic's leverage is the gap between the deductive and the inductive. No amount of past observation logically constrains the future; the demand for a rational grounding runs into either logical possibility of failure (deductive route) or circularity (inductive route).

Exercise 3 (medium, short answer).

Explain why Nozick's tracking theory must deny epistemic closure. Illustrate with the brain-in-a-vat scenario.

Hint

Apply the sensitivity condition to both "I have hands" and "I am not a BIV." Do they give the same verdict?

Answer

The sensitivity condition says $S$ knows $p$ only if, were $p$ false, $S$ would not believe $p$ . For "I have hands": were you handless (in normal circumstances — an accident, an amputation), you would not believe you have hands, because the perceptual process would not deliver the belief. So the belief is sensitive — you know it.

For "I am not a BIV": were you a BIV, the vat would stimulate you identically and you would still believe you were not a BIV. So the belief fails sensitivity — you do not know it.

But "I have hands" entails "I am not a BIV." If closure held, knowing the first would give you knowledge of the second. Since tracking says you know the first but not the second, closure must fail. This is the cost of preserving ordinary knowledge.

Exercise 4 (medium, multiple choice).

Moore's "Proof of an External World" is best described as:

A. A transcendental argument proving that experience requires external objects.

B. A modus tollens that privileges common-sense certainty over skeptical premises.

C. A verificationist argument that "external world" is meaningful only if verifiable.

D. A reliabilist appeal to the reliability of perception.

Hint

Moore does not offer a deep theory; he offers a proof whose premises he claims are more certain than the skeptic's.

Answer

Option B. Moore inverts the skeptic's modus ponens. The skeptic reasons: if you cannot rule out skeptical scenarios, you do not know you have hands. Moore replies: I know I have hands; so I can rule out the scenarios. The force is not a new theory of knowledge but a claim about epistemic priority — common-sense beliefs are more secure than the philosophical premises arrayed against them.

Exercise 5 (hard, critique).

Contextualism claims to "solve" the skeptical paradox by saying the skeptic and the ordinary attributor are both right relative to their respective contexts. In 5-7 sentences, reconstruct the strongest objection to this "solution."

Hint

Does the contextualist dissolution address the skeptic's actual worry, or does it change the subject? Consider what happens when the skeptic raises the BIV possibility in your context.

Answer

The strongest objection is that contextualism changes the subject rather than answering the skeptic. The skeptic's worry is not about the word "knows" or about how attributions function in different conversations — it is about whether the subject actually has the epistemic standing the skeptic denies. When the BIV possibility is raised in a conversation, the contextualist concedes that, relative to the now-elevated standards, the subject does not know. But that is precisely the conclusion the skeptic wanted; the contextualist has granted it rather than refuted it.

A second objection is empirical: Schiffer and others argue that the alleged context-sensitivity of "knows" is not supported by the linguistic evidence. If "knows" does not actually shift meaning with context, the contextualist dissolution rests on a false semantic premise. The rival view — interest-relative invariantism (Stanley) — relocates the stakes-dependence from the word to the fact of knowing, preserving a single semantic meaning while allowing practical interests to affect whether knowledge is present.

Exercise 6 (hard, defence).

In 5-7 sentences, evaluate Putnam's semantic argument against the brain-in-a-vat scenario. Does it refute skepticism?

Hint

Putnam argues that a BIV cannot refer to trees, vats, or brains. What does this show about the BIV hypothesis, and what does it not show?

Answer

Putnam's argument (in Reason, Truth and History, 1981) applies semantic externalism: the reference of a term depends on causal contact with the kind of thing it denotes. An actual BIV — a brain that has never interacted with trees, vats, or brains in the relevant physical sense — cannot refer to trees, vats, or brains; its word "tree" refers at most to its own tree-like mental images or to the computer program generating them. So when the BIV thinks "I am a BIV," it is not thinking the thought it seems to think; the sentence is systematically disinterpreted.

Applied to us: if we are BIVs, our sentence "we are BIVs" cannot express the skeptical hypothesis, because our words lack the right referents. So the hypothesis "I am a BIV" is, if true, unstateable by me; and if stateable by me, it is false. Putnam concludes we are not BIVs.

The argument is elegant but limited. It shows that the BIV hypothesis is self-refuting in a particular semantic sense. It does not establish that we have knowledge of the external world — it establishes only that a specific global skeptical scenario is incoherent. Local skepticism (about particular beliefs, about induction, about other minds) is untouched. And some philosophers (including Stroud) argue that the argument presupposes externalism, which is itself a substantive and contestable semantic thesis. The BIV is refuted only if causal theories of reference are correct — a premise a determined skeptic may deny.

Advanced results Master

The closure debate

Epistemic closure is the fulcrum of contemporary debates about skepticism. The principle — that knowledge is closed under known entailment — is so natural that rejecting it seems to cost more than accepting it. Yet two of the most influential theories of knowledge are committed to its failure.

Dretske's "Epistemic Operators" (1970) introduced the zebra case and argued that knowledge is not closed under known entailment. You can know the animal is a zebra without knowing it is not a cleverly disguised mule, even though the first entails the second. Dretske's diagnosis: knowledge requires ruling out relevant alternatives, and the set of relevant alternatives is not closed under entailment. "Zebra" and "not a cleverly disguised mule" are evaluated against different contrast classes ^{[Dretske 1970]}.

Nozick's tracking theory (1981) independently delivers the same verdict. Sensitivity is not preserved under known entailment: a belief can be sensitive while an entailed belief is not. You can know you have hands (the belief is sensitive) without knowing you are not a handless BIV (the belief is not sensitive). Nozick embraces the denial of closure as the price of preserving ordinary knowledge against the skeptic ^{[Nozick 1981]}.

Hawthorne (Knowledge and Lotteries, 2004) defends closure and argues against Dretske and Nozick ^{[Hawthorne 2004]}. Hawthorne's lottery problem sharpens the issue: given any proposition $p$ with a high but non-unit probability, there is a lottery under which believing $p$ on that evidence is not knowledge. If knowledge is not closed, one could know $p$ , know $p \to q$ , yet fail to know $q$ — even when $q$ is entailed. This produces an unacceptable fragmentation of knowledge across deductive inference. Hawthorne prefers a version of invariantism that preserves closure but accepts that the standards for knowledge are very high — high enough that we know less than we thought.

Timothy Williamson's Knowledge and its Limits (2000) defends closure from a different direction ^{[Williamson 2000]}. Williamson's knowledge-first epistemology holds that knowledge is unanalysable and that evidence is knowledge ( $E = K$ ). On this view, closure is not a separate principle to be defended but a structural feature of knowledge itself: if $p$ is part of your evidence, and $p$ entails $q$ , then $q$ is part of your evidence. Denying closure would fragment the evidence base in ways that make rational inquiry impossible. Williamson combines this with unus multorum: there is a single knowledge relation, and the word "knows" does not shift meaning across contexts. This puts him in opposition to contextualism as well as to Dretske-Nozick.

Jonathan Kvanvig has argued that closure is preserved in ordinary knowledge attributions even if it fails in some theoretical cases. The contemporary consensus, if there is one, treats closure as default-preserving: the burden of proof is on the theorist who would deny it, and the cost of denial (fragmented knowledge, broken deductive inference) is widely judged too high.

Putnam's semantic externalism and the BIV

Hilary Putnam's argument in Reason, Truth and History (1981) attempts a semantical refutation of the brain-in-a-vat scenario ^{[Putnam 1981]}. The argument applies the causal theory of reference, on which a term refers to the kind of entity causally responsible for the speaker's uses of that term. An actual BIV — a brain that has never stood in appropriate causal contact with vats, brains, or trees — cannot use "vat," "brain," or "tree" to refer to vats, brains, or trees. At most, its words refer to its own internal states or to the generating program.

The argument then turns on a self-referential twist. Suppose I am a BIV. Then my thought "I am a BIV" does not refer to brains or vats; it refers to something else (my internal brain-like states, say). So the thought that I am a brain in a vat is not the thought I take it to be. If I am a BIV, the skeptical hypothesis is unstateable by me. If I can state it (as I appear to be doing now), I am not a BIV. Putnam concludes: we are not brains in vats.

The argument is powerful but limited. It shows that one specific global skeptical scenario is semantically self-undermining. It does not establish that we have knowledge of the external world, nor does it touch local skepticism (about particular beliefs, about induction, about other minds). Stroud has argued that the argument begs the question against the skeptic: it presupposes semantic externalism, which is a substantive philosophical thesis the skeptic may reject. If internalism about reference is correct — if what my words mean is determined by my internal states — then the BIV can refer to vats after all, and Putnam's argument fails.

Contextualism debates

Contextualism about "knows" — developed by DeRose, Lewis, Cohen, and others — holds that the truth-conditions of knowledge attributions vary with the standards of the attributor's context ^{[DeRose 1995; Lewis 1996]}. DeRose's "solving the skeptical problem" programme presents contextualism as a dissolution of the skeptical paradox: the skeptic and the ordinary attributor are both right, relative to their respective standards.

Lewis's "Elusive Knowledge" offers the most systematic version. Knowledge, for Lewis, is true belief that eliminates every possibility that is not properly ignored — subject to rules of relevance that determine what may be ignored. The most striking rule is the Rule of Attention: any possibility to which the subject attends cannot be ignored. The skeptic's move — explicitly raising the BIV scenario — makes that possibility attended-to, and so not properly ignorable. Once attended to, the BIV alternative cannot be eliminated; "you don't know" becomes true. In ordinary contexts, where the BIV is not attended to, it may be ignored, and ordinary knowledge is preserved.

The contextualist programme faces several objections. Schiffer's linguistic objection argues that the alleged context-sensitivity of "knows" is not supported by standard semantic tests. If "knows" were context-sensitive in the way contextualists claim, we should expect the kind of disagreement patterns we see with indexicals ("I am here now"), but we do not find them. Two speakers disagreeing about whether the bank depositor "knows" seem to be disagreeing about the same proposition, not talking past each other.

Stanley's interest-relative invariantism (Knowledge and Practical Interests, 2005) relocates the stakes-dependence ^{[Stanley 2005]}. It is not the word "knows" but the fact of knowing that depends on practical interests. Whether $S$ knows $p$ is sensitive to the stakes for $S$ : high stakes raise the threshold. This preserves a single semantic meaning for "knows" while accommodating the bank-case intuitions. Stanley's view is invariantist (one meaning) but subject-sensitive (stakes matter).

Rysiew's pragmatic encroachment defends a related position: the stakes-sensitivity is pragmatic rather than semantic. The proposition expressed by " $S$ knows $p$ " is invariant, but the conversational implicature varies with context, producing the illusion of semantic shift. The dispute between contextualism, interest-relative invariantism, and pragmatic encroachment turns on whether the variation occurs at the level of semantics, fact, or pragmatics.

The "How possible?" question

Stroud's The Significance of Philosophical Scepticism (1984) reframes the skeptical challenge in a way that makes externalist responses look question-begging ^{[Stroud 1984]}. Stroud argues that the skeptic is not simply making a mistake about the word "knows" or about context-relative standards. The skeptic asks a philosophical question: how is knowledge of the external world so much as possible, given that our evidence is compatible with its absence?

On Stroud's reading, the externalist responses (reliabilism, tracking, safety) do not answer this question — they change it. To say that $S$ knows $p$ when produced by a reliable process presupposes that there is a reliable process, which presupposes that there is an external world in which the process operates. But the existence of the external world is precisely what the skeptic questions. The externalist helps herself to what the skeptic denies, then announces that knowledge is unproblematic. For Stroud, this is not a refutation of skepticism but a refusal to engage with it.

Stroud's challenge has shaped the literature. The internalist tradition takes it as confirming that only an internally accessible grounding can answer the skeptic. The externalist tradition responds that the demand for an internally accessible grounding is itself the skeptic's premise — one we have independent reason to reject. The standoff is real: the two traditions disagree about what would count as an answer to the skeptical question.

Transcendental arguments

Transcendental arguments — inaugurated by Kant and revived in the analytic tradition by P. F. Strawson — attempt to answer the skeptic by showing that experience is possible only if certain conditions obtain. Strawson's Individuals (1959) argues that the identification of particulars in experience presupposes a framework of reidentifiable spatial objects: to make sense of our own experience, we must already inhabit a world of enduring things in space ^{[Strawson 1959]}.

The strategy is to move from "we must think in terms of an external world" to "there is an external world." The first step is relatively secure — the conditions of coherent experience can be argued for transcendentally. The second step is the difficulty. Stroud's critique (1968, and developed in his 1984 book) argues that transcendental arguments establish at most what we must believe, not what is true. Even if we cannot help thinking in terms of an external world, it remains logically possible that there is none — the skeptical hypothesis is not refuted, only shown to be unbelievable.

This produces what Stroud calls the "transcendental stalemate": transcendental arguments can establish the indispensability of certain concepts but cannot establish the truth of the beliefs framed in those concepts. The best they can do is show that skepticism is unlivable — that we cannot help believing in the external world even if we cannot prove it exists. Whether this counts as a refutation depends on whether the standard for refutation is truth or viability.

Skepticism about other minds

The problem of other minds asks how we can know that other human beings have inner experiences — that the people around us are not automata, philosophical zombies, or empty shells displaying outward behavior without any felt interiority. The problem is structurally similar to the BIV: each person's access to other minds is mediated by behavior, and behavior is compatible with the absence of experience.

Three responses dominate. The analogical argument holds that since my own behavior is caused by my own mental states, and since others behave similarly, they probably have similar mental states. The argument is an induction from a single case (my own), which weakens it — and the skeptic presses the point that the induction has no positive instances beyond the self.

The argument from best explanation (in the style of inference to the best explanation) holds that positing other minds is the best explanation of the complex, coordinated behavior of others. The hypothesis that others have experiences explains language, cooperation, and emotional expression better than the automaton hypothesis. This is a broadly abductive response — and like all abduction, it presupposes that explanatory virtue is a guide to truth.

Wittgenstein's later philosophy dissolves the problem rather than answering it. The private language argument (in the Philosophical Investigations) argues that the very idea of a purely private mental state, accessible only to its owner and unverifiable by anyone else, is incoherent. Meaning and mental-state attribution are governed by public criteria — by forms of life in which the concept of pain, for instance, is tied to observable behavior and learned linguistic practices. The demand for an inference from behavior to inner state misconstrues the grammar of mental concepts: we do not infer that someone is in pain, we see it in their writhing and their crying out.

Skepticism about induction and Goodman's new riddle

Nelson Goodman's Fact, Fiction, and Forecast (1954) deepens the Humean problem with the new riddle of induction ^{[Goodman 1954]}. Define a predicate "grue": an object is grue iff it is examined before some future time $t$ and is green, or is not examined before $t$ and is blue. Every observed emerald is green — and every observed emerald is equally grue (since all examined emeralds were examined before $t$ ). Why, then, should we project "green" rather than "grue"?

The puzzle is not about the logical form of induction — both projections conform to the same enumerative pattern. It is about which predicates are projectible — which are legitimately usable in inductive inference. Goodman's answer is entrenchment: a predicate is projectible to the extent that its projection is entrenched in past successful practice. "Green" is well-entrenched; "grue" is not (it is a gerrymandered predicate defined in terms of an arbitrary time $t$ ). Projectibility is not a logical property but a historical and pragmatic one.

The new riddle sharpens the Humean problem: even granting that induction is a legitimate inferential pattern, the question of which hypotheses induction supports is underdetermined without a prior account of projectible predicates. And entrenchment, as a pragmatic criterion, inherits the circularity worry — past entrenchment is a guide to future projectibility only if the future resembles the past, which is the very presupposition Hume questioned.

Moral skepticism

Moral skepticism denies that moral knowledge is available, or that moral claims are true. Two main forms dominate.

Error theory (J. L. Mackie, Ethics: Inventing Right and Wrong, 1977) holds that ordinary moral discourse purports to be objective — moral claims purport to describe stance-independent, categorical facts — but no such facts exist. Moral properties would be "queer" (Mackie's term): entities with intrinsic prescriptive force, unlike anything else in the natural world. Since the moral facts that discourse purports to track do not exist, all positive moral claims are false. Error theory is a realist view about what moral discourse claims but an eliminativist view about what exists ^{[Mackie 1977]}.

Noncognitivism holds that moral claims do not purport to state facts at all — they express attitudes. A. J. Ayer's emotivism (Language, Truth, and Logic, 1936) analyses "murder is wrong" as akin to "murder — boo!" — an expression of disapproval, not a proposition capable of truth or falsity. Charles Stevenson developed a more sophisticated version that tied moral language to both expressive and recommendation functions. Contemporary expressivism (Simon Blackburn, Allan Gibbard) refines the view: moral claims express noncognitive states but can participate in logical inference, disagreement, and truth-like discourse without stating facts.

Responses include moral realism in several forms. Cornell realism (Richard Boyd, Nicholas Sturgeon, David Brink) holds that moral properties are natural properties, identified a posteriori by the role they play in mature moral theory — analogously to how water is identified with H₂O. Non-natural realism (G. E. Moore, recently T. M. Scanlon, Derek Parfit) holds that moral properties are sui generis, irreducible to natural properties. The debate between error theory, noncognitivism, and realism turns on whether moral discourse is best modelled as fact-stating, expressive, or systematically mistaken — and on the metaethical question of what would count as vindicating one model over another.

Religious epistemology and skepticism

Religious epistemology intersects skepticism at two points. The evidentialist challenge (following Locke and Clifford) holds that belief in God is justified only if supported by sufficient evidence. Since the evidence for God's existence is (on this view) inadequate, religious belief is epistemically defective. The challenge is a form of skepticism about religious knowledge.

Reformed epistemology (Alvin Plantinga, developing themes in Calvin and Aquinas) denies that evidence is required. Plantinga argues that belief in God can be properly basic — warranted without inferential support — when produced by a functioning sensus divinitatis in the appropriate cognitive environment ^{[Plantinga 2000]}. This extends externalism to religious belief: the believer need not be able to prove God exists, any more than the perceiver need prove the external world exists. The warrant flows from the proper functioning of the relevant faculty, not from the subject's reflective access.

J. L. Schellenberg's divine hiddenness argument poses a distinctively skeptical challenge ^{[Schellenberg 1993]}. If a loving God exists, reasonable non-resistant non-belief should not occur — a loving God would make his existence manifest to anyone open to relationship with him. Yet reasonable non-belief does occur. Therefore, a loving God does not exist. The argument turns the absence of evident divine presence into evidence against God's existence — a skeptical challenge pitched at the level of divine self-disclosure rather than human cognition.

Evolutionary debunking arguments

Evolutionary debunking arguments (EDAs) use facts about the evolutionary origins of our beliefs to undermine their claim to truth. The structure:

P1. Our belief-forming faculties regarding domain $D$ were shaped by evolutionary pressures directed at fitness, not truth.

P2. If evolution selected for fitness rather than truth in domain $D$ , then the fact that we have the beliefs we do is insensitive to whether those beliefs are true.

C. We have no warrant for our $D$ -beliefs — their reliability is undercut by their genealogy.

Sharon Street's evolutionary debunking of moral realism (2006) applies this to morality ^{[Street 2006]}. Our moral beliefs were shaped by selection pressures (kin altruism, reciprocal altruism, group cohesion) that are fitness-tracking, not truth-tracking. On moral realism — which holds that moral facts are stance-independent — the evolutionary explanation of our moral beliefs is independent of whether those beliefs are true. So the reliability of our moral beliefs is left unexplained, and the beliefs are debunked.

Richard Joyce (The Evolution of Morality, 2006) argues more strongly that evolution gives us no reason to think moral beliefs are true, and some reason to think the very concept of categorical moral reasons is a useful fiction. Responses include Copp (moral truths supervene on natural truths that evolution does track), Wielenberg (ethical truths are necessary truths, like mathematical truths, and evolution selects for the cognitive capacities that grasp them), and Brosnan (the debunking argument proves too much — it would equally debunk mathematics and science).

Plantinga's Evolutionary Argument Against Naturalism (EAAN) generalises the worry ^{[Plantinga 2011]}. If both naturalism and evolution are true, then our cognitive faculties were selected for fitness, not truth. The probability that such faculties are reliable — Plantinga argues — is low (he suggests below one-half). But any argument we give for naturalism and evolution presupposes the reliability of our cognitive faculties. So the conjunction of naturalism and evolution undercuts the rationality of believing it. Critiques (O'Connor; Fitelson and Sober) target Plantinga's probability assignment and argue that truth-tracking faculties are fitness-enhancing, so evolution would select for them. The general question — does knowing our cognitive faculties evolved undermine their reliability? — remains one of the liveliest intersections of philosophy of biology and epistemology.

Connections Master

Connection to 20.01.01 — Knowledge, justification, and truth

The skeptical challenge presupposes the analysis of knowledge developed in 20.01.01. The JTB analysis and its Gettier successors (tracking, safety, causal theories) are the resources against which skeptical arguments are evaluated: Nozick's tracking theory was developed partly as a response to skepticism, and Sosa's safety condition handles both Gettier cases and skeptical scenarios in a unified framework. The closure debates that organise this unit inherit the Gettier-era search for a fourth condition on knowledge.

Connection to 20.01.02 — Theories of justification

Theories of justification frame the skeptical challenge directly. Foundationalism faces the skeptic's demand for indubitable foundations — the Cartesian project of finding a bedrock that resists radical doubt. Coherentism faces the isolation objection: a coherent fiction is as internally supported as a true account, which the skeptic exploits. Reliabilism faces the new evil demon problem — an envatted brain with internally identical processes is unreliable in its world yet seems as justified as we are. The theory of justification one adopts largely determines which skeptical pressures are most threatening.

Connection to 20.08.NN — Philosophy of science: induction and underdetermination

Hume's problem of induction and the new riddle of induction (Goodman) are central to both epistemology and philosophy of science. The Duhem-Quine thesis (a hypothesis cannot be tested in isolation) is a structural analog of the skeptical underdetermination argument: the evidence underdetermines theory choice, just as the experiences underdetermine the BIV hypothesis. Bayesian confirmation theory, covered in the philosophy of science curriculum, provides the formal substrate for contemporary responses to Hume.

Connection to 25-logic and 42-mathematical-logic — Epistemic and modal logic

The closure principle and the counterfactual analyses of sensitivity and safety require the formal apparatus of modal logic. Sensitivity ( $\neg p □ \to \neg B_{S} p$ ) is a counterfactual conditional in the tradition of Lewis and Stalnaker; safety is a universal quantification over nearby possible worlds. Epistemic logic (Hintikka) provides the operator $K_{S} p$ and the possible-worlds semantics that make closure precise. A formal treatment of the skeptical argument requires the tools of §25 (modal logic) and §42 (advanced modal and epistemic logic).

Connection to 20.02.NN — Ethics: moral skepticism and metaethics

Moral skepticism (error theory, noncognitivism, expressivism) and the evolutionary debunking arguments covered in this unit connect directly to the ethics curriculum. The metaethical debates about realism, anti-realism, and quasi-realism provide the background against which moral epistemology operates. Plantinga's EAAN and Street's evolutionary debunking argument raise questions that span epistemology, ethics, and philosophy of biology.

Historical and philosophical context Master

Skepticism is among the oldest philosophical positions. The ancient tradition divides into two schools. Pyrrhonism traces its origin to Pyrrho of Elis (c. 360-270 BCE), who accompanied Alexander the Great's campaigns to India and is reported to have been impressed by the gymnosophists — naked ascetic philosophers whose suspension of judgment Pyrrho adapted. Pyrrho wrote nothing; his thought was systematised three centuries later by Aenesidemus and, most fully, by Sextus Empiricus (c. 160-210 CE), whose Outlines of Pyrrhonism (Pyr/rhoneioi hypotyposeis) is the surviving classic of the school. Sextus presents Pyrrhonism as a therapeutic practice: by setting opposing arguments in balance (isostheneia) and suspending judgment (epoché), the Pyrrhonian attains tranquility (ataraxia). The Ten Modes of Aenesidemus and the Five Modes of Agrippa codify the patterns of argument by which the Pyrrhonian induces suspension — including the regress mode, which would become Agrippa's trilemma in the modern literature ^{[Sextus Empiricus]}.

Academic skepticism developed in Plato's Academy under Arcesilaus (c. 315-240 BCE) and Carneades (c. 214-129 BCE). Where the Pyrrhonians suspended judgment on whether knowledge is possible, the Academics argued positively that it is not — while allowing that some beliefs are more persuasive (pithanon) than others and can guide action. The dispute between Pyrrhonian and Academic skepticism, reconstructed by Cicero in the Academica, set the terms of debate for the next millennium.

The skeptical tradition lay largely dormant in the medieval period, though Augustine's Contra Academicos (386 CE) engaged the Academic position. The modern revival came with Michel de Montaigne's Apologie de Raymond Sebond (1575), which drew on Sextus to argue for the limits of human reason. Montaigne's influence on Descartes was direct and acknowledged: Descartes' Meditations on First Philosophy (1641) was written partly in response to the Montaignean skeptical challenge ^{[Descartes 1641]}.

Descartes' Meditations gave skepticism its modern form. The method of doubt — treating every belief that could conceisbly be false as if it were false — was designed to find an indubitable foundation. The dream argument (Meditation I) and the evil demon hypothesis (Meditation I, developed in Meditation III) target perception and mathematics respectively. The cogito ("I think, therefore I am") was the bedrock: a belief whose truth is guaranteed by the act of thinking it. The further project — rebuilding knowledge from the cogito outward — depended on proving God's existence and veracity, which would guarantee the reliability of clear and distinct ideas. The proof of God in Meditation III and the causal proof in Meditation V were immediately controversial. The Cartesian Circle — the charge that Descartes uses the reliability of clear and distinct ideas to prove God, then uses God to guarantee the reliability of clear and distinct ideas — was raised by Arnauld in the Fourth Objections and has haunted the Meditations ever since ^{[Descartes 1641]}.

The British empiricists inherited and worsened the skeptical pressure. Locke's Essay (1689) grounded knowledge in ideas of sensation and reflection, but the gap between ideas and the external world opened by Locke's representative theory of perception invited skeptical exploitation. Berkeley closed the gap by denying material substance — esse est percipi — but at the cost of idealism. Hume pushed empiricism to its skeptical conclusion. The Treatise of Human Nature (1739-40) and the Enquiry Concerning Human Understanding (1748) argued that causal inference rests on habit, that induction lacks rational foundation, and that the self is a bundle of perceptions rather than a substance. Hume's skepticism was not merely theoretical — it was a personal crisis, described in the Treatise's conclusion as a "philosophical melancholy" from which Hume rescued himself only by leaving the study and socialising ^{[Hume 1748]}.

Kant's Critique of Pure Reason (1781) was the great attempt to answer Hume. Kant argued that knowledge requires both sensory input (the matter of experience) and conceptual form (the categories — causality, substance, unity). The mind contributes a priori structures that organise experience; without them, experience would be a "rhapsody" of unconnected sensations. Synthetic a priori knowledge — knowledge that is informative yet independent of experience — is possible because the mind's own contribution can be known independently of particular experiences. The Kantian synthesis defined the terms of epistemological debate for the next century and a half, and the transcendental method — arguing from the possibility of experience to the conditions that make it possible — became a permanent resource in anti-skeptical argument ^{[Strawson 1959]}.

The twentieth century reopened the skeptical question with new tools. G. E. Moore's "Proof of an External World" (1939) and "A Defence of Common Sense" (1925) offered the common-sense response: the premises of common sense are more certain than the philosophical principles deployed against them. Moore did not refute the skeptic in the sense of proving the external world exists from neutral premises; he argued that no neutral premises are available, and that the skeptic's demand for them is itself question-begging.

The post-war period saw the development of the formal apparatus that organises contemporary debate. Dretske's relevant alternatives (1970), Goldman's reliabilism (1976), Nozick's tracking theory (1981), and Putnam's semantic externalism (1981) appeared within a single decade — a remarkable concentration that reshaped the field. The 1980s and 1990s brought the contextualist programme (DeRose 1992, Lewis 1996), Stroud's reconstructive work (1984), and Sosa's virtue epistemology (1991 onward). The 2000s saw the closure debate renewed by Hawthorne (2004) and Williamson (2000), the pragmatic-encroachment literature (Stanley 2005, Fantl and McGrath 2009), and the application of evolutionary arguments to moral and religious epistemology (Street 2006, Plantinga 2011).

The contemporary state of the field is one of structured disagreement. No single response to skepticism commands consensus, but the leading positions are now precisely characterised: Moorean, relevant-alternatives, tracking, safety, contextualist, pragmatic-encroaching, transcendental, and semantic-externalist. The skeptical challenge remains, as it has for two and a half millennia, the stress-test against which every theory of knowledge is measured.

Bibliography Master

Sextus Empiricus. (c. 200 CE). Outlines of Pyrrhonism (Pyrrhoneioi hypotyposeis). English trans. J. Annas and J. Barnes, Cambridge University Press (2000).
Cicero. (45 BCE). Academica. English trans. C. Brittain, Oxford University Press (2006).
Augustine. (386 CE). Contra Academicos. English trans. D. J. Kavanagh, in Against the Academicians and The Teacher, Hackett (1995).
Montaigne, M. de. (1575). "Apologie de Raymond Sebond." In Essais, Book II, Ch. 12.
Descartes, R. (1641). Meditationes de Prima Philosophia. English trans. J. Cottingham, Meditations on First Philosophy, Cambridge University Press (1996).
Locke, J. (1689). An Essay Concerning Human Understanding.
Berkeley, G. (1710). A Treatise Concerning the Principles of Human Knowledge.
Hume, D. (1739-40). A Treatise of Human Nature.
Hume, D. (1748). An Enquiry Concerning Human Understanding.
Kant, I. (1781). Critik der reinen Vernunft. English trans. P. Guyer and A. Wood, Critique of Pure Reason, Cambridge University Press (1998).
Russell, B. (1912). The Problems of Philosophy. Oxford University Press.
Moore, G. E. (1925). "A Defence of Common Sense." In Contemporary British Philosophy, 2nd series. Reprinted in G. E. Moore, Philosophical Papers, Allen and Unwin (1959).
Moore, G. E. (1939). "Proof of an External World." Proceedings of the British Academy 25, 273-300.
Ayer, A. J. (1936). Language, Truth, and Logic. Gollancz.
Reichenbach, H. (1949). The Theory of Probability. University of California Press.
Wittgenstein, L. (1953). Philosophical Investigations. Blackwell.
Goodman, N. (1954). Fact, Fiction, and Forecast. Harvard University Press.
Strawson, P. F. (1959). Individuals: An Essay in Descriptive Metaphysics. Methuen.
Sellars, W. (1956). "Empiricism and the Philosophy of Mind." In Minnesota Studies in the Philosophy of Science, vol. 1, pp. 253-329.
Hintikka, J. (1962). Knowledge and Belief: An Introduction to the Logic of the Two Notions. Cornell University Press.
Mackie, J. L. (1977). Ethics: Inventing Right and Wrong. Penguin.
Goldman, A. I. (1976). "Discrimination and Perceptual Knowledge." The Journal of Philosophy 73, 771-791.
Dretske, F. (1970). "Epistemic Operators." The Journal of Philosophy 67, 1007-1023.
Dretske, F. (1981). "The Pragmatic Dimension of Knowledge." Philosophical Studies 40, 363-378.
Putnam, H. (1981). Reason, Truth and History. Cambridge University Press.
Nozick, R. (1981). Philosophical Explanations. Harvard University Press.
Sosa, E. (1991). Knowledge in Perspective: Selected Essays in Epistemology. Cambridge University Press.
Stroud, B. (1968). "Transcendental Arguments." The Journal of Philosophy 65, 241-256.
Stroud, B. (1984). The Significance of Philosophical Scepticism. Oxford University Press.
DeRose, K. (1995). "Solving the Skeptical Problem." The Philosophical Review 104, 1-52.
Lewis, D. (1996). "Elusive Knowledge." Australasian Journal of Philosophy 74, 549-567.
Blackburn, S. (1993). Essays in Quasi-Realism. Oxford University Press.
Gibbard, A. (1990). Wise Choices, Apt Feelings. Harvard University Press.
Boyd, R. (1988). "How to Be a Moral Realist." In Essays on Moral Realism, ed. G. Sayre-McCord, pp. 181-228. Cornell University Press.
Brink, D. (1989). Moral Realism and the Foundations of Ethics. Cambridge University Press.
Williamson, T. (2000). Knowledge and its Limits. Oxford University Press.
Schellenberg, J. L. (1993). Divine Hiddenness and Human Reason. Cornell University Press.
Hawthorne, J. (2004). Knowledge and Lotteries. Oxford University Press.
Stanley, J. (2005). Knowledge and Practical Interests. Oxford University Press.
Street, S. (2006). "A Darwinian Dilemma for Realist Theories of Value." Philosophical Studies 127, 109-166.
Joyce, R. (2006). The Evolution of Morality. MIT Press.
Plantinga, A. (2000). Warranted Christian Belief. Oxford University Press.
Plantinga, A. (2011). Where the Conflict Really Lies: Science, Religion, and Naturalism. Oxford University Press.
BonJour, L. (2010). Epistemology: Classic Problems and Contemporary Responses, 2nd ed. Rowman and Littlefield.
Fantl, J. and McGrath, M. (2009). Knowledge in an Uncertain World. Oxford University Press.
Copp, D. (2008). "Darwinian Skepticism about Moral Realism." Philosophical Issues 18, 186-206.
Wielenberg, E. J. (2010). "On the Evolutionary Debunking of Morality." Ethics 120, 441-464.
Sosa, E. (1999). "How to Defeat Opposition to Moore." Philosophical Perspectives 13, 141-153.