Skepticism, Certainty, and Closure
The problem
Can I know there is a computer in front of me? The Cartesian challenge says no, or at least not with certainty, because at least three scenarios are compatible with exactly the experience I'm having right now while the computer is absent: I could be dreaming, I could be the victim of a deceiver, or I could be a brain in a vat (BIV) fed a simulated stream of experience. These scenarios are epistemically seamless — by construction, nothing available to me from the inside distinguishes them from the ordinary case. If knowing requires ruling out every scenario incompatible with what I believe, then the mere availability of these alternatives seems to cost me knowledge of the computer.
The three alternatives are not quite interchangeable. The deceiver and the BIV hypothesis are phenomenologically identical, but Hilary Putnam argued they may not be semantically identical: under an externalist theory of reference, a real brain in a vat's word "vat" would never have been causally hooked up to actual vats, so the sentence "I am a brain in a vat," asserted from inside the scenario, would be false or meaningless. That gives the BIV hypothesis a self-undermining flavor the demon hypothesis doesn't obviously share, even though the two feel identical from within.
Why probability doesn't rescue the situation
A natural response is to say the skeptical scenarios are merely improbable. This doesn't work as cleanly as it sounds. Assigning any credence — even a low one — to "I am a BIV" requires trusting the very faculties (evidence-weighing, sense of likelihood, judgments of simplicity) that the scenario itself puts in question. Using those faculties to certify their own reliability is circular. This is why Descartes did not attempt a probabilistic response to the demon: an omnipotent deceiver could rig judgments of likelihood as easily as judgments of fact, so no discount, however small, is safe. Contemporary attempts to assign the BIV hypothesis a low prior via simplicity or parsimony run into the same wall one level up, since the truth-tracking status of parsimony itself is exactly what a thoroughgoing skeptic can question. The honest diagnosis is not that the probability of being deceived is high, low, or unknowable — it's that probability-talk presupposes a stable background the global hypothesis is specifically built to undermine, so the question is closer to a category error than an open empirical matter.
Moore's proof and where it strains
G. E. Moore's 1939 "proof of an external world" runs: here is one hand, here is another, therefore at least two external objects exist, therefore an external world exists. The inference is valid, and the premises are things Moore claims to know with more certainty than he could ever have for any skeptical premise denying them.
The proof strains in three connected places. First, classical sense-datum epistemology (which Moore himself helped build, alongside Russell) holds that what's immediately given in perception is appearance, and that material objects are inferred as the probable cause of that appearance. Moore's proof skips this inferential step entirely — "here is a hand" is simply asserted as a premise, which only works if you already accept a rival, direct-realist epistemology on which perception itself delivers non-inferential knowledge of physical objects. The proof never argues for that epistemology; it assumes it. Second, and relatedly, the proof is dialectically question-begging: the skeptic doesn't dispute the validity of the inference, only the premise, so simply asserting the premise with confidence doesn't move the argument forward for anyone who doubted it in the first place.
Third, and most precisely, there is an equivocation in the word "certain." Psychological certainty is maximal confidence — more conviction than in any rival claim, with no live doubt. Epistemic or logical certainty is indefeasibility — the impossibility of error given one's grounds. The skeptic's challenge targets the second kind; Moore explicitly concedes elsewhere that he isn't claiming logical impossibility of error, only that doubting the hand is less reasonable than doubting the skeptical premise. That is a comparative, psychological claim wearing the vocabulary of a metaphysical one. This maps onto a real fork in epistemology between infallibilist knowledge (grounds must entail truth — the classical, Cartesian standard) and fallibilist knowledge (good-enough grounds suffice even though error remains logically possible — the mainstream contemporary view). Moore's ordinary perceptual knowledge is fallibilist knowledge, which is perfectly good knowledge on the majority view, but it was never going to supply the indefeasibility the skeptical challenge specifically asks for. Demanding indefeasibility consistently, it's worth noting, doesn't just sink Moore's hand — it sinks essentially all empirical knowledge, since no perceptual or inductive belief is ever logically guaranteed by its evidence.
Wittgenstein's reclassification
In On Certainty, Wittgenstein argues that "here is a hand" isn't known, in this context, in either an inferred or a directly-evidenced sense — it functions as a hinge: not a move within the practice of doubting and justifying, but part of what makes that practice possible at all, the way the hinges of a door aren't among the things that move when the door opens. Hinge status isn't a fixed property of the sentence; the same words are an ordinary, checkable claim in an anatomy class and a hinge when deployed against radical doubt — what changes is the role the sentence is playing, not its content.
This is a related but distinct phenomenon from Wittgenstein's remarks on first-person avowals like "I am in pain," where "I know" misfires for a different reason: there is no evidential gap at all between having a sensation and being disposed to report it, so there's nothing for "know" to add. Hinge propositions about the external world misfire because they are the shared, inherited backdrop that empirical inquiry runs on, not because they are avowals. Both are cases where "know" fails to get traction, for different underlying reasons.
Applied to Moore: Wittgenstein vindicates the instinct that no inference from sense-data is needed, while denying Moore's own description of what he has as knowledge held with certainty. Both halves of that description borrow grammar that doesn't belong to hinge propositions, which are neither known nor unknown in the ordinary sense.
The closure principle and its discontents
The skeptical argument's real logical engine is the closure principle: if you know P, and know that P entails Q, you know Q. Applied here: knowing you have a hand entails knowing you are not a handless brain in a vat; you don't know the latter; therefore, by modus tollens, you don't know the former.
Robert Nozick's tracking theory denies closure rather than accepting the conclusion. On this view, knowledge requires that a belief track the truth: roughly, in the nearest worlds where P is false you wouldn't believe P, and in the nearest worlds where P is true you would. "I have a hand" tracks the truth this way — if you lacked a hand, you wouldn't believe you had one. "I am not a BIV" does not track anything — in the nearest world where you are a BIV, you would believe exactly what you believe now, since the vat is built to produce identical conviction either way. Same entailment relation, opposite tracking status, so closure simply fails for this pair without contradiction. Relevant-alternatives and contextualist accounts (Dretske, Lewis, DeRose) reach a similar place by a different route: knowing P only requires ruling out the alternatives that are live in the context of the claim, and the BIV hypothesis isn't live in ordinary contexts — it only becomes relevant once a conversation like this one raises it. Moore's own move can be read as the closure-preserving alternative: run the inference in the other direction — since I know I have a hand, and that entails I'm not a BIV, I know I'm not a BIV — trading the same premise for the opposite conclusion.
None of this is free. Closure is not an arbitrary rule; it's close to what valid deductive extension of knowledge is supposed to be, and denying it has been attacked (by Williamson among others) as costing at least as much as accepting skepticism. The honest picture is a genuine fork: keep closure and accept the skeptical conclusion, or drop closure and keep ordinary knowledge, with serious epistemologists on both sides and no cost-free third option.
A related worry, from Timothy Williamson's "Cognitive Homelessness," generalizes the circularity noted above about probability: the skeptic's aspiration to a neutral, presupposition-free standpoint from which to evaluate all belief-forming methods at once is unavailable to any thinker, because evaluating a method already requires using some method. The skeptic ends up demanding a vantage point that doesn't exist for anyone, including the skeptic.
Where this leaves things
There are two historically serious resting points, not one. The first is to adopt a substantive epistemology — denying closure, restricting relevant alternatives, defending direct realism — that blocks the skeptical inference at some specific joint, at a real and acknowledged cost. The second, associated with Hume and, in a different register, with Wittgenstein, is to notice that the skeptical arguments, however hard to refute on their own terms, don't and structurally can't alter the framework within which ordinary belief, evidence, and action already operate. Asking "what difference does it make" is not a way of dodging the puzzle; it is one of the field's two real endpoints, alongside the technical machinery of tracking, contextualism, and closure-denial. Neither resolves the question so much as names, precisely, where argument runs out.