A Review of Williamson's 'Knowledge and Its Limits' — Epoche B2
Anti-Luminosity, Not Unanalysability: Reading Williamson's Knowledge and Its Limits What the book is usually said to say Timothy Williamson's Knowledge and Its Limits (2000 [1] ) is normally introduced with a single sentence: knowledge cannot be analysed into belief, truth and further conditions, so we should start from knowledge instead. That is accurate as far as it goes, and it makes the book sound like a refusal — as though, after four decades of failed analyses, Williamson had proposed giving up. This review argues that the summary picks the wrong load-bearing part. Unanalysability is a conclusion Williamson draws, not an argument he offers, and by itself it would be weak: the failure of a research programme is evidence against it, but not much. What makes the book more than a report of failure is the anti-luminosity argument [2] . That argument does not say the analysis of knowledge is hard [3] ; it says the picture that made the analysis look mandatory is false. Everything distinctive in the rest of the book — the identification of evidence with knowledge, the knowledge norm of assertion, the treatment of scepticism — depends on it. A reader who takes away only "knowledge is primitive" has taken away the part that could most easily be denied. The programme Williamson inherited The setting is familiar. The tripartite analysis holds that a person knows a proposition when they believe it, it is true, and they are justified in believing it [4] . Edmund Gettier showed in 1963 that the three conditions can be met without knowledge [5] : a justified belief can be false, a truth can be validly inferred from it, and the resulting belief is true, justified, and plainly not knowledge. What followed was a thirty-year search for a fourth condition, or for a repaired third one. Some proposals were internalist, adding requirements on the believer's grounds: that no false lemma be used, that no true proposition would defeat the justification. Others were externalist, most influentially Alvin Goldman's suggestion in "What Is Justified Belief?" (1979) that the belief be produced by a reliable process. Every proposal met a counterexample, and by the 1990s the pattern was itself the topic — Linda Zagzebski argued that counterexamples of Gettier's shape can be generated for any analysis that pairs a truth condition with a fallible condition of another sort [6] . The knowledge-first reversal Williamson's diagnosis is that the programme rested on an unstated picture. On that picture the mind has an interior, and belief lies inside it; truth and reliability lie outside; knowledge is a composite, made by adding external conditions to an internal state. The analysis is then a matter of finding the right glue. He denies the picture. Knowing, he argues, is itself a mental state — the most general factive one. Where seeing that it is raining and remembering that it is raining each entail that it is raining, knowing is the weakest state that does so, and the others are special cases of it. If knowing is a state in its own right rather than a compound, there is no reason to expect it to decompose, and the failure of the programme is explained rather than merely recorded. Stated this way, the reversal is a proposal, and an opponent may simply decline it. What forces the issue is the next argument. The anti-luminosity argument Call a condition luminous when, necessarily, if it obtains one is in a position to know that it obtains. The internalist picture requires that a good many conditions be luminous: feeling cold, believing something, having such-and-such seem to be the case. These are supposed to constitute a cognitive home, a region where being in a state and being able to tell that you are in it come to the same thing. Williamson's argument is a sorites. You feel cold at dawn and hot by noon, and the change is gradual. Divide the interval into instants a millisecond apart. At the first you feel cold; at the last you do not. Now add a constraint on knowledge: knowledge requires a margin for error. Where your evidence is a matter of degree and your discriminations are imprecise, you cannot know that a condition obtains if it only just obtains, because you could very easily have been in an almost identical state in which it did not, and would have believed the same thing. Combine the two. Suppose feeling cold is luminous. Then at any instant at which you feel cold, you are in a position to know it. But if you know it at one instant, the margin-for-error constraint requires that you still feel cold a millisecond later, since your state then is almost indistinguishable. So feeling cold at any instant implies feeling cold at the next. Run this along the series and you get: you feel cold at dawn, therefore you feel cold at noon. That is false. So feeling cold is not luminous. Nothing in the argument turns on temperature. It generalises to any condition that can obtain to a greater or lesser degree and can change gradually, which is to say to almost every condition of interest. The cognitive home is uninhabited. Why this is the argument that carries the book Take the anti-luminosity result seriously and the rest follows more easily. If no interesting internal condition is transparently available, then the internalist's motivation for building knowledge out of such conditions is gone; the analysis was not merely hard but pointless. That is a stronger conclusion than induction from past failures. It also underwrites the book's most-quoted thesis, the identification of evidence with knowledge. If evidence were a class of internal states we could always tell we were in, it would make sense to treat evidence as the base and knowledge as what is built on it. Once that is denied, the question of what one's evidence is becomes as substantive as the question of what one knows, and Williamson's proposal — that one's total evidence is one's total knowledge — is no longer paradoxical. It has real consequences: two people in the