Williamson's Knowledge and Its Limits: Knowledge-First Epistemology

Imagine that a colleague asks whether a report has been submitted. You say yes. Perhaps you watched the successful submission and checked the receipt. Perhaps you merely hoped someone had done it. In both versions the report really has been submitted, and you believe that it has. Yet your colleague has good reason to care which version occurred. The truth of your answer does not settle whether you knew it.

One philosophical approach starts with that contrast and asks what must be added to true belief to produce knowledge. Timothy Williamson's Knowledge and Its Limits proposes reversing the investigation. Knowledge should help us understand belief, evidence, and justification, rather than receive its definition from a combination of those concepts. This is the central proposal of knowledge-first epistemology. The book was first published in 2000; the Oxford online edition is dated 2002. Williamson's Oxford bibliography and the publisher's account of the book distinguish those reference points.

The reversal is demanding because knowledge seems harder to establish than belief. You can sincerely report your conviction and still be wrong about whether you know. Williamson accepts that difficulty. His project asks whether a good theory of knowledge must begin with a state whose presence we can always recognize. The book's account of the limits of self-knowledge challenges that requirement.

The familiar proposal is that knowing something involves believing it, its being true, and having justification for believing it. Justification concerns the reasons or grounds supporting a belief. A correct answer reached through a good observation differs from an equally correct answer produced by wishful thinking. Philosophers seeking an analysis want conditions that capture the difference without simply repeating the word knowledge.

An invented example shows why the project is difficult. You check an office calendar that normally records meetings accurately. It says that the director is in Singapore. Unknown to you, today's entry was copied from an old schedule. The director happens to be in Singapore for a different meeting. Your belief has a respectable-looking basis and is true, but its truth results from an accident disconnected from the entry you consulted. This example does not settle exactly which condition knowledge requires. It makes the problem of accidental correctness visible.

Williamson's response is methodological. Repeated trouble finding a satisfactory analysis gives us reason to investigate a different starting point; it does not constitute a deductive proof that every possible analysis must fail. On his approach, knowing is a fundamental epistemic condition whose relations to other concepts can be explained without reducing it to independently understood ingredients. The book's overall abstract presents that reversal explicitly.

There is a useful distinction between giving an analysis and giving an explanation. A doctor can explain how an ability develops, when it fails, and what it enables without defining it through a short list of necessary and sufficient conditions. The analogy does not prove Williamson right. It shows why abandoning a reductive definition need not end inquiry. The question becomes whether starting with knowledge produces better explanations elsewhere.

Williamson treats knowing as a mental state. Its being factive means that knowing that something is the case entails its truth. If you know the report was submitted, it was submitted. Believing the report was submitted does not have that implication. Knowledge and belief therefore differ even when confidence feels the same from the inside.

His claim is stronger than the ordinary observation that knowledge requires truth. The proposed mental state itself involves the world being a certain way. A person who knows that the door is open and a deceived person who merely believes that it is open can differ mentally, in this broad sense, despite having indistinguishable experiences. This is an externalist proposal about mental states: what state you occupy can depend partly on circumstances beyond your skin. The publisher's summary identifies this extension of externalism as a central commitment.

Consider another hypothetical pair. One worker sees a genuine submission receipt. Another sees a perfect-looking forgery and has no reason to suspect it. Both may describe their experience identically. Williamson's proposal allows the first worker's knowing to matter to the explanation of what happens next. The second worker's convincing experience remains psychologically important, but it does not establish the same relation to the submitted report.

The point does not require calling the deceived worker foolish. An account of what someone knows differs from an assessment of whether they are blameworthy. Nor does it show that everything mental lies outside the head. It challenges the requirement that every genuine mental state be completely determined by what the person and their environment look like from a strictly internal perspective. Whether that challenge succeeds remains a philosophical question.

Chapter 9 defends the equation commonly written as E = K: a subject's total evidence consists of the propositions that subject knows. This concerns the subject's body of evidence, not the claim that every known proposition supports every hypothesis. Williamson also distinguishes what belongs to one's evidence from the relation of its supporting some further proposition. His discussion appears in “Evidence,” especially sections 9.1–9.3.

Return to the report. You may know that the receipt displays a particular reference number, that the submission page opened, and that your colleague received a notification. These propositions can support a further conclusion about successful submission. A confident but false belief that the report was accepted cannot itself belong to your evidence under E = K. A different true proposition, that the screen displayed an acceptance message, might belong to your evidence if you know it.

That distinction keeps the proposal from demanding that inquiry begin with certainty about every conclusion. We can know observations while remaining uncertain about what explains them. We can also use a known fact about probabilities as evidence without knowing which uncertain outcome will occur. An investigator can reason from what is known while exploring a hypothesis that remains unproved.

The controversial step is identifying all evidence with knowledge. A critic can ask whether a careful but deceived observer has evidence for a false conclusion. Williamson's framework permits that observer to know some things, including facts about appearances, while denying that every apparent fact belongs to their evidence. An internalist critic may regard this as giving too little weight to the observer's perspective. The disagreement concerns what evidence is, rather than whether deception can be convincing.

Chapter 11 proposes that assertion has a knowledge rule: assert a proposition only if you know it. Assertion is the act of putting something forward as true. The rule is a proposed constitutive standard for that act, rather than a claim that speakers always obey it. The chapter abstract and section list distinguish the knowledge account from accounts requiring only truth or reasonable belief.

Our report example separates those standards. A person who confidently guesses that submission succeeded might tell the truth without knowing it. If asked for a basis, admitting that it was a guess changes how the listener should use the answer. A qualified prediction, an explicit supposition, and an unqualified assurance make different contributions to a conversation. The knowledge rule helps explain why those differences matter.

The rule supplies a necessary condition, not a complete permission to speak. Knowing a private fact does not settle whether disclosing it would be considerate or lawful. Conversely, a sincere mistake can violate a standard without making the speaker dishonest or deserving severe blame. These distinctions prevent an epistemic account of assertion from becoming an indiscriminate moral verdict on every mistaken sentence.

A lottery sharpens the issue. Knowing that one ticket has a very low chance of winning seems different from knowing that it will lose. Williamson uses lottery cases in arguing against weaker assertion rules. The relevant contrast is between a probability claim and an outright claim about the outcome, rather than between a mathematical probability of one and every other probability. His account does not replace ordinary knowledge with absolute certainty.

The book's anti-luminosity argument concerns an apparently secure kind of access to one's own mind. A condition is luminous if, whenever it obtains, the person is in a position to know that it obtains. Being in a position to know differs from actually knowing: someone might never have considered the question. The issue is whether attention could always remove the ignorance. See chapter 4, sections 4.1–4.3.

Williamson asks us to imagine feeling cold at dawn, gradually warming up, and no longer feeling cold by noon. Throughout the process the person carefully attends to how they feel. The changes are so slow that successive moments differ too little for the person to discriminate reliably. Feeling cold seems a promising candidate for luminosity: surely, whenever you feel cold, you could recognize that you do.

The argument adds a reliability requirement. In this carefully described process, knowing that you feel cold at one moment requires enough margin for error that you still feel cold at the next, extremely similar moment. Otherwise almost unchanged confidence would be mistaken immediately nearby. If feeling cold always put the attentive person in a position to know it, feeling cold at one stage would yield knowledge at that stage, and the margin requirement would carry feeling cold into the next stage.

Repeat that reasoning and it carries feeling cold all the way to noon, contradicting the stipulated endpoint. Williamson rejects the luminosity assumption. This is a reconstruction of his argument, not an empirical finding about a measured temperature threshold. Its force depends on the margin premise as well as the described continuity of the process. Rejecting that premise is a substantive way to resist the conclusion; declaring introspection obviously transparent simply restates what the argument challenges.

Chapter 5 applies related reasoning to the principle often called KK: if a person knows a proposition, that person knows that they know it. Williamson argues against even a suitably restricted guarantee of being in a position to achieve that further knowledge. Higher-order knowledge is an additional achievement. “Margins and Iterations,” section 5.1 develops the argument through inexact visual judgments about a tree's height.

An ordinary illustration helps distinguish the levels. You can recognize that a distant tree is taller than a chair without identifying its height to the nearest centimetre. Moving from a reliable broad judgment to an exact account of which judgments your eyesight supports requires more information. The illustration does not identify a specific KK failure by itself. It shows why the extra question about the limits of your knowledge deserves separate attention.

Margins for error belong to a family of ideas about reliability, safety, and the ease of making a mistake. They should not be converted into the requirement that a proposition remain true in every vaguely similar possible world, regardless of how the belief was formed. Circumstances, the basis of judgment, and the relevant possibilities matter. Nor does adding a safety condition automatically supply a complete analysis of knowledge; that would reverse the methodological direction Williamson is exploring.

The conclusion is compatible with much self-knowledge. It denies a universal guarantee. Someone may often know that they know a colleague's name while lacking the same assurance about a borderline perceptual judgment. This also differs from overconfidence, which involves taking oneself to know something one does not know. KK failure concerns knowledge that is present without the further knowledge. Treating the two phenomena as identical would obscure the argument.

On Williamson's world-involving account, a successful perceiver and a deceived counterpart can have indistinguishable experiences while differing in what they know. Given that possibility and E = K, their bodies of evidence can differ. Nontransparency alone does not establish the possibility of such experiential duplicates. This challenges a skeptical argument that grants exactly the same evidence to both subjects. Williamson disputes that evidential starting point. The book's summary connects his criticism of skepticism to assumptions about access to evidence.

Suppose the successful worker knows that the report was accepted while the deceived worker knows only that an acceptance message appeared. On E = K, their bodies of evidence can differ. Describing the experiences as indistinguishable does not by itself erase that difference. A skeptic needs an argument for the stronger evidential equality, rather than obtaining it automatically from the experiential resemblance.

This response will not convince someone who insists that legitimate evidence must be internally recognizable. It also does not give the worker a magical test that rules out every forgery. Its purpose is to challenge one route to skepticism. A philosophical theory can distinguish knowing from deception without providing a decision procedure that a deceived person could reliably apply to discover which situation they occupy.

The strongest methodological objection asks what knowledge-first explanation buys us. Declaring knowledge fundamental would be unhelpful if it merely renamed the problem. A defender must show that the proposal clarifies evidence, assertion, mind, or rational inquiry. The argument should be assessed through those explanatory results, not through the impression that a basic concept is automatically either profound or empty.

A second objection targets the demanding account of self-access. Why should a continuous change in confidence prevent a distinctive form of introspective recognition? The anti-luminosity reconstruction makes clear where this challenge lands: on the reliability reasoning connecting one moment with its neighbour. One can question the generality of that reasoning without denying that people sometimes misunderstand their own minds. Equally, acknowledging occasional mistakes does not by itself establish Williamson's stronger conclusion.

A third objection concerns assertion. An audience sometimes needs a well-informed answer when the speaker has uncertainty, and a qualified recommendation may be more useful than silence. The knowledge account's defender must distinguish asserting an uncertain outcome from asserting a known probability, giving advice, or explicitly making a conjecture. Whether those distinctions handle every case is open to argument. Practical usefulness alone does not settle which act was performed or which epistemic rule governs it.

For a hypothetical project team, a useful habit is to separate an observed result from an inference and a forecast. “The receipt arrived” can be assessed separately from “the submission succeeded” and “the client will approve it.” This practice is a contemporary application of the distinctions, not a management technique tested or prescribed by Williamson. Its value is that colleagues can inspect the grounds for each step instead of treating confidence as a substitute for them.

The same care helps when reading an automated summary. Suppose it confidently reports that a contract contains a renewal clause. Checking the contract can establish something about the clause; the summary's fluent delivery cannot by itself do so. This example makes no assumption about whether a machine can have mental states or knowledge. It concerns the human reader's grounds for relying on the answer and the difference between a generated claim and a checked source.

In a friendship, the distinctions can make an assurance more careful. Someone asks whether an invitation was intentionally withheld. You may know that no invitation arrived while only suspecting the explanation. Taking the relationship seriously can require preserving that gap. Epistemic humility here is a willingness to examine what supports an interpretation, not a declaration that ordinary knowledge is impossible.

Williamson's challenge leaves us with a demanding combination: use knowledge as an explanatory starting point while accepting limits to our ability to recognize its presence. The approach offers no universal checklist that settles every claim. It gives reasons to ask which facts support a conclusion, what a speaker commits to, and why another layer of reflection may still leave something unsettled.

• Timothy Williamson, Knowledge and Its Limits. Begin with the introduction, then read chapters 4–5 on self-access, chapter 9 on evidence, and chapter 11 on assertion.
• Williamson, “Précis of Knowledge and its Limits”. A brief statement of the project's motivation in the author's own voice.
• Patrick Greenough and Duncan Pritchard, editors, Williamson on Knowledge. A collection of critical discussions; its role here is further exploration, not primary evidence for Williamson's text.