20.01.01 · philosophy / epistemology

Epistemology: knowledge, justification, and truth

shipped3 tiersLean: none

Anchor (Master): primary sources: Plato Theaetetus, Gettier 1963, Goldman 1967, Sosa 1991, Williamson 2000

Epistemology is the branch of philosophy that asks what knowledge is, how we acquire it, and what it means for a belief to be justified or true. These questions look simple. They are not. The attempt to give a precise answer — to say exactly what knowledge is — has generated one of the richest and most contested literatures in analytic philosophy. This unit follows that attempt from Plato's Theaetetus through the twentieth-century Gettier problem and its aftermath: reliabilism, virtue epistemology, contextualism, knowledge-first, the structure of justification, and the challenge of skepticism. The thread connecting all of it is a single question: what does it take to go beyond merely having a true belief, to actually knowing something?

Intuition [Beginner]

You know a lot of things. You know that you are reading this text right now. You know that water freezes at zero degrees Celsius. You know that the Earth orbits the Sun. But what exactly are you doing when you know these things? What makes knowledge different from guessing, or from having an opinion that happens to be right?

Start with a simple picture. Imagine your friend tells you it will rain tomorrow, and it does rain. Did your friend know it would rain? Maybe. But now imagine your friend had no reason to think it would rain — they just got lucky. In that case we would say they had a true belief, but not knowledge. The belief was right, but it was not justified by any good reason.

This suggests a first answer: knowledge is justified true belief. You know something when (a) you believe it, (b) it is true, and (c) you have a good reason — a justification — for believing it. This three-part analysis, called JTB for short, goes back to Plato's dialogue Theaetetus ^{[Plato Theaetetus]}, where Socrates considers and ultimately rejects the suggestion that knowledge is "true belief with an account." Despite Plato's rejection, the JTB picture dominated Western epistemology for more than two thousand years.

Each of the three components does real work. Belief is the psychological state: you cannot know something you do not even believe. (There are philosophers who contest this, but set that aside for now.) Truth is a matter of how the world is: you cannot know something false, no matter how strongly you believe it or how good your reasons are. The ancients believed the Earth was the centre of the solar system; they were justified by the best evidence available; they were wrong, so they did not know it.

Justification is the link between you and the truth: it is what makes your belief responsible rather than accidental. A jury that convicts based on overwhelming evidence has a justified belief about the defendant's guilt, even if new evidence later emerges. A juror who flips a coin and happens to agree with the verdict has a true belief, but not a justified one.

The JTB picture is clean, intuitive, and — as it turns out — false. In 1963, a young philosopher named Edmund Gettier published a three-page paper in the journal Analysis with the unassuming title "Is Justified True Belief Knowledge?" ^{[Gettier 1963]}. His answer was: not necessarily. Gettier described short scenarios in which someone has a belief that is justified and true, but where the truth of the belief is, intuitively, accidental. These Gettier cases broke the JTB analysis and set off more than sixty years of attempts to fix it.

Here is a simplified Gettier case. Imagine you look at a clock that reads 2:00, and you form the belief "it is 2:00." The clock has always been reliable before. Your belief is justified — you have good evidence. As it happens, the clock stopped exactly twelve hours ago, at 2:00 AM. But it is now 2:00 PM. So your belief is true. You have a justified true belief that it is 2:00.

But you do not know that it is 2:00. Your belief is true only because of a cosmic coincidence — the clock happened to stop at the time that matches reality. The justification and the truth came apart in the wrong way.

The Gettier problem is not a trick or a technicality. It reveals something deep about the structure of knowledge: justification and truth can come apart even when both are present. The challenge, for every epistemologist since 1963, has been to explain what additional condition — or what replacement condition — closes the gap.

One family of responses says: what matters is not just that your belief is justified, but that it was formed by a reliable process. If your belief was produced by a method that generally gets things right — like looking at a working clock, or checking a trustworthy source, or using your eyes under normal lighting conditions — then it counts as knowledge. This is reliabilism, associated especially with Alvin Goldman ^{[Goldman 1967]}. The stopped-clock case is handled naturally: a broken clock is not a reliable time-telling process, so beliefs formed by looking at it are not knowledge, even when they happen to be true.

Another family of responses says: what matters is the intellectual virtue of the knower. On this view, knowledge is a cognitive achievement — a belief that arises from the exercise of stable intellectual character traits like carefulness, open-mindedness, and thoroughness. This is virtue epistemology, developed in its modern form by Ernest Sosa ^{[Sosa 1991]} and Linda Zagzebski. The stopped-clock believer lacks an intellectual virtue at the crucial moment: they failed to check whether the clock was working. Their belief was true, but it was not an achievement — it was a fluke.

A third response changes the subject entirely. Timothy Williamson ^{[Williamson 2000]} argues that knowledge is not decomposable into belief plus truth plus something else. Knowledge is a primitive mental state — not further analysable into components, just as belief itself is not analysable into anything simpler. On this knowledge-first view, the Gettier problem is a pseudo-problem: it arises only because we tried to build knowledge out of pieces that do not add up to it. The right move is to stop trying to analyse knowledge and instead use knowledge to analyse other things — to explain belief and justification in terms of knowledge, rather than the other way around.

These are not the only responses, and none of them is uncontested. The debate over the nature of knowledge remains one of the liveliest in contemporary philosophy. What makes it compelling is that it touches everything: science, law, education, religion, and everyday life all depend on assumptions about what people know and how they know it. Getting clear about knowledge is not a parlor game — it is foundational for understanding how we relate to the world and to each other.

Visual [Beginner]

The visual emphasises the point where the JTB analysis breaks down. Everything flows smoothly until the final step: justified true belief is supposed to be knowledge, but Gettier cases show that it might not be. The gap is where all the philosophical action is.

Worked example [Beginner]

Consider the following scenario, a variant of one of Gettier's original cases.

Smith and Jones have applied for the same job. Smith has strong evidence that Jones will get the job and that Jones has ten coins in his pocket. (Smith counted the coins earlier, and the company president told Smith that Jones would be hired.) From this evidence, Smith infers: "The person who will get the job has ten coins in his pocket."

As it turns out, Smith gets the job, not Jones. And Smith, unknowingly, also has ten coins in his pocket. So Smith's belief — "the person who will get the job has ten coins in his pocket" — is true. It is also justified: Smith had perfectly good evidence. But it is not knowledge. The truth of the belief depends on facts Smith was completely unaware of (that he himself would get the job and that he happened to have ten coins).

Step through the JTB components:

Belief: Yes. Smith believes the proposition.
Truth: Yes. The person who got the job (Smith) does have ten coins.
Justification: Yes. Smith's evidence was strong and reasonable.

All three conditions are satisfied, yet Smith does not know. This is the Gettier problem in its starkest form. The belief is true, but not because of Smith's justification. The justification points to Jones; the truth comes from Smith. The connection between justification and truth has been severed.

Check your understanding [Beginner]

Exercise (easy, multiple choice).

According to reliabilism, what matters most for whether a belief counts as knowledge?

A. How strongly the believer holds the belief. B. Whether the belief was produced by a process that tends to get things right. C. Whether the believer can explain why the belief is true. D. Whether the belief coheres with the believer's other beliefs.

Hint

Reliabilism is about the process that generates the belief, not about the believer's internal states or their ability to articulate reasons.

Answer

Option B. Reliabilism holds that a belief counts as knowledge when it is produced by a reliable cognitive process — one that yields a high proportion of true beliefs over a range of situations. The strength of the believer's conviction (A), their explanatory ability (C), and coherence with other beliefs (D) are not the central reliabilist criterion.

Exercise (medium, short answer).

Construct your own Gettier case. Describe a scenario in which someone has a justified true belief that is not knowledge. Make sure all three JTB conditions are met and that the intuitive lack of knowledge is apparent.

Hint

The recipe for a Gettier case is: (1) the subject has good evidence for a false proposition, (2) from this they infer something that is in fact true, but (3) the truth of the inferred proposition is coincidental relative to the evidence.

Answer

A worked example: Maria is driving through the countryside and sees a barn-shaped structure in a field. She believes "there is a barn in that field." Unbeknownst to Maria, the local area is full of barn facades — realistic fake barns — and only one real barn exists, which happens to be the one she is looking at. Her belief is true and justified (she has visual evidence under normal conditions). But she does not know there is a barn: had she looked at any fake barn, she would have formed the same false belief. The truth is lucky, not secured by her evidence.

This is the well-known "fake barns" case due to Alvin Goldman (1976), and it illustrates the Gettier structure: justification points one way, truth comes from another, and the connection between them is fragile.

Formal definition [Intermediate+]

The JTB analysis can be stated precisely. Let $S$ be a subject (a cognitive agent), and let $p$ be a proposition. Then:

$S knows that p ⟺ (1) S believes that p, (2) p is true, and (3) S is justified in believing that p .$

Each clause requires unpacking.

Belief. Belief is a propositional attitude: to believe that $p$ is to hold $p$ as true in one's cognitive repertoire. Belief comes in degrees — you can believe something tentatively or firmly — but for the JTB analysis, what matters is the binary distinction between believing and not believing. Dispositional accounts treat belief as a disposition to act as if $p$ were true; occurrent accounts restrict belief to propositions currently before the mind.

Truth. The truth condition requires a theory of truth. The default in epistemology is correspondence: $p$ is true if and only if the state of affairs described by $p$ obtains. Alternatives — coherence theories of truth, pragmatic theories, deflationary theories — are available, but they are not the standard background assumption in the JTB debate. What matters for the analysis is that truth is objective: it depends on the world, not on the subject's mental state.

Justification. Justification is the most contested component. At minimum, justification requires that the subject's belief is based on evidence or reasons. But what counts as adequate evidence? Here the literature splits into internalism and externalism about justification.

Internalism holds that the justifying factors for a belief must be internally accessible to the subject — the subject must be able to reflectively access the grounds of their belief. On this view, justification is a matter of the subject's conscious (or potentially conscious) perspective.
Externalism holds that what justifies a belief need not be accessible to the subject; what matters is whether the belief was formed by a process that is objectively reliable, regardless of whether the subject can identify that process. Reliabilism is the paradigm externalist theory.

Formal definition: the Gettier schema

Gettier's 1963 paper ^{[Gettier 1963]} did not just present counterexamples; it relied on two assumptions that were widely held in mid-century epistemology:

Assumption 1 (Fallibility of justification). It is possible for a person to be justified in believing a proposition that is false. Justification does not guarantee truth.

Assumption 2 (Deductive closure of justification). If $S$ is justified in believing $p$ , and $S$ knows that $p$ entails $q$ , and $S$ believes $q$ on the basis of this entailment, then $S$ is justified in believing $q$ .

Given these assumptions, the Gettier schema is:

$S$ is justified in believing a false proposition $p$ (by Assumption 1, this is possible).
$p$ entails a true proposition $q$ (where $q$ is in fact true, though $p$ is false).
$S$ deduces $q$ from $p$ , and believes $q$ on this basis (by Assumption 2, $S$ is justified in believing $q$ ).
$q$ is true (by construction).
So $S$ has a justified true belief in $q$ . But $S$ does not know $q$ , because $S$ 's justification rests on the false proposition $p$ .

The schema shows that Gettier cases are not isolated curiosities. Whenever justification is fallible (Assumption 1) and closed under known entailment (Assumption 2), Gettier cases can be generated systematically. This is why the problem is structural, not merely a matter of fixing a few counterexamples.

Key theorem: the Gettier impossibility result

A precise formulation of the Gettier problem can be stated as a conditional impossibility theorem.

Theorem. No analysis of knowledge as justified true belief can be correct if justification satisfies both of the following conditions:

(J1) Fallibility: For some proposition $p$ , $S$ is justified in believing $p$ and $p$ is false.
(J2) Closure: If $S$ is justified in believing $p$ , and $S$ correctly deduces $q$ from $p$ , then $S$ is justified in believing $q$ .

Proof sketch. Suppose (J1) and (J2) hold. By (J1), there exists a proposition $p$ such that $S$ is justified in believing $p$ and $p$ is false. Since $p$ is false, the disjunction $p \lor q$ can be made true by choosing $q$ to be any true proposition. Since $S$ is justified in believing $p$ , and $p$ entails $p \lor q$ , by (J2) $S$ is justified in believing $p \lor q$ . So $S$ has a justified true belief in $p \lor q$ . But $S$ 's justification for $p \lor q$ depends entirely on $S$ 's justification for the false proposition $p$ . The truth of $p \lor q$ depends entirely on the true proposition $q$ . The justification and the truth are not appropriately connected. So $S$ 's justified true belief in $p \lor q$ is not knowledge. Therefore JTB (with justification satisfying J1 and J2) is not sufficient for knowledge.

This result shows that the Gettier problem is not a peripheral glitch. Any epistemological theory that endorses both fallibility and deductive closure of justification will face Gettier cases. Responses to Gettier must therefore reject at least one of J1, J2, or the JTB analysis itself.

Counterexamples to common slips

"Gettier showed that knowledge does not require justification." This is exactly backwards. Gettier cases show that justification is not enough — but every major response to Gettier retains a role for justification (or something justification-adjacent like reliability or intellectual virtue). No one seriously argues that knowledge is just true belief.
"The Gettier problem is just about lucky true beliefs." Luck is a useful heuristic but not a precise criterion. Many knowledge-constituting beliefs involve an element of luck (you were lucky to have looked at the right moment, lucky your sources were reliable). What makes Gettier cases distinctive is not luck per se but the structural disconnect between the justification and the truth — the justification supports one route to the proposition, the truth comes through another.
"Gettier cases are artificial and have nothing to do with real knowledge." The artificiality is by design: philosophy constructs clean cases to isolate a conceptual feature. The real-world analogues are pervasive. Eyewitness misidentification (where the witness correctly identifies the culprit but for the wrong reasons), medical diagnoses that are right by coincidence, and scientific inferences that happen to align with the truth despite flawed reasoning are all Gettier-style phenomena.

Exercises [Intermediate+]

Exercise 1 (easy, argument reconstruction).

Reconstruct Gettier's argument in formal premise-conclusion form. State each premise explicitly and identify which premise a reliabilist would reject.

Hint

The argument has two key premises about justification (fallibility and closure) and a conclusion that JTB is insufficient. A reliabilist response targets the conception of justification, not the logical structure.

Answer

P1. (Fallibility of justification): It is possible for a subject $S$ to be justified in believing a false proposition $p$ .

P2. (Closure of justification): If $S$ is justified in believing $p$ , and $S$ deduces $q$ from $p$ , then $S$ is justified in believing $q$ .

P3. (Truth of the disjunction): If $p$ is false and $q$ is true, then $p \lor q$ is true.

C1. By P1, there exists a false $p$ that $S$ is justified in believing. By P3, for any true $q$ , $p \lor q$ is true. By P2, $S$ is justified in believing $p \lor q$ . So $S$ has a justified true belief in $p \lor q$ .

C2. But $S$ does not know $p \lor q$ (because $S$ 's justification depends on the false $p$ ). So JTB is not sufficient for knowledge.

A reliabilist does not reject P1 or P2 as such — they reject the internalist conception of justification that underlies them. The reliabilist says: the process that led from $p$ to $p \lor q$ is not reliable in the relevant sense, because the initial belief in $p$ was false and the process did not track the truth of $p \lor q$ . The reliabilist's "justification" is not the same as the internalist's.

Exercise 2 (medium, multiple choice).

A detective has strong evidence pointing to suspect A. The detective believes "the person who committed the crime owns a red car." Suspect A does own a red car. But the actual criminal is suspect B, who also owns a red car. The detective's belief is true and justified. On which response to Gettier does this case bear most directly?

A. Virtue epistemology — the detective exercised intellectual virtue. B. The "no false lemmas" response — the detective's inference relies on a false lemma ("suspect A is the criminal"). C. Contextualism — the standards for knowledge are context-dependent. D. Knowledge-first — knowledge is primitive and unanalysable.

Hint

The detective's reasoning goes through a false intermediate step: they believe the culprit is suspect A. This is the false lemma. Which response to Gettier specifically targets the role of false intermediate beliefs?

Answer

Option B — the "no false lemmas" response. This response (associated with Gilbert Harman and others) proposes that knowledge requires justified true belief where no essential step in the reasoning is false. The detective's inference depends on the false lemma "suspect A is the criminal." If this lemma is essential to the reasoning, then the "no false lemmas" condition is not met, and the case is handled. The response is historically important but faces counterexamples of its own: cases where the subject's reasoning does not go through a false lemma, yet the Gettier-style disconnect remains.

Exercise 3 (medium, defence).

Write a paragraph (4-6 sentences) defending the reliabilist response to Gettier. Explain how reliabilism handles Gettier cases and identify one advantage it has over internalist accounts of justification.

Hint

Focus on the core reliabilist move: replace "justification" with "reliability of the belief-forming process." The advantage is that reliability is objective — it does not depend on whether the subject can articulate their reasons.

Answer

Reliabilism handles Gettier cases by shifting the criterion for knowledge from internal justification to the objective reliability of the belief-forming process. In the stopped-clock case, the process — consulting a broken clock — is unreliable: across a range of possible scenarios, it would yield mostly false beliefs. In the original Gettier cases, the subject's reasoning depends on a false proposition, and the process of inferring a disjunction from a false conjunct does not reliably produce true beliefs. The advantage over internalism is that reliabilism anchors knowledge in an objective feature of the world (the track record of the cognitive process) rather than in the subject's potentially misleading perspective. This makes knowledge less vulnerable to lucky coincidences and better aligned with scientific standards of evidence, where what matters is whether a method produces correct results, not whether the investigator can introspectively access all the relevant factors.

Exercise 4 (medium, critique).

Now write a paragraph (4-6 sentences) critiquing reliabilism. Pick a single line of objection and develop it.

Hint

The two main lines of objection are: (a) the generality problem (how do you specify which "process" is the relevant one?), and (b) the new evil demon problem (a reliable process in the actual world might be unreliable in a sceptical scenario, yet the subject seems equally justified).

Answer

The generality problem is the most persistent objection to reliabilism. To evaluate whether a belief was formed by a reliable process, we must specify which process is the relevant one. But processes can be described at many levels of generality: "looking at a clock," "looking at a stopped clock," "looking at this particular stopped clock at 2:00 PM on a Tuesday," and so on. At some levels of description the process is reliable; at others it is not. The reliabilist must provide a principled way to select the right level of description, and no consensus exists on how to do this. Without such a principle, the reliabilist analysis is incomplete: it tells us that knowledge requires reliability, but it cannot tell us of what. The problem is not that reliability is the wrong criterion, but that the criterion is underspecified until the process is individuated — and individuating the process is itself a philosophical problem that reliabilism was supposed to bypass.

Exercise 5 (hard, argument reconstruction).

Reconstruct Williamson's "knowledge-first" argument against the JTB analysis. The argument has two parts: (a) the failure of JTB, and (b) the positive claim that knowledge is unanalysable. State each part with explicit premises and identify where a critic would push back.

Hint

Part (a) is the Gettier-style point. Part (b) is the deeper claim: that knowledge is a fundamental epistemic notion, not constructible from belief plus truth plus any further condition. The critic's pushback is usually that "unanalysable" is too strong — even if JTB fails, some other multi-component analysis might work.

Answer

Part (a): JTB fails.

P1. If knowledge were analysable as JTB, then every justified true belief would be knowledge (modulo Gettier adjustments).

P2. Decades of attempted fixes to JTB (reliabilism, virtue epistemology, defeasibility conditions, "no false lemmas," sensitivity conditions, safety conditions) have all failed: for each proposed "fourth condition," counterexamples have been produced.

C1. The repeated failure of JTB-style analyses provides strong inductive evidence that no such analysis will succeed.

Part (b): Knowledge is unanalysable.

P3. The most natural explanation for the systematic failure of compositional analyses of knowledge is that knowledge is not compositional — it is not built out of belief plus other ingredients.

P4. Knowledge plays a distinctive role in our cognitive economy: it is the factive mental state that serves as the basis for assertion, action, and practical reasoning.

P5. Other fundamental mental states (belief, desire, intention) are standardly treated as unanalysed primitives in philosophy of mind.

C2. Knowledge should be treated as a primitive — unanalysable into more basic components — and used to explain belief, justification, and evidence, rather than being explained by them (the "E = K" thesis: evidence is knowledge).

Where a critic pushes back. The inductive argument in C1 is not deductive. The history of failed analyses does not entail that no correct analysis exists — it only shows that the attempts so far have failed. A persistent internalist might argue that the right fourth condition has simply not been found yet, and that the project of analysing knowledge is incomplete rather than impossible. Williamson's move from "all attempts have failed" to "the concept is unanalysable" is an inference to the best explanation, and alternative explanations are available: perhaps the concept is complex and we have not yet been clever enough, or perhaps the concept is context-sensitive in a way that resists a single analysis without being primitive.

The Gettier problem and responses [Master]

The Gettier problem is the single most important development in twentieth-century epistemology. Before 1963, the JTB analysis was near-universal — not because it had been carefully defended, but because it seemed so obvious that it did not need defending. Gettier's paper ^{[Gettier 1963]} changed this overnight. The paper is three pages long and contains almost no argument of its own: it simply presents two cases and notes that they are counterexamples. The impact was seismic. A 1997 survey of professional philosophers by the journal Philosophy and Phenomenological Research found that Gettier's paper was the most widely discussed article in the preceding fifty years of epistemology.

The philosophical challenge is not merely to find a "fourth condition" that blocks Gettier cases. It is to understand what knowledge is — and, more deeply, what it means for a belief to be connected to the truth in the right way. Every response to Gettier reflects a different conception of this connection.

Defeasibility theories

One of the earliest responses was the defeasibility approach: a justified true belief is knowledge unless there exists a true proposition that, if the subject knew it, would defeat their justification. On this view, Gettier cases involve "defeating" propositions — facts that would undermine the subject's reason for believing. In the Smith-Jones case, the defeating proposition is "Jones will not get the job." If Smith knew this, his justification for believing "the person who will get the job has ten coins" would collapse.

The challenge for defeasibility theories is to specify which defeating propositions are relevant without trivialising the analysis. There are always true propositions that would somehow undermine a subject's confidence (the subject might learn that they are a brain in a vat, or that their cognitive faculties are unreliable). If any true defeating proposition suffices to block knowledge, then knowledge becomes impossibly demanding — a form of skepticism. Defeasibility theorists must distinguish genuine defeaters from irrelevant or misleading ones, and this distinction has proven difficult to draw precisely.

Sensitivity and safety

Robert Nozick's ^{[Nozick 1981]} sensitivity condition holds that $S$ knows that $p$ only if, were $p$ false, $S$ would not believe that $p$ . Formally: $S$ 's belief that $p$ is sensitive just in case in the nearest possible worlds where $p$ is false, $S$ does not believe $p$ . In the stopped-clock case, the belief "it is 2:00" is not sensitive: in nearby worlds where it is not 2:00 (but the clock still reads 2:00), the subject would still believe it is 2:00. So the belief fails the sensitivity condition and is not knowledge.

Sensitivity handles many Gettier cases, but it faces two well-known problems. First, it fails for necessary truths: if $p$ is necessarily true, there are no possible worlds where $p$ is false, so the sensitivity condition is vacuously satisfied — but we want to say that not every believed necessary truth is known. Second, sensitivity is not closed under known entailment: $S$ might know $p$ by sensitivity, know that $p$ entails $q$ , and believe $q$ on this basis, yet $q$ might fail the sensitivity condition (because there are worlds where $q$ is false that are not worlds where $p$ is false).

The safety condition, developed by Duncan Pritchard and others, reverses the modal direction: $S$ 's belief that $p$ is safe just in case in nearby possible worlds where $S$ believes $p$ , $p$ is true. Safety avoids some of sensitivity's problems (it handles necessary truths correctly, since there are no nearby worlds where the belief is false), but it faces its own counterexamples. In particular, there are cases where a belief is safe (in nearby worlds the subject would not have formed the belief if it were false) but intuitively not knowledge, because the safety is accidental.

Reliabilism and its problems

Alvin Goldman's reliabilism ^{[Goldman 1967]} holds that $S$ knows that $p$ iff $S$ 's belief in $p$ is produced by a reliable cognitive process — one that yields a high ratio of true beliefs to false beliefs across a suitable range of cases. Reliabilism is an externalist theory: the subject need not be able to identify or describe the reliable process; what matters is the objective track record.

The most important objection to reliabilism is the generality problem (described in the Intermediate exercises). A second is the new evil demon problem, due to Stewart Cohen and Alvin Goldman (in a different paper). Imagine a subject whose cognitive faculties are perfectly reliable in the actual world but who is (unknown to them) being deceived by an evil demon. In this scenario, the subject's beliefs are produced by unreliable processes (the demon distorts everything), yet the subject is intuitively justified in their beliefs — they are doing everything an epistemically responsible agent should do. If reliabilism says the subject is not justified, it delivers the wrong verdict. Reliabilists have responded in various ways (distinguishing justification from warrant, appealing to "normal world" reliability), but the problem remains a central point of contention.

Virtue epistemology [Master]

Virtue epistemology ^{[Sosa 1991]} reframes knowledge in terms of the intellectual character of the knower rather than the properties of individual beliefs. On Sosa's version, knowledge is apt belief: belief that is accurate (true) because adroit (produced by a competent cognitive faculty). The "because" is crucial: the truth of the belief must be attributable to the exercise of the cognitive virtue, not to luck.

Sosa's framework distinguishes three levels of epistemic assessment:

Accurate: the belief is true.
Adroit: the belief is produced by a competent faculty (e.g., vision, memory, logical reasoning).
Apt: the belief is accurate because adroit — the truth is creditable to the competence.

A Gettier case, on this analysis, is a belief that is accurate and adroit but not apt: the belief is true, and it is produced by a competent faculty, but the truth is not because of the competence. In the stopped-clock case, the subject's visual competence is in good order, but the truth of the belief does not flow from that competence — it flows from the coincidence that the clock stopped at the right time.

Linda Zagzebski's responsibilist variant of virtue epistemology emphasises the role of intellectual character traits — curiosity, thoroughness, intellectual courage — rather than cognitive faculties. On Zagzebski's view, knowledge requires that the subject's intellectual motivation is properly directed toward truth and that the belief arises from the exercise of these motivations.

Virtue epistemology handles many Gettier cases elegantly, but it inherits the generality problem in a new form: what counts as a "competent faculty" or an "intellectual virtue" requires specification, and the boundary between virtuous and non-virtuous cognition is not always sharp.

Contextualism and the standards for knowledge [Master]

Contextualism, developed by David Lewis, Keith DeRose ^{[DeRose 1995]}, and Stewart Cohen, holds that the truth-conditions of knowledge attributions (" $S$ knows that $p$ ") vary with the context of the attributor. In ordinary conversation, the standards for knowledge are relatively low: "I know the bank is open on Saturday" is true in casual contexts even if the speaker has not double-checked. In high-stakes contexts — a sceptical philosophy seminar, a courtroom — the standards rise, and the same attribution may become false.

Contextualism addresses two issues simultaneously. First, it offers a treatment of Gettier-style cases: in some contexts, the standards for knowledge are low enough that even Gettiered beliefs count (if the conversational context does not raise the possibility of broken clocks or fake barns). In other contexts, the standards exclude them. Second, contextualism promises a response to scepticism: the sceptic raises the standards for knowledge so high that no belief can meet them, but this does not show that knowledge-attributions are false in ordinary contexts — only that they are false in the sceptic's artificially elevated context.

The main objection to contextualism is subject-sensitive invariantism (or pragmatic encroachment): the view that it is not the attributor's context that matters but the subject's practical situation. On this view, whether $S$ knows that $p$ depends in part on what is at stake for $S$ : if the cost of being wrong is very high, higher evidential standards are required for knowledge. Jeremy Fantl and Matthew McGrath have developed this position, arguing that knowledge is a practical state — it licenses action — and that the stakes of the action affect whether the subject has sufficient evidence.

Foundationalism vs coherentism [Master]

A separate but related debate in epistemology concerns the structure of justification. Even if we set aside the Gettier problem and grant that justification is necessary for knowledge, we need an account of how justification is organised. Do all justified beliefs rest on a foundation of basic, self-justifying beliefs, or does justification form a web of mutual support?

Foundationalism ^{[BonJour 1985]} holds that the structure of justification is hierarchical. At the base are basic beliefs — beliefs that are justified without being inferred from other beliefs. These might include perceptual beliefs ("I see a red patch"), introspective beliefs ("I am in pain"), or beliefs about simple logical truths. Non-basic beliefs are justified by being supported, directly or indirectly, by basic beliefs. The resulting structure is a pyramid or a tree: basic beliefs at the bottom, derived beliefs above.

Classical foundationalism (Descartes, Locke) required that basic beliefs be incorrigible — immune to doubt. This proved too restrictive: very few beliefs meet this standard, and classical foundationalism slides toward skepticism. Moderate foundationalism (Chisholm, Audi) relaxes the requirement, allowing that basic beliefs need only be prima facie justified — justified unless there are specific defeaters.

Coherentism holds that justification is a matter of coherence among beliefs. A belief is justified if it fits well into the subject's overall web of beliefs — if it is supported by multiple connections to other beliefs and if the total system is internally consistent and comprehensive. There are no basic beliefs; justification is holistic. The metaphor is a web or a raft (Neurath's boat): every plank supports the others, and none rests on the sea floor.

The central objection to coherentism is the isolation problem: a perfectly coherent set of beliefs might have no connection to the external world. A well-constructed fairy tale is internally coherent, but its propositions are not justified. Coherentism seems to lack an input from the world — a way to explain why this coherent system corresponds to reality rather than some other. Foundationalists argue that without basic beliefs anchored in perception or introspection, coherentism cannot distinguish genuine justification from sophisticated fiction.

Foundationalists face their own challenge: the regress problem. If every non-basic belief is justified by other beliefs, what justifies the basic beliefs? If basic beliefs are self-justifying, what makes them so? And if some beliefs can be self-justifying, why cannot all beliefs be? The regress problem is the mirror image of the isolation problem, and the two together define the dialectical space for theories of epistemic structure.

A third option, infinitism, holds that the regress of justification does not terminate — it goes on forever. Each belief is supported by another, ad infinitum, without circularity or a foundational stopping point. Infinitism has few defenders, because finite minds cannot entertain infinitely many justifying beliefs, but it avoids the problems of both foundationalism (no need for special basic beliefs) and coherentism (no need for circular support). Peter Klein is the most prominent contemporary infinitist.

Skepticism [Master]

The sceptical challenge is the hardest problem in epistemology — and also the oldest. In its strongest form, skepticism holds that we cannot know anything about the external world. The challenge is not that we are sometimes wrong; it is that no amount of evidence ever suffices to rule out the possibility that we are systematically deceived.

The classical sceptical argument runs through a scenario: the evil demon (Descartes), the brain in a vat (Putnam), or the Matrix (the film). In each scenario, the subject's experiences are indistinguishable from normal experience, but the external world is radically different from what the subject believes. The sceptic argues:

P1. If $S$ knows that $p$ , then $S$ can rule out all alternatives to $p$ (including sceptical scenarios).

P2. $S$ cannot rule out sceptical scenarios (because the experiences are indistinguishable).

C. Therefore, $S$ does not know that $p$ .

The argument is valid. Every response to skepticism must reject one of the premises. Mooreanism (G. E. Moore) rejects P2: Moore famously held up his hands and claimed to know that they existed — thereby claiming to have ruled out the sceptical scenario. The cost is that the response feels question-begging: the sceptic will ask how Moore knows his hands exist, and the dialectic reopens.

Contextualism rejects P1 by arguing that in ordinary contexts, the relevant alternatives do not include sceptical scenarios. The sceptic raises the standards for "knowing" by introducing radical possibilities that ordinary speakers do not typically consider. In ordinary contexts, we do know things; in the sceptic's elevated context, perhaps we do not.

Relevant alternatives theory (developed by Goldman and others) rejects P1 by arguing that knowledge requires ruling out only relevant alternatives, not all logically possible ones. Sceptical scenarios are not relevant in ordinary contexts, so they do not need to be ruled out. The challenge is to specify which alternatives are relevant without either collapsing into skepticism (all alternatives are relevant) or trivialising knowledge (only "easy" alternatives count).

Externalism (including reliabilism and virtue epistemology) rejects P2 in a different way: even if the subject cannot rule out the sceptical scenario by reflection, the subject's cognitive faculties are in fact reliable (in the actual world), and this objective fact suffices for knowledge. The subject does not need to prove they are not a brain in a vat; they just need to be reliably connected to the truth. The sceptic's response is that this amounts to assuming what needs to be proved — assuming the reliability of faculties that might be unreliable.

The debate between sceptics and anti-sceptics has been enriched by the closure principle for knowledge: if $S$ knows $p$ , and $S$ knows that $p$ entails $q$ , then $S$ knows $q$ . If closure holds, and the subject cannot know they are not a brain in a vat, then by closure the subject cannot know anything that entails they are not a brain in a vat — which is nearly everything about the external world. Denying closure is a possible response (Nozick's sensitivity theory denies it), but closure is deeply intuitive, and denying it has high philosophical costs.

Social epistemology [Master]

Traditional epistemology focuses on the individual knower: what does it take for me to know something? Social epistemology asks how knowledge is produced, transmitted, and assessed in communities. The shift from individual to social is not merely a change of topic; it raises new questions about testimony, disagreement, trust, and the division of cognitive labour.

Testimony is the primary mechanism by which most human knowledge is transmitted. You know that the Earth orbits the Sun not because you have verified it yourself, but because you learned it from teachers, books, and other sources. The epistemology of testimony asks: under what conditions does testimony generate knowledge? Reductionists (Fricker, Lyons) hold that testimony is epistemically valuable only insofar as it can be reduced to non-testimonial evidence (you trust your doctor because doctors have a track record of being right). Anti-reductionists (Burges, Coady) hold that testimony is a basic source of knowledge, not reducible to perception, memory, or inference — we are entitled to trust testimony unless there are specific reasons not to.

Peer disagreement poses a challenge to individual epistemology. Suppose you and an epistemic peer — someone equally intelligent, informed, and careful — disagree about a proposition. The conciliationist response (Elga, Christensen) holds that you should reduce your confidence in the disputed proposition: the disagreement itself is evidence that you might be wrong. The steadfast response (Kelly, van Inwagen) holds that you are entitled to maintain your original confidence: the fact that a peer disagrees does not, by itself, undermine your reasons. The debate has implications for scientific consensus, political disagreement, and religious belief.

The division of cognitive labour (Kitcher, Strevens) concerns the social organisation of inquiry. No individual can master all the evidence for all their beliefs; knowledge production is necessarily distributed across communities with specialised expertise. The epistemological question is how this distribution ought to be organised: how should trust be allocated among experts, and how should non-experts decide which experts to believe?

Epistemic injustice, developed by Miranda Fricker, concerns the ways in which social power distorts epistemic practices. Testimonial injustice occurs when a speaker's testimony is given less credibility than it deserves because of prejudice (e.g., racial or gender bias). Hermeneutical injustice occurs when a group lacks the conceptual resources to make sense of their own experiences, because the dominant interpretive frameworks exclude them. Fricker's work shows that epistemology is not politically neutral: the social conditions under which knowledge is produced and transmitted affect who can be a knower and what can be known.

Knowledge-first and the E=K thesis [Master]

Timothy Williamson's Knowledge and its Limits (2000) ^{[Williamson 2000]} is the most influential epistemology book of the twenty-first century so far. Williamson's central thesis, E = K, holds that one's evidence is what one knows: $Evidence (S) = {p : S knows that p}$ . This reverses the traditional direction of analysis: instead of analysing knowledge in terms of evidence, Williamson analyses evidence in terms of knowledge.

The E = K thesis has far-reaching consequences. It implies that evidence is always factive — you cannot have evidence for a false proposition, because knowing requires truth. It implies that two subjects who believe the same false proposition have different evidence (because neither knows it, so it is not evidence for either). And it implies that the sceptical challenge — "how do you know your evidence is not misleading?" — is malformed: evidence just is knowledge, so the question reduces to "how do you know what you know?", which is circular.

Williamson's positive view is that knowledge is a prime condition — not composite, not built from simpler components, and not subject to analysis in terms of belief plus truth plus a further condition. This is the knowledge-first programme: treat knowledge as fundamental and use it to explain other epistemic notions (belief, justification, evidence, assertion) rather than the other way around.

The knowledge-first programme has generated a large literature. Critics object that treating knowledge as unanalysable is an admission of defeat — it gives up on the explanatory project of saying what knowledge is. Defenders reply that some theoretical primitives are inevitable (physics has mass, charge, and spin; epistemology can have knowledge), and that knowledge is a better primitive than belief or justification because it is factive and because it plays a central role in our cognitive and social practices.

Connections [Master]

Logic and formal methods [20.01.NN] (pending) connects via epistemic logic: the possible-worlds semantics for knowledge operators ( $K_{a} p$ : "agent $a$ knows that $p$ ") developed by Hintikka (1962) and refined in distributed systems (Fagin-Halpern-Moses-Vardi 1995) provides a formal model of knowledge that complements the philosophical analysis. A formalisation of the Gettier schema in modal logic is a natural target for §20.01.
Philosophy of science 20.07.01 pending (pending) connects via the analysis of scientific knowledge. The Gettier problem and its responses apply to scientific claims: a scientific theory might be empirically successful (justified) and true, yet known for the wrong reasons. The debate over scientific realism involves epistemological questions about whether our best theories give us knowledge of unobservable entities.
Philosophy of mind 20.06.01 pending (pending) connects via the nature of belief. If knowledge is factive (entails truth) and belief is not, then knowledge is not a species of belief on the knowledge-first view — it is a different kind of mental state entirely. The relationship between knowledge, belief, and other propositional attitudes is a shared concern of epistemology and philosophy of mind.
Ethics [20.08.NN] (pending) connects via epistemic injustice (Fricker) and the ethics of belief (Clifford, "The Ethics of Belief," 1877): whether there are moral obligations governing what one believes, and how social power shapes who is permitted to be a knower.
Philosophy of language [20.09.NN] (pending) connects via the semantics of knowledge attributions. Contextualism is as much a thesis in the philosophy of language as in epistemology: it claims that the truth-conditions of "S knows that p" vary with the context of utterance. The debate over contextualism engages with work in linguistics on indexicality and context-sensitivity (Kratzer, Kaplan, Lewis).

Cross-domain to philosophy of physics 20.03.01 pending: the measurement-problem debate involves epistemological questions about what observers can know about quantum systems and whether knowledge of a system's state is possible when the state is in superposition. The QBist interpretation (Fuchs, Mermin, Schack) explicitly treats quantum states as epistemic — degrees of belief rather than physical facts — and is in direct conversation with the epistemological tradition covered in this unit.

Historical & philosophical context [Master]

The Western epistemological tradition begins with Plato's Theaetetus (c. 369 BCE) ^{[Plato Theaetetus]}, where Socrates considers and rejects three definitions of knowledge: perception, true belief, and true belief with an account (logos). Plato's own view, developed in the Republic and the Meno, is that knowledge is of the Forms — eternal, unchanging objects of rational insight — and that sense perception yields only opinion (doxa). The distinction between knowledge and true belief is already present: the Meno offers the analogy of the guided statues of Daedalus, which stay in place only when tied down — just as true beliefs "run away" from the mind unless "tied down" by reasoning (i.e., justified).

The medieval period added little to the analysis of knowledge per se, though Augustine's theory of divine illumination (knowledge is made possible by God's interior light) and Aquinas's synthesis of Aristotelian empiricism with Christian theology developed the background assumptions about perception and reason that early modern epistemology would challenge.

Descartes's Meditations (1641) introduced the modern sceptical challenge in its most powerful form. By entertaining the hypothesis of an evil demon who systematically deceives him, Descartes found that he could doubt nearly everything — except his own existence as a thinking thing (cogito ergo sum). The Cartesian project of rebuilding knowledge from an indubitable foundation set the agenda for foundationalism and for the internalist tradition in epistemology.

Locke, Berkeley, and Hume — the British empiricists — challenged the rationalist assumption that knowledge is grounded in pure reason. Locke's theory of ideas (all knowledge derives from experience), Berkeley's idealism (existence is perception), and Hume's radical empiricism (causation is mere habit of association) progressively weakened the claim that reason alone can deliver knowledge of the external world. Hume's problem of induction — that no amount of observed instances can logically guarantee the next instance — remains a central challenge for any theory of empirical knowledge.

Kant's Critique of Pure Reason (1781) attempted to resolve the rationalist-empiricist standoff by arguing that knowledge requires both sensory input and innate conceptual structures (the categories). The synthetic a priori — propositions that extend knowledge yet are knowable independently of experience — was Kant's answer to Hume. Whether the Kantian programme succeeds is itself a contested epistemological question.

The twentieth century saw the professionalisation of epistemology as a subdiscipline of analytic philosophy. The logical positivists (Carnap, Schlick) attempted to ground all knowledge in the verification principle: a proposition is meaningful only if it is analytically true or empirically verifiable. This programme collapsed under its own weight (the verification principle itself is neither analytically true nor empirically verifiable), but it sharpened the distinction between knowledge and meaning and influenced the development of Bayesian epistemology.

Gettier's 1963 paper ^{[Gettier 1963]} is the watershed moment for contemporary epistemology. The subsequent literature — Goldman's reliabilism ^{[Goldman 1967]}, Nozick's tracking theory ^{[Nozick 1981]}, Sosa's virtue epistemology ^{[Sosa 1991]}, Williamson's knowledge-first ^{[Williamson 2000]} — constitutes the mainstream of the field. The journals Journal of Philosophy, Philosophical Review, Analysis, Philosophy and Phenomenological Research, and Episteme are the primary venues for contemporary epistemological research.

Social epistemology emerged as a distinct subfield in the late 1980s, with Alvin Goldman's Knowledge in a Social World (1999) and the work of Fuller, Kitcher, and Longino establishing the social dimensions of knowledge production as a legitimate domain of epistemological inquiry. Fricker's Epistemic Injustice (2007) expanded the field to include questions of power, identity, and oppression.

Bibliography [Master]

Foundational and historical:

Plato — Theaetetus (c. 369 BCE), trans. Levett, rev. Burnyeat (Hackett, 1990). [Need to source.]
Descartes, R. — Meditationes de Prima Philosophia (1641); English trans. Meditations on First Philosophy (Cambridge University Press, 1996). [Need to source.]
Hume, D. — An Enquiry Concerning Human Understanding (1748). [Need to source.]
Kant, I. — Critik der reinen Vernunft (1781); English trans. Critique of Pure Reason (Cambridge University Press, 1998). [Need to source.]

Gettier and the JTB debate:

Gettier, E. — "Is Justified True Belief Knowledge?", Analysis 23, 121–123 (1963). [Need to source.]
Goldman, A. — "A Causal Theory of Knowing", Journal of Philosophy 64, 357–372 (1967). [Need to source.]
Goldman, A. — "Discrimination and Perceptual Knowledge", Journal of Philosophy 73, 771–791 (1976). (The fake barns case.) [Need to source.]
Clark, M. — "Knowledge and Grounds: A Comment on Mr. Gettier's Paper", Analysis 24, 46–48 (1963). [Need to source.]

Reliabilism and externalism:

Goldman, A. — Epistemology and Cognition (Harvard University Press, 1986). [Need to source.]
Plantinga, A. — Warrant and Proper Function (Oxford University Press, 1993). [Need to source.]

Virtue epistemology:

Sosa, E. — Knowledge in Perspective: Selected Essays in Epistemology (Cambridge University Press, 1991).
Zagzebski, L. — Virtues of the Mind (Cambridge University Press, 1996). [Need to source.]
Sosa, E. — A Virtue Epistemology: Apt Belief and Reflective Knowledge, Vol. I (Oxford University Press, 2007). [Need to source.]

Contextualism and pragmatic encroachment:

DeRose, K. — "Contextualism and Knowledge Attributions", Philosophy and Phenomenological Research 52, 913–929 (1992). [Need to source.]
Lewis, D. — "Elusive Knowledge", Australasian Journal of Philosophy 74, 549–567 (1996). [Need to source.]
Fantl, J. & McGrath, M. — Knowledge in an Uncertain World (Oxford University Press, 2009). [Need to source.]

Knowledge-first:

Williamson, T. — Knowledge and its Limits (Oxford University Press, 2000).
Williamson, T. — "Knowledge First, Knowledge Second", in Knowledge First (ed. Carter, Gordon, and Jarvis, Oxford University Press, 2017). [Need to source.]

Structure of justification:

BonJour, L. — The Structure of Empirical Knowledge (Harvard University Press, 1985).
Haakonsson, K. & Winther, R. (eds.) — Epistemology: Classic Problems and Contemporary Responses (Rowman & Littlefield, 2010). [Need to source.]

Skepticism:

Nozick, R. — Philosophical Explanations (Harvard University Press, 1981).
Pritchard, D. — Epistemic Luck (Oxford University Press, 2005). [Need to source.]
Stroud, B. — The Significance of Philosophical Scepticism (Oxford University Press, 1984). [Need to source.]

Social epistemology:

Goldman, A. — Knowledge in a Social World (Oxford University Press, 1999). [Need to source.]
Fricker, M. — Epistemic Injustice: Power and the Ethics of Knowing (Oxford University Press, 2007). [Need to source.]
Kitcher, P. — The Advancement of Science: Science without Legend, Objectivity without Illusions (Oxford University Press, 1993). [Need to source.]

Textbooks and surveys:

Pritchard, D. — What is this Thing Called Knowledge?, 5th ed. (Routledge, 2023). [Need to source.]
Steup, M. & Neta, R. (eds.) — Contemporary Debates in Epistemology, 2nd ed. (Wiley-Blackwell, 2014). [Need to source.]
Nagel, T. — What Does It All Mean? (Oxford University Press, 1987). [Need to source.]

Wave 2 epistemology seed unit, produced 2026-05-21. No hooks_out declared; future units in epistemology (e.g., 20.01.02 on sceptical arguments, 20.01.03 on Bayesian epistemology) may declare incoming hooks. Status is shipped per curriculum plan. External epistemology reviewer per PHILOSOPHY_PLAN §9 is pending.

Prerequisites

none — this is a leaf unit

Tier anchors

beginner: Any intro philosophy textbook; Nagel, What Does It All Mean?
intermediate: Pritchard, What is this Thing Called Knowledge?; Steup & Neta, Contemporary Debates in Epistemology
master: primary sources: Plato Theaetetus, Gettier 1963, Goldman 1967, Sosa 1991, Williamson 2000

References

TODO_REF pending
Gettier, E. — Is Justified True Belief Knowledge? (Analysis, 1963) · full text · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF
Gettier, E. — Is Justified True Belief Knowledge? (Analysis, 1963) [non-pending marker — canonical three-page counterexample to JTB] · full text
TODO_REF pending
Plato — Theaetetus (c. 369 BCE) · 201c–210a (knowledge as true belief with an account) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
Goldman, A. — A Causal Theory of Knowing (Journal of Philosophy, 1967) · full text · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
Sosa, E. — Knowledge in Perspective (Cambridge University Press, 1991) · Ch. 1–5 (virtue epistemology); Ch. 13 (intellectual virtue) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
Williamson, T. — Knowledge and its Limits (Oxford University Press, 2000) · Ch. 1 (E=K); Ch. 4–5 (knowledge-first) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
Nozick, R. — Philosophical Explanations (Harvard University Press, 1981) · Ch. 3 (knowledge and skepticism; tracking theory) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
BonJour, L. — The Structure of Empirical Knowledge (Harvard University Press, 1985) · Ch. 3–5 (foundationalism); Ch. 8 (coherentism critique) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
Pritchard, D. — What is this Thing Called Knowledge? 5th ed. (Routledge, 2023) · Ch. 1–4 (JTB and Gettier); Ch. 7–8 (reliabilism, virtue epistemology) · see docs/catalogs/NEED_TO_SOURCE.md#epistemology
TODO_REF pending
DeRose, K. — Contextualism and Knowledge Attributions (Philosophy and Phenomenological Research, 1995) · full text · see docs/catalogs/NEED_TO_SOURCE.md#epistemology

Reviewer

TBD

Estimated time

beginner: 20m
intermediate: 40m
master: 60m