Knowledge

Knowledge is justified true belief — or so Plato claimed in the Meno and Theaetetus, and so epistemology taught for two millennia, until Edmund Gettier published a three-page paper in 1963 that refuted it. The refutation was trivially simple, which is why it had not been published sooner: no one thought to look for counterexamples to a definition that felt obviously correct. This is itself a datum about knowledge: we are very bad at knowing when we know things.

The Classical Analysis and its Destruction

The tripartite analysis holds that S knows P if and only if: (1) P is true, (2) S believes P, and (3) S is justified in believing P. Gettier cases are constructed situations in which all three conditions are satisfied but intuition firmly denies that the person knows. The simplest: you look at a clock that has stopped, and it reads 3:00. It is in fact 3:00. You believe it is 3:00, and you are justified — clocks are usually right. But you do not know it is 3:00, because the clock is broken and your belief is accidentally true.

Fifty years of attempted repairs have produced no consensus replacement. Causal theories require the right causal connection between fact and belief. Reliability theories require that the process producing the belief is generally reliable. No-false-lemma theories require that the belief is not based on any false intermediate step. Each repair faces new counterexamples. Epistemology's central problem remains open.

The lesson may be that 'knowledge' is not a natural kind — not a single thing that a unified analysis will capture. It may be a family of related epistemic successes that human language groups under one word: perceptual knowledge, memorial knowledge, testimonial knowledge, theoretical knowledge, procedural knowledge (knowing-how). These may require different analyses, or none at all.

Justified Belief and the Regress Problem

Any belief S claims to know must be justified. What justifies it? Other beliefs, presumably. But what justifies those? If justification requires further justification indefinitely, we face an infinite regress. If the regress terminates in unjustified foundational beliefs (foundationalism), we must explain why these beliefs get a free pass. If beliefs mutually support each other without a foundation (coherentism), we must explain why coherent systems of false beliefs do not count as knowledge.

Reliabilism (Goldman) cuts this knot by replacing 'justified' with 'produced by a reliable process' — the question of justification becomes a question of causal history, not of reasons. But this generates its own problems: what counts as a reliable process? Reliable in what reference class? A process can be reliable in one reference class and unreliable in another. The specification problem for reliability mirrors the original regress problem, but one level up.

The Bayesian approach converts justification into probability: S is justified in believing P to degree r if S's credence in P, updated by all available evidence via Bayes' theorem, equals r. This is the most mathematically tractable framework available. It handles evidence accumulation, degrees of belief, and confirmation relations with precision no other framework matches. Its vulnerability: the priors must come from somewhere, and Bayesianism has no account of what makes prior probabilities rational that is not itself circular.

Knowledge and Action

Pragmatists — Peirce, James, Dewey — argued that the question 'what is knowledge?' cannot be separated from 'what is knowledge for?' Knowledge is not a spectator relation between a mind and the world; it is an instrument for successful action. A belief counts as knowledge when it serves as a reliable guide to achieving one's aims.

This dissolves the Gettier problem: on the pragmatist view, knowledge is not a special species of true belief. It is any belief that successfully functions as a guide to action in the domain where it is deployed. The stopped clock case does not generate knowledge because the belief about the time does not reliably guide clock-related action (you would look at the wrong clock again).

The pragmatist move is powerful but raises a cost: it makes knowledge agent-relative and purpose-relative. What functions successfully for one agent's purposes may function unsuccessfully for another's. The shared-knowledge structures that science, law, and democracy depend on — the idea that there are facts accessible to multiple agents — become harder to ground.

The question 'what is knowledge?' has proven more durable than almost any other in philosophy. This durability is not evidence that the question is unanswerable — it is evidence that the concept of knowledge is doing more work in our cognitive economy than a single analysis can capture. Any theory that makes the Gettier problem disappear by redefinition has not solved the problem — it has changed the subject.