Three Pages Against 2,000 Years: The Gettier Problem and What Knowledge Really Is

The Hook: A Paper Shorter Than This Introduction That Shook an Entire Discipline

In June 1963, the journal Analysis published a paper barely three pages long. Its author, Edmund Gettier, was a 36-year-old, largely unknown philosopher at Wayne State University who had until then published almost nothing. The legend—which Gettier himself later took relish in feeding—holds that he wrote the piece only because his university wanted to see publications from him and the pressure was mounting. He had one small idea, wrote it down, and submitted it, almost reluctantly. For the rest of his career he published hardly anything else of note. He didn't need to. Those three pages were enough.

The paper bore the plain title Is Justified True Belief Knowledge? And its answer, delivered in two short, almost comically constructed thought experiments, was: no. With that, Gettier toppled a definition of knowledge that in its essentials had formed the foundation of epistemology since Plato—that is, for roughly 2,400 years. Overnight—well, over a few years—what had looked like a settled question became one of the liveliest research programs in modern philosophy. To this day, more than sixty years later, there is no generally accepted repair. The "Gettier problem" remains open.

Why should this interest you, beyond the intellectual appeal? Because the question "What is knowledge?" is anything but academic. It sits in the engine room of everything you do professionally. When does a monitoring system "know" that a server is about to fail—and when is it merely right by accident? When did a language model "know" something, and when did it merely produce a plausible sentence that happened to be true? When may an auditor write in a compliance report that a company "knew" about a risk? In every one of these cases, we intuitively distinguish between real knowledge and lucky hits that look like knowledge. Gettier laid bare exactly this distinction—and showed that we understand it far less well than we thought.

This article takes you the whole way: from Plato's original definition through Gettier's two famous counterexamples and their cunning inner logic, through the most important repair attempts of the past sixty years—No False Lemmas, the causal theory, reliabilism, Nozick's truth-tracking, defeasibility—to Williamson's radical proposal to turn the whole question inside out, and finally to what all of it means for AI, data, and our own approach to security.

Part 1: The Classical Definition – Knowledge as Justified True Belief

Plato's Legacy

The question of what distinguishes knowledge from mere opinion is ancient. Plato poses it sharply in two dialogues, the Meno and the Theaetetus. In the Meno he asks why knowledge should be more valuable than a correct opinion. If two travelers both know the way to Larissa—one because he knows it, the other because he correctly guesses it—both arrive all the same. Wherein lies the added value of knowledge? Plato's famous answer: a merely correct opinion is fleeting; it "runs away like the statues of Daedalus" unless one "ties it down by grasping the cause." Knowledge is tied-down, secured true opinion, anchored by reasons.

From this intuition later philosophy distilled an elegant tripartite structure, the so-called JTB definition (justified true belief). It holds: a person S knows that p if and only if three conditions are met.

First, the truth condition: p is true. You cannot know something false. If I "know" that the Earth is flat, then I do not know it—I merely believe it, mistakenly. Knowledge is factive.

Second, the belief condition: S believes that p. You cannot know something you don't even hold to be true. Knowledge is a form of taking-to-be-true.

Third, the justification condition: S is justified in believing that p. Here lies the decisive filter. It separates knowledge from the lucky hit. Someone who bets on red at the casino and wins had a true belief ("red will come up")—but they did not know it, because they had no good reason. Only justification—the evidence, the good grounds, the reliable source—turns accidentally being right into genuine knowledge.

The Apparent Perfection of the Definition

These three conditions seem not only plausible but almost complete. They are, in the language of logic, individually necessary and jointly sufficient. Necessary means: drop one and it is no longer knowledge. A false belief is not knowledge (truth missing). A true fact I do not believe, I do not know (belief missing). A true belief without grounds is mere guessing (justification missing). And sufficient means: if all three are met, it must be knowledge. So at least went the assumption that for two millennia hardly anyone seriously doubted. A. J. Ayer restated it canonically once more in 1956, and Roderick Chisholm did so similarly. The question "What is knowledge?" was considered essentially answered. Only the fine details remained.

It was precisely this assumption of sufficiency that Gettier attacked. His thesis was surgically precise: you can satisfy all three conditions—a true, believed, justified belief—and still not have knowledge. The definition leaves open a back door through which luck slips in.

Part 2: Gettier's Two Cases – How Luck Comes In Through the Back Door

Case 1: Smith, Jones, and the Ten Coins

Imagine that Smith and Jones have both applied for the same job. Smith has two pieces of information he regards as solid. First, the company president has credibly assured him: "Jones will get the job." Second, Smith happens to have counted the coins in Jones's pocket—there were exactly ten. From these two observations, Smith draws a logically entirely correct inference and forms the belief:

(e) The man who will get the job has ten coins in his pocket.

This belief is excellently justified. It rests on the president's statement and on Smith's own count. A reasonable person would consider (e) well grounded.

Now the twist. In reality it is not Jones who gets the job, but Smith himself. And—unknown to Smith—Smith too happens to have exactly ten coins in his own pocket. So the belief (e) is true: the man who gets the job (Smith) does in fact have ten coins in his pocket. It is believed. And it is justified. All three JTB conditions are met.

And yet—this is Gettier's point—we would hesitate to say that Smith knows that the man who gets the job has ten coins. For his belief became true only through a double coincidence: that he himself gets the job and that he himself has ten coins. His justification was aimed at Jones; the truth came from Smith. Grounds and fact have come apart. Smith simply got lucky.

Case 2: Jones's Ford, or Brown's Whereabouts in Barcelona

The second case is even more cunning, because it turns logic itself against the definition. Smith has strong evidence for the belief: "Jones owns a Ford." Jones has always driven a Ford, just gave Smith a ride in a Ford, and so on. Now Smith has another friend, Brown, of whose whereabouts he has not the faintest idea. Out of pure logical play—a true statement implies any statement joined to it by "or"—Smith forms three disjunctions, among them:

(h) Either Jones owns a Ford, or Brown is in Barcelona.

Smith believes (h), and he is justified, because he correctly derives it from his well-grounded Ford belief. Now the double twist. First: Jones in fact owns no Ford at all—he is currently driving a rental. So Smith's original belief was false. Second, and here is the kicker: Brown is in fact—by sheer coincidence—right now in Barcelona. This makes the disjunction (h) true, because its second disjunct is true. Smith believes (h), is justified, and (h) is true. Again all three conditions are met—and again no one would say Smith knows that (h) holds. He has no idea that Brown is in Barcelona. His belief is derived from a false premise and rescued only by an independent coincidence.

The Common Pattern: Double Luck

Both cases share a structure one can call double luck. First something goes wrong: Smith's justification rests on something false (Jones gets the job / Jones owns a Ford). Then, independently, something goes accidentally right (Smith has ten coins / Brown is in Barcelona). The first piece of bad luck is offset by the second piece of good luck, and at the end stands a true, justified belief—which is nonetheless not knowledge, because the connection between justification and truth has been severed. The belief is true, but not true because the justification was good. It is true by accident.

It is precisely this "by accident"—epistemologists speak of epistemic luck—that is the core of the problem. Knowledge, runs the sharpened intuition after Gettier, is incompatible with luck. It must not be a matter of chance that my justified belief is true. But the JTB definition does not guarantee that.

Part 3: The Phenomenon Is Bigger Than Gettier – Related Cases

It quickly emerged that Gettier's construction was no isolated trick but a pattern that can be varied at will. Three classic examples make this vivid.

The stopped clock (traceable to Bertrand Russell). You look at a wall clock that looks reliable, and it reads 3:00. You form the belief that it is 3:00—and indeed it is 3:00. But the clock stopped exactly twelve hours ago. You have a true, justified belief, but only because you looked at the one right moment. Knowledge? Hardly.

The sheep in the field (Roderick Chisholm). You stand at a field and see something that looks exactly like a sheep. You believe: "There is a sheep in the field." And there is a sheep in the field—only behind a hill, invisible to you. What you actually see is a dog disguised as a sheep. Your belief is true and justified, but what makes it true is something you never perceived.

The fake barns (Alvin Goldman). Henry is driving through the countryside and sees something that looks exactly like a barn. He thinks: "A barn." And it is a real barn. But Henry does not know that the whole area is full of perfect barn façades—mere fronts. Had he looked at one of the fakes, he would have been deceived. That he happens to be looking at the one real barn is luck. This "fake barn" case is especially interesting because here perception works perfectly normally—the problem sits in the environment, not in the head. Some epistemologists therefore do not count it among the genuine Gettier cases; the boundaries of the concept are themselves blurry.

This variety shows: Gettier had not found an isolated case but a fault line in the architecture of the concept of knowledge. And half a century of philosophy has tried to mend it.

Part 4: The Repair Attempts – and Why None Quite Succeeds

The responses fall broadly into two strategies. One tightens the justification so that Gettier situations never arise in the first place. The other adds a fourth condition to truth, belief, and justification. Both strategies have clever proponents—and both run into the same wall.

Repair 1: Infallibility

The most radical proposal: the problem is fallible justification. In all Gettier cases the evidence was good but not conclusive—it left open the possibility of error. If instead we demand infallible justification, i.e. grounds that logically exclude error, the cases vanish. This is the classic Cartesian position: knowledge requires certainty.

The price, however, is devastating. Almost nothing we believe we know in everyday life rests on infallible evidence. My senses can deceive me, my memory can betray me, every measurement has a tolerance. If knowledge required infallibility, we would have virtually none—the repair leads straight into skepticism. Whoever tries to solve Gettier with infallibility saves the concept of knowledge by emptying it almost entirely. A bad bargain.

Repair 2: No False Lemmas

A far more economical idea: in both Gettier cases Smith derives his true belief from a false intermediate assumption. "Jones gets the job" is false; "Jones owns a Ford" is false. So we demand as a fourth condition: genuine knowledge may not rest on a false premise. This No False Lemmas condition is intuitively attractive and disposes of Gettier's original cases elegantly.

But it has two weaknesses. First, it does not catch every case: in the stopped clock and the disguised-sheep cases there is no clearly identifiable false premise one infers from—the belief arises directly from the (deceptive) perception, not from a derivation. Second, more generally: if you demand that no falsehood whatsoever plays a part in the evidence, skepticism threatens again, because our thinking is almost always full of small, incidental errors. Keith Lehrer therefore weakened the requirement to "no significant and ineliminable false core premises." But then the old problem of vagueness begins: how significant is significant enough?

Repair 3: The Causal Theory

Alvin Goldman proposed a different route in 1967. What is missing in Gettier, he argued, is the right causal connection between the fact and the belief. In genuine perceptual knowledge, the fact—the tree in front of me—causes my belief via a normal chain (light, eye, optic nerve). In Gettier this chain is cut: that Smith has ten coins in no way caused his belief (e); it arose from Jones's coins. So: knowledge requires that the fact which makes p true also helps cause, in an appropriate way, the belief that p.

This too has a deep snag. First, it fails for abstract knowledge: that 2 + 2 = 4 has no causal effects—numbers cause nothing—and yet we know it. The causal theory fits only empirical knowledge. Second, and more seriously: what exactly is an appropriate causal chain? One can construct Gettier cases in which the truth-making fact is indeed involved in the genesis of the belief—via a deviant causal chain—and yet we still would not speak of knowledge. The theory shifts the riddle from "What is justification?" to "What is a non-deviant causal chain?"—and the vagueness remains.

Repair 4: Reliabilism

Closely related but more general is reliabilism, also developed largely by Goldman. Its core thought: what makes a belief knowledge is not that the believer can produce reasons, but that the process that generated the belief is reliably truth-conducive. A healthy eye in daylight is a reliable process; guessing is not. In Gettier cases the process is precisely not reliable: in an area full of barn façades, "look and believe" mostly produces false beliefs—Henry just got lucky.

Reliabilism remains one of the most influential theories to this day, precisely because it "externalizes" justification: you need not know your reasons; the process merely has to work. Yet it too struggles with how to delimit the relevant process and its relevant environment (the famous "generality problem"): do I count "perception," "perception from 50 meters," or "perception in this façade-filled area"? Depending on the choice, the verdict about reliability comes out differently.

Repair 5: Nozick's Truth-Tracking

Robert Nozick proposed an elegant counterfactual condition in 1981: knowledge is a belief that tracks the truth. Two conditions formalize this. Sensitivity: if p were false, S would not believe that p. And adherence: if p remained true in slightly altered circumstances, S would still believe it. In Gettier cases sensitivity fails: had Smith not happened to have ten coins, he would still believe (e)—his belief does not track the truth, it merely hits it by accident.

Nozick's theory is powerful and explains many cases, but it has notorious side effects—for instance, that under it knowledge is no longer reliably "inherited through logical entailment," which many philosophers consider too high a price.

Repair 6: Defeasibility – the Doctrine of Undefeated Grounds

Another major strand, represented for example by Lehrer and Paxson (1969), is the defeasibility theory. Its idea: a justified true belief is knowledge if there is no defeater—no fact that, were the believer to know it, would seriously undermine their justification. In Smith's case such a defeater exists: the fact that Jones does not get the job. If Smith knew that, his grounds for (e) would collapse. Knowledge, accordingly, is undefeated justified true belief.

Here too the same weakness returns: vagueness. How strongly must a defeater weaken the justification for knowledge to vanish? Which facts of the environment must the evidence "take into account," and which not? There is no sharp line.

Part 5: The Pattern Behind the Failure – and Williamson's Reversal

If you lay the six repairs side by side, an uncomfortable suspicion forces itself upon you. Each solves Gettier's original cases. And against each, a new, more cunning case can be constructed—or it pays with skepticism, or with a new, equally vague basic condition. The following overview sums it up.

The recurring word in the right-hand column is vagueness. It looks as though the concept "knowledge" cannot be decomposed into clean, sharply outlined building blocks without a new Gettier case always seeping in at the edge. Some epistemologists draw from this the conclusion that the project of "analyzing knowledge"—decomposing knowledge fully into simpler components—may be inherently hopeless.

It was precisely this conclusion that Timothy Williamson drew in his influential book Knowledge and Its Limits (2000), turning the whole question around. Ever since Plato, everyone had tried to assemble knowledge from simpler ingredients—belief, truth, justification. Williamson's provocative thesis: that is the mistake. Knowledge is not assembled. It is itself the simplest, most fundamental building block—a state in its own right, the "most general factive mental state," which cannot be decomposed into belief-plus-extras. In his "Knowledge First" epistemology the direction of explanation reverses: knowledge is not explained through justification, but—roughly—justification through knowledge. That Gettier cases are not knowledge is then not a defect of a definition but simply a datum that any theory must respect.

Whether Williamson is right is contested. But his reversal marks how deep Gettier's three pages dug: they did not merely refute a definition but raised the question of whether defining is even the right approach here.

Part 6: Why This Matters for Technology, Data, and Security

It would be a mistake to dismiss all this as philosophical bead-game play. The Gettier structure—justified, true, and yet only accidentally on target—appears wherever we ascribe knowledge to systems or to ourselves.

Take a monitoring or alerting system. It raises an alarm "because" CPU load is above 90 percent, and indeed the service fails shortly afterward. Was the system "right"? Sometimes yes—but sometimes the CPU load came from a harmless batch job, and the outage came from an entirely different corner (a certificate expiring at the same time). The warning was true and justified and still applied only by accident. It is precisely this Gettier-like "lucky hit" that is why one evaluates alerts not only by whether they fire in an emergency, but by whether they fire for the right reason. Sensitivity and adherence in Nozick's sense are here not philosophy but alert quality.

Or think of large language models. An LLM produces a sentence that happens to be true but stems from a "false premise" within—from a confabulation that coincidentally aligns with the truth. Here the parallel to Gettier's No False Lemmas cases is striking: a true result derived from something unreliable. The question of whether a model "knows" something or merely generates a plausible, accidentally correct string is at bottom the same question Gettier asked in 1963. That is exactly why it is not enough to measure a model only by its hit rate; one wants to know whether the right answer comes from the right internal "reason"—a question that leads directly into mechanistic interpretability (see The Ghost in the Machine: How to Read a Neural Network From the Inside).

And in the compliance and audit context: when a report finds that an organization "knew" about a risk, a full concept of knowledge is implicitly at work. Is it enough that somewhere in the company a true, plausibly grounded document existed—even if the connection between that document and the actual danger was only coincidental? Lawyers and auditors wrestle, without calling it that, constantly with Gettier questions: when is knowing genuine knowing and when merely a retrospectively justified lucky hit?

The shared lesson runs: truth plus good reasons does not suffice if the connection between the two is brittle. Robustness—that the belief would still be true even if circumstances were slightly different—is the real difference between knowledge and luck. That is an intuition an engineer grasps at once.

The Central Takeaway

The central, practically usable insight of the Gettier problem is not the depressing observation that after 2,400 years we still cannot cleanly define "knowledge." It is more constructive: a belief can be true and well grounded and still worthless, because it is true only by luck—and the difference between knowledge and luck lies in the robustness of the connection between your reasons and reality.

Translate that into your everyday life, technical and personal. When your test is green—is it green because the code is correct, or by accident (for instance because it never runs the faulty section)? When your architecture decision has proven itself—did it prove itself for the reasons you cited, or did an unobserved variable rescue you? When a forecast came true—did it come true because your model captures reality, or because two errors happened to cancel out? The Gettier lens trains a particular skepticism: ask not only "Is it true?" and "Did I have good reasons?" but "Is it true because my reasons were good—or would I have been wrong too, had the world been only slightly different?" Whoever checks this way distinguishes stable knowledge from fragile lucky hits—and that is exactly the difference that counts when it matters.

Reflection Question

Think of an important belief you rely on professionally—an architectural assumption, a security measure, an assessment of a system or a person. It is (so you assume) true, and you have good reasons for it. But now ask yourself the Gettier inverse: In how many slightly altered versions of the world would your belief still be true—and in how many would it be saved only by the same lucky coincidence that happens to hold right now? And if, on honest reflection, you notice the connection is thinner than you thought: what would you have to measure or test to turn a lucky hit into genuine knowledge?

Cross-References in the Vault

• The Ghost in the Machine: How to Read a Neural Network From the Inside – The question of whether an AI model "knows" or is merely accidentally right demands a look at its inner mechanisms; a direct technical echo of the Gettier question.
• Spooky Action at a Distance: Quantum Entanglement from Einstein to the Quantum Internet – This too is about the limits of what we can know with certainty about the world when correlation and cause come apart.
• The Cosmic Tension: Why the Universe Seems to Have Two Rates of Expansion – A case in which two well-justified beliefs contradict each other—the empirical variant of the question of when justification truly holds.

Sources

• Edmund L. Gettier (1963): Is Justified True Belief Knowledge? In: Analysis 23(6), pp. 121–123. Original PDF: https://www.ditext.com/gettier/gettier.html
• The Analysis of Knowledge, Stanford Encyclopedia of Philosophy: https://plato.stanford.edu/entries/knowledge-analysis/
• Gettier Problems, Internet Encyclopedia of Philosophy: https://iep.utm.edu/gettier/
• Reliabilist Epistemology, Stanford Encyclopedia of Philosophy: https://plato.stanford.edu/entries/reliabilism/
• Gettier problem, Wikipedia (overview and case variants): https://en.wikipedia.org/wiki/Gettier_problem
• Timothy Williamson (2000): Knowledge and Its Limits. Summary of the "Knowledge First" position: https://iep.utm.edu/knowledge-first-theories-of-justification/

← All articles