209 – Knowledge vs. belief
The most important distinction we are probably not making
In ordinary conversation, the words “know” and “believe” are used almost interchangeably. A person who says “I know the restaurant is on the left” and a person who says “I believe the restaurant is on the left” are taken to be expressing different degrees of confidence about the same kind of thing: the first more certain than the second, but both making a claim about a matter of fact that could in principle be checked. The distinction, on this everyday understanding, is a matter of degree: knowledge is confident belief, belief is hesitant knowledge, and the line between them is drawn by how sure we feel.
This understanding is wrong, not in a technical philosophical sense that can be safely ignored by anyone without a taste for academic puzzles, but in a way that has direct and practical consequences for how we evaluate claims, how we argue, and how we decide what to act on. The difference between knowing something and believing it is not primarily a difference of degree. It is a difference of kind. And the collapse of the distinction (the treatment of strong belief as though it were knowledge, and of knowledge as though it were merely one belief among others) is responsible for a family of errors that this series has been documenting from several directions without yet confronting directly.
This article confronts it directly.
What the philosophical tradition says, and why it matters
The classical definition of knowledge was proposed by Plato in the dialogue Theaetetus and has organized the philosophy of knowledge for two and a half millennia: knowledge is justified true belief.¹ To know something is not merely to believe it: it must be true, and we must have adequate justification for believing it. Three conditions, all necessary, none alone sufficient.
The truth condition says that we cannot know something that is false. If we believe that the restaurant is on the left and it is in fact on the right, we do not know where the restaurant is; we believe something incorrect. This seems obvious, but its implications are not always followed. A very large number of things that people describe as knowledge (of politics, of history, of other people’s motivations, of the consequences of policies) turn out, on examination, to be beliefs that may or may not be true, held with a degree of confidence that outstrips the available justification.
The justification condition says that a belief does not become knowledge merely by being true. If we believe the restaurant is on the left because a fortuneteller told us so, and it happens to be on the left, we do not know where the restaurant is; we have a true belief arrived at by an unreliable method. The truth was not connected to our belief in the right way. This condition is the one most frequently violated in practice, because the feeling of confidence that attaches to a belief is systematically disconnected from the quality of the justification for it. Beliefs acquired through reliable methods can feel uncertain. Beliefs acquired through unreliable methods (through emotional resonance, social pressure, wishful thinking, or the mere repetition that produces familiarity) can feel rock-solid. The feeling is not the evidence.
The belief condition says that knowledge requires that we actually hold the position: that it be genuinely part of our cognitive repertoire, not merely something we can recite. This condition is more interesting than it appears, because it draws attention to the gap between what people say they believe and what their behavior reveals they actually believe. Peterson’s observation that we can only find out what we actually believe by watching how we act, discussed in article 212, is a claim about this gap. The belief condition of knowledge is not satisfied by the ability to assent to a proposition when asked. It requires genuine commitment, which shows up in behavior and not only in verbal report.
Together, these three conditions define knowledge in a way that is considerably more demanding than the everyday usage suggests. Most of what we describe as knowledge (of politics, of economics, of history, of medicine, of other people) is, on examination, belief in various stages of justification, ranging from the highly justified to the barely justified to the entirely unjustified. This is not a counsel of despair. It is a demand for precision. Knowing which of our beliefs constitute knowledge and which do not is the precondition for holding them with the calibrated confidence that the series has been recommending throughout.
The Gettier problem, and why it matters for this series
In 1963, the philosopher Edmund Gettier published a three-page paper that dismantled the classical definition of knowledge with two counterexamples so simple that they can be stated in a sentence.² Consider: we look at a clock on the wall and see that it reads 3:15. We form the belief that it is 3:15. The belief is true; it is in fact 3:15. And we have justification; we looked at the clock, which has always been reliable before. But the clock stopped exactly 12 hours ago, and we happened to look at it at the one moment in 24 hours when it shows the correct time by coincidence. Do we know that it is 3:15?
The intuitive answer is no. We have a justified true belief, but we do not know the time; the truth of our belief is connected to our justification by a coincidence that we have no access to. The Gettier problem shows that justified true belief is not sufficient for knowledge. Something more is required: some appropriate connection between the justification and the truth, some non-accidental relationship between the reliability of the method and the correctness of the conclusion.
The Gettier problem is not merely a technical puzzle for academic philosophers. Its practical import is this. A great deal of what we take ourselves to know is Gettier-knowledge: true beliefs arrived at by methods that happened to produce the right answer in this case but that are not reliably connected to truth in the relevant domain. The investor who correctly predicted the 2008 financial crisis because they were generally pessimistic about markets, rather than because they had identified the specific mechanism of failure, has a true belief with a justification, but not knowledge of the kind that would reliably generalize. The person who correctly identified a dishonest politician because they distrust all politicians, rather than because they identified specific evidence of dishonesty, has a true belief with a justification, but not knowledge that would reliably distinguish honest from dishonest politicians in future cases. In each instance, the belief was true and the person had reasons for holding it, but the connection between the reasons and the truth was accidental in the way that the stopped-clock case is accidental, and the method would not be reliable in the next case.
This is the domain in which intuition and track record diverge in the specific way that Philip Tetlock’s research on expert forecasting documented: some people have good track records not because they have reliable methods but because their characteristic error patterns happened to align with the direction of events for a period. When the events change direction, the Gettier knowledge collapses because it was never genuine knowledge to begin with. The method was not connected to the truth in the right way.
The spectrum of justification
If the truth condition and the Gettier problem suggest that knowledge is rarer than ordinary usage implies, the justification condition adds the further complication that justification is not binary: it is not either present or absent, but comes in degrees that vary continuously across a very wide range.
At one end of the spectrum are claims whose justification is so strong that it is difficult to imagine what further evidence could add: that the Earth is approximately 4.5 billion years old, that the germ theory of disease correctly describes the transmission of infectious illness, that the Holocaust occurred. These claims are supported by multiple independent lines of evidence, tested by methods whose reliability is itself well established, and consistent with everything else we know about the relevant domains. Their justification approaches what we ordinarily mean by certainty, and revising them would require revising so much of what supports them that the cost would be enormous.
At the other end are claims supported by nothing but the feeling that they are true, the say-so of a single unreliable source, or the motivated reasoning that the previous articles have documented. Between these extremes lies a continuous spectrum of belief states (some closer to genuine knowledge, some closer to bare opinion) and the practice that this series recommends is the calibration of confidence to position on that spectrum. The belief whose justification is strong warrants high confidence. The belief whose justification is weak warrants low confidence, regardless of how strongly it is felt.
The practical difficulty is that the feeling of confidence is not a reliable guide to position on the justification spectrum. The mechanisms that produce high confidence are largely independent of the mechanisms that produce good justification. High confidence comes from repetition, emotional resonance, social reinforcement, narrative coherence, and the familiarity that the knowledge illusion (article 202) mistakes for understanding. A belief can be strongly felt and weakly justified. A belief can be weakly felt (held tentatively, with genuine uncertainty) and strongly justified. The calibration the series recommends requires distinguishing the feeling from the justification, which requires asking not “how confident do I feel?” but “what is my evidence, and how reliable is the method that produced it?”
Why beliefs arrive the way they do
The philosophical account of knowledge as justified true belief describes what knowledge would look like if it were properly constituted. It does not describe how beliefs actually form in the minds of human beings. The gap between these two accounts is the subject of much of this series, but it is worth stating the gap precisely in the context of this article.
Beliefs, in the actual human case, do not typically arrive as conclusions of evidential assessment. They arrive as products of experience, social influence, emotional response, narrative coherence, and the specific biases documented throughout the series. The child who grows up in a religious household and develops religious beliefs, the young person who adopts the political positions of their social circle, the professional who comes to believe that the methods of their discipline are the most reliable methods available for the questions they care about: none of these beliefs is arrived at primarily through the weighing of evidence. They arrive through processes that have their own logic, their own reliability, and their own characteristic failure modes.
This is not, in itself, a problem. The social transmission of beliefs (the adoption of positions that one’s community has developed and refined) is one of the most powerful mechanisms available for rapidly acquiring a large body of approximately reliable beliefs about the world. Cultural learning, as the anthropologist Joseph Henrich has argued, is the mechanism through which human beings access the accumulated cognitive work of previous generations, and it is responsible for most of what any individual knows.³ The problem arises not in the acquisition of beliefs through social and experiential processes but in the failure to subject those beliefs, once acquired, to the kind of retrospective justification that would allow the well-supported to be distinguished from the poorly supported.
What happens in practice is that beliefs acquired through social and experiential processes are held with a degree of confidence that reflects the vividness of the experience or the strength of the social pressure rather than the quality of the justification. The person who witnessed an event feels certain about what they saw, even when eyewitness memory is notoriously unreliable. The person whose community unanimously holds a position feels certain about it, even when social unanimity is a product of selective exposure rather than independent confirmation. The person who has held a belief for a long time (who has organized a large part of their experience around it and used it to make decisions that turned out well) feels certain about it, even when the belief’s track record is confounded with the many other factors that contributed to the good outcomes.
The feeling of certainty, in each case, is real. It is a genuine psychological state that influences behavior, shapes attention, and resists revision. What it is not is evidence of good justification. And the series’ central practice (holding beliefs with confidence proportional to the evidence) requires maintaining the distinction between the feeling and its object, between how certain one is and how certain one ought to be.
The specific danger of treating strong belief as knowledge
There is a specific practical danger in the conflation of strong belief with knowledge that goes beyond the personal epistemic costs. It is the social and political danger of treating one’s own strong beliefs as a ground for overriding the legitimate interests of people who hold different strong beliefs.
If one knows something, then the person who denies it is in error, and it is at least arguable that acting on what one knows (even against the preferences of those who are wrong) is justified by the fact that one is right and they are not. Knowledge, in this usage, is a trump card: it ends the conversation about whose preferences should prevail, because one party has a claim to truth that the other lacks. If one merely believes something, even strongly, then the person who believes otherwise has the same epistemic standing one does, and the question of whose preferences should prevail cannot be settled by appealing to the superior epistemic status of one’s belief.
The conflation of strong belief with knowledge is therefore not merely an intellectual error. It is a move that has political and moral consequences: that transforms a legitimate disagreement between people with different but equally well-founded beliefs into a conflict between knowledge and error, between the enlightened and the mistaken, between those who see correctly and those who need to be educated or overridden. The history of the twentieth century contains numerous catastrophic examples of political projects organized around the claim that one group possessed knowledge that others lacked (knowledge of the direction of history, knowledge of the nature of the good society, knowledge of the class or race or religion whose interests were identical with the interests of humanity) and that this knowledge justified overriding the preferences of those who disagreed, for their own benefit and for the benefit of all.
A more mundane but equally instructive example is provided by the investment management industry. Professional fund managers (people who have spent years studying financial markets, who possess advanced degrees in finance and economics, who have access to sophisticated analytical tools and proprietary data) hold with great conviction the belief that their expertise enables them to identify stocks that will outperform the market. This belief is the foundation of an industry that manages tens of trillions of dollars and charges fees commensurate with the value of the expertise it claims to provide. The belief is also, in the aggregate, wrong in a specific and well-documented way.
The observation was formalized by the economist Burton Malkiel in his 1973 book A Random Walk Down Wall Street, where he proposed that a blindfolded monkey throwing darts at a newspaper’s financial pages could select a portfolio that would perform as well as one carefully chosen by experts.⁴ The Wall Street Journal tested the claim directly from 1988 onwards in a series of Dartboard Contests, pitting professional fund managers against journalists throwing darts at stock tables. Over 142 six-month contests, the professionals did come out slightly ahead, but only by a margin that disappeared almost entirely once the higher risk of the professionals’ portfolios was accounted for. Choosing riskier stocks produces higher average returns without requiring any analytical skill, and the professionals had systematically chosen riskier stocks. Their edge was not expertise. It was risk appetite.
The most rigorous version of the test was conducted by Rob Arnott and colleagues at Research Affiliates in 2012, who simulated 100 random monkey portfolios of 30 stocks each, drawn from the 1,000 largest stocks by market capitalization, and tracked their performance from 1964 to 2010. Ninety-six of the 100 random portfolios outperformed the market index over that period, by an average of 1.7 percent per year.⁵ The structural reason is precise: market-cap-weighted indices like the S&P 500 disproportionately weight the largest and most expensive stocks, which tend to appreciate more slowly than smaller stocks. A random portfolio naturally includes more smaller stocks and therefore outperforms the index not because of skill but because of a structural feature of how the index is constructed. The monkey does not win because the monkey is clever. The monkey wins because the benchmark against which professional expertise is measured has been constructed in a way that makes it relatively easy to beat.
The knowledge claim of the fund management industry is not merely that managers can select good stocks. It is that their analytical expertise produces returns that justify their fees: returns above what a randomly selected portfolio would achieve. The evidence, aggregated across the industry over decades, does not support this claim. The strong belief persists because the professional identity, the fee structure, the institutional apparatus, and the client relationships of the entire industry are built on it. Revising the belief would require dismantling the edifice. The belief is not maintained because the evidence supports it. It is maintained because too much depends on it, which is the definition of a crony belief, held not for epistemic reasons but for the social and institutional costs of abandoning it.
These are instructive cases across very different scales. But the underlying move (from strong belief to claimed knowledge, and from claimed knowledge to the justification of overriding others) is not confined to either extreme. It appears wherever the confidence that attaches to a belief outstrips the justification for it, and wherever that confidence is used to dismiss the legitimate perspectives of people who hold different beliefs. The practice this series calls The Conscious Look is partly a practice of resisting this move: of maintaining the distinction between what one strongly believes and what one genuinely knows, and of engaging with those who believe differently on the assumption that they may be seeing something that one’s own framework misses.
The practical application: two questions
The argument of this article reduces to two questions that can be asked of any belief held with confidence. The first is the justification question: what is my evidence for this, and how reliable is the method that produced it? The second is the Gettier question: even if my belief is true and I have reasons for holding it, is the connection between my reasons and the truth the kind of connection that would reliably produce correct beliefs in similar cases?
The justification question surfaces the gap between felt certainty and evidential warrant: between how confident one is and how confident the evidence warrants being. The Gettier question surfaces the gap between having reasons and having reliable reasons: between the belief being true and the method being the kind of method whose outputs can be trusted across a range of cases.
Neither question is comfortable to apply to beliefs one holds with conviction. The discomfort is the point. The Conscious Look, applied to the knowledge-belief distinction, is the practice of asking these questions seriously, not as an exercise in universal doubt, which leads nowhere, but as a specific diagnostic for identifying where genuine knowledge ends and strongly felt belief begins. The boundary is not always clear. It is always worth looking for.
Further reading
Plato’s Theaetetus, in any modern translation, is the original and still the most searching examination of what knowledge is: written in the form of a dialogue that explicitly refuses to arrive at a settled definition, and that is more honest about the difficulty of the question than most subsequent treatments. The Meno and the Phaedo develop related themes and are equally readable.
Edmund Gettier’s “Is Justified True Belief Knowledge?” published in Analysis in 1963, is three pages long and freely available online. Reading it is recommended: the counterexamples are stated simply, the argument is clear, and the experience of watching a two-and-a-half-thousand-year-old consensus dissolve in three pages is instructive in itself.
Paul Boghossian’s Fear of Knowledge: Against Relativism and Constructivism (2006) defends the objectivity of knowledge against the relativist and constructivist positions that deny it: arguing that the claim “there is no objective knowledge” is self-undermining and that the rejection of the knowledge-belief distinction in the name of epistemic humility produces, in practice, the opposite of humility. It is short, precise, and philosophically rigorous without being inaccessible.
Burton Malkiel’s A Random Walk Down Wall Street (1973, updated regularly) is the original and most readable account of the efficient market hypothesis and its implications for investment management, including the blindfolded monkey metaphor that has organized subsequent empirical testing of the claim that professional expertise produces returns above what random selection would achieve.
Joseph Henrich’s The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter (2015) provides the evolutionary and anthropological framework for understanding why social transmission of belief is not simply a distortion of ideal rational inquiry but a powerful and largely reliable mechanism for accessing the accumulated cognitive work of previous generations, and for understanding where that mechanism fails.
Philip Tetlock and Dan Gardner’s Superforecasting: The Art and Science of Prediction (2015) provides the empirical foundation for the calibration argument: what it actually looks like to hold beliefs with confidence proportional to evidence, and how the practice of tracking one’s predictions against outcomes produces a progressive alignment of felt certainty with actual justification.
Notes
¹ Plato’s formulation of knowledge as justified true belief appears most explicitly in the Theaetetus (201c-210b), though the analysis there is presented as a proposal that Socrates then proceeds to examine and ultimately reject. The formulation that has passed into philosophical currency (“knowledge is justified true belief”) is a reconstruction from the Platonic text rather than a direct quotation, and Plato himself did not regard it as a satisfactory definition. The subsequent tradition that treated it as the standard analysis of knowledge was therefore, in a specific sense, more confident in the definition than its original source warranted. This is itself an instance of the knowledge-belief distinction being violated: the philosophical tradition believed it knew what Plato had established, when Plato had explicitly declined to establish it.
² Gettier, E. L. (1963). Is justified true belief knowledge? Analysis, 23(6), 121-123. The paper generated an enormous subsequent literature attempting to add a fourth condition to the justified true belief analysis that would block Gettier-style counterexamples. The most prominent proposals include the reliability condition (the belief must be produced by a reliable cognitive process), the no-false-lemmas condition (the justification must not depend on any false intermediate belief), and the safety condition (the belief could not easily have been false given the same evidence). None of these proposals has achieved consensus, and the epistemological community has largely concluded that the justified true belief analysis cannot be repaired by simple addition and that a more fundamental reconception of knowledge may be required. For the purposes of this series, the important point is the practical one: that having reasons for a belief is not the same as having reasons that are reliably connected to truth in the relevant domain.
³ Henrich, J. (2015). The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter. Princeton University Press. Henrich’s argument that human cognitive superiority rests primarily on cultural learning rather than individual intelligence has direct implications for the epistemology of belief: it means that most of what any individual believes has been socially transmitted rather than individually derived, and that the reliability of those beliefs depends on the reliability of the cultural transmission process rather than on the reliability of individual reasoning. The cultural transmission process is in general highly reliable for beliefs about the practical management of the natural and social environment: the accumulated wisdom of generations of people dealing with similar problems. It is less reliable for beliefs about politically contested questions, where the social transmission process is shaped by the interests and biases of those who control the transmission, and for beliefs about novel situations for which the accumulated wisdom has no established template.
⁴ Malkiel, B. G. (1973). A Random Walk Down Wall Street. W. W. Norton and Company. The book introduced the efficient market hypothesis to a general audience and remains in print in regularly updated editions. The central claim (that stock prices already incorporate all available information, making consistent outperformance through analysis impossible) was controversial in 1973 and remains so, though the evidence has consistently supported the weak and semi-strong forms of the hypothesis. The Wall Street Journal‘s Dartboard Contest, begun in 1988 and running for 142 contests until 2002, was directly inspired by Malkiel’s challenge. The contest’s results (professionals winning 61 of the first 100 contests against dart-picked portfolios, but with a margin that shrank significantly once risk differences were controlled for) are reviewed in Liang, B. (1999). Price pressure: Evidence from the Dartboard Column. Journal of Business, 72(1), 119-134.
⁵ The Research Affiliates simulation is described in Arnott, R. D., Hsu, J., Kalesnik, V., and Tindall, P. (2013). The surprising alpha from Malkiel’s monkey and upside-down strategies. Journal of Portfolio Management, 39(4), 91-105. The finding that 96 of 100 random portfolios beat the cap-weighted market index over a 46-year period is less a refutation of the efficient market hypothesis than a demonstration that the standard market-cap-weighted index is a structurally biased benchmark. The monkey does not outperform because markets are inefficient; it outperforms because the benchmark underweights smaller stocks, which have historically produced higher returns. The practical implication (that low-cost equal-weight index funds outperform most actively managed funds) is now widely accepted and is reflected in the ongoing flow of assets from active to passive management.