The Sources of Knowledge

“We can know more than we can tell.”

— Michael Polanyi, The Tacit Dimension (1966)

Where does knowledge come from? Classical Indian philosophers called the sources of knowledge pramāṇas and argued about how many there are (see Chapter 2). Western philosophers have usually listed perception, memory, introspection, reason, and testimony, with inference as the way we extend knowledge from these sources.

Each source is generally reliable and each fails in characteristic ways. A critical thinker needs to know both. When someone says “I saw it with my own eyes,” “I remember it clearly,” “I just know what I feel,” “It’s obvious,” or “I read it in a reputable source,” they are appealing to one of these sources, and each appeal can be evaluated.

Basic and derived sourcesListen

Robert Audi (Epistemology, 3rd ed. 2011) distinguishes:

• Basic sources: sources that can produce knowledge or justification without depending on another source. Audi lists perception, memory, introspection (consciousness), and reason.
• Derived sources: sources whose deliverances depend on other sources. On Audi’s view, testimony is derived: to learn from what you tell me, I must perceive your words.

Another useful distinction is between generative sources, which produce new knowledge (perception, reason), and preservative sources, which keep or pass on knowledge already acquired (memory, and on some views testimony). Whether memory and testimony can ever generate knowledge is debated below.

PerceptionListen

Perception is the most basic source of knowledge about the physical world. Philosophers ask two questions about it: What do we directly perceive? and How does perception justify beliefs?

Naive and direct realism

Naive realism is the commonsense view: we perceive physical objects directly, and they are as they appear, colors and all.

Direct realism keeps the core idea, that we perceive physical objects themselves and not some inner intermediary, but drops the claim that objects are always as they appear. Perception involves complex physical and neural processes, but what we are aware of is the object, not the process. Thomas Reid defended direct realism against the “way of ideas” (see Reid and common sense), and it has many defenders today.

The argument from illusion and hallucination

Two classic arguments try to show that we do not perceive physical objects directly.

The argument from illusion. A straight stick half submerged in water looks bent. What you see is bent. The stick is not bent. So what you see is not the stick but something else: a sense-datum, an image in the mind. Bertrand Russell opens The Problems of Philosophy (1912) with a version: the color of his table seems to vary with the light and his position, so “the colour is not something which is inherent in the table, but something depending upon the table and the spectator and the way the light falls on the table.” A. J. Ayer (The Foundations of Empirical Knowledge, 1940) and H. H. Price (Perception, 1932) developed the theory of sense-data.

The argument from hallucination. Macbeth hallucinates a dagger. His experience can be subjectively indistinguishable from seeing a real dagger. In hallucination, there is no dagger, so he is aware of something else, perhaps a mental image. Since the veridical experience is indistinguishable from the hallucination, it is natural to conclude that the same kind of thing is its immediate object. (This is sometimes called the “spreading step” or the “common factor” assumption.)

The time-lag argument. When you look at a distant star, you see light that left it thousands of years ago. The star may no longer exist. So what you see, it is argued, is not the star as it is now, but something else.

J. L. Austin (Sense and Sensibilia, 1962) attacked the argument from illusion. When the stick looks bent, we see a stick that looks bent. Nothing bent is seen. The argument trades on running together “looks bent” and “is a bent-looking thing.” Austin also insisted that illusions (where a real object is misperceived) differ fundamentally from hallucinations (where there is no object), and that lumping them together distorts the analysis.

Indirect realism and sense-data

Indirect realism (representative realism) holds that we perceive physical objects, but only indirectly, by being directly aware of mental representations (ideas, sense-data, percepts) that are caused by and represent those objects. Descartes, Locke, and Russell held versions of this view.

Indirect realism fits well with science: light reflected from objects enters the eye, is transduced into neural signals, and processed by the brain into a representation. But it has a serious epistemological problem, the veil of perception: if all I directly perceive are my own representations, how can I know that they resemble, or even correspond to, anything outside? I can never step outside my representations to compare them with the world. Berkeley exploited this to argue for idealism, and it underlies Cartesian skepticism.

Idealism and phenomenalism

If we only ever perceive ideas, why believe in anything beyond them?

• Berkeley’s idealism: physical objects just are collections of ideas, sustained by God’s perception. See Berkeley.
• Phenomenalism (John Stuart Mill, An Examination of Sir William Hamilton’s Philosophy, 1865; later A. J. Ayer): physical objects are “permanent possibilities of sensation.” “There is a table in the next room” means roughly “If I were to go into the next room, I would have table-like experiences.” Phenomenalism was abandoned in the mid-20th century because the translations could never be completed, and the conditions (“if I were to go into the next room”) themselves refer to physical objects.

Disjunctivism

Disjunctivism (J. M. Hinton, 1967; Paul Snowdon; John McDowell, “Criteria, Defeasibility, and Knowledge,” 1982; M. G. F. Martin) rejects the “common factor” assumption. A veridical perception and a matching hallucination may be subjectively indistinguishable, but they are fundamentally different kinds of mental state. When you see a dagger, you are related to the dagger itself; when you hallucinate, you are not. What they have in common is only a disjunction: “either I am seeing a dagger, or it merely seems to me that I am.”

Epistemological disjunctivism (McDowell; Duncan Pritchard, Epistemological Disjunctivism, 2012) applies this to justification. In the good case, your reason for believing there is a dagger is that you see that there is a dagger, a reason that guarantees truth (it is factive). In the bad case, you have only the weaker reason that it seems to you there is a dagger. This view offers a distinctive response to skepticism. See Epistemological disjunctivism.

Other theories of perceptual experience include adverbialism (Roderick Chisholm, C. J. Ducasse), which says that seeing red is sensing “redly,” not being aware of a red object, and intentionalism or representationalism (Gilbert Harman, Michael Tye), which says that experiences represent the world as being a certain way, as beliefs do, and can be accurate or inaccurate.

Primary and secondary qualities

Galileo (The Assayer, 1623) wrote that tastes, odors, and colors “reside only in the consciousness”; if the living creature were removed, all these qualities would be annihilated. Robert Boyle and John Locke developed the distinction:

• Primary qualities (shape, size, motion, number, solidity) are in objects as we perceive them.
• Secondary qualities (color, sound, taste, smell, warmth) are powers in objects to produce experiences in us. The experiences do not resemble anything in the objects.

Locke’s argument: put one hand in hot water and the other in cold, then put both in lukewarm water. The same water feels warm to one hand and cold to the other. It cannot be both warm and cold, so warmth is not in the water as it is felt.

Berkeley replied that the same relativity applies to primary qualities (an object looks larger up close), so the distinction collapses. Modern color science has kept the debate alive. Color physicalists identify colors with surface reflectance properties; dispositionalists identify them with powers to cause experiences; eliminativists say nothing in the world is literally colored.

A vivid modern example is “the dress”, a photograph that went viral in 2015. Some people saw a white and gold dress, others blue and black (it was physically blue and black). The disagreement arises because the brain automatically corrects for the color of the ambient light (color constancy), and people’s brains made different assumptions about the lighting in the ambiguous photograph. Perception is not a passive recording. It is an interpretation.

Cognitive penetration and theory-ladenness

Is what you perceive affected by what you believe, expect, or want?

Theory-ladenness. Norwood Russell Hanson (Patterns of Discovery, 1958) asked us to imagine Tycho Brahe and Johannes Kepler watching the sunrise. Tycho believed the Sun orbits the Earth; Kepler believed the Earth orbits the Sun. Do they see the same thing? Hanson suggested that in a sense they do not: observation is shaped by the concepts and theories the observer brings to it. Thomas Kuhn compared scientific revolutions to gestalt switches, like the duck–rabbit figure, which can be seen as a duck or as a rabbit but not both at once. See Theory-ladenness of observation.

Modularity. Jerry Fodor (The Modularity of Mind, 1983) argued that early perception is informationally encapsulated: it is not influenced by beliefs. His evidence was illusions like the Müller-Lyer illusion (two lines of equal length, one with arrowheads pointing in and one with arrowheads pointing out, look unequal). Even after you measure them and know they are equal, they still look unequal. Your belief does not penetrate your perception.

Cognitive penetration. Others argue that higher-level states can affect perception. Susanna Siegel’s example (see Phenomenal conservatism) of Jack, whose fear that Jill is angry makes her face look angry to him, raises the question of whether such penetrated perceptions can justify beliefs. Empirical research is contested. One well-known line of research concerns racial bias in split-second judgments: in studies such as B. Keith Payne’s (2001), participants primed with Black faces more often misidentified tools as guns under time pressure. Whether such effects occur in perception itself or in later judgment is debated, but either way, the lesson is that “I saw it” is not immune to bias.

Perception in practice: how the senses mislead

Perception is generally reliable for medium-sized objects in good conditions. But it is constructive. Hermann von Helmholtz (1867) described perception as unconscious inference: the brain infers the most likely causes of its sensory input. Modern predictive processing theories (such as Andy Clark, “Whatever Next?,” 2013) develop this idea: the brain constantly predicts its input and updates on the error. Mostly, this makes perception fast and accurate. Sometimes it produces systematic errors:

• Inattentional blindness. In Daniel Simons and Christopher Chabris’s famous 1999 experiment, participants counted basketball passes in a video. About half failed to notice a person in a gorilla suit walking through the scene and thumping their chest. We see far less than we think we do.
• Change blindness. People often fail to notice large changes in a scene if the change occurs during a brief interruption, such as a blink or a cut. In some experiments, people giving directions failed to notice that the person they were talking to had been replaced by someone else.
• The McGurk effect (McGurk and MacDonald, 1976). When you watch a video of someone saying “ga” dubbed with the sound “ba,” you typically hear “da.” What you see changes what you hear.
• Expectation. Wine tasters described a white wine dyed red using the vocabulary of red wines (Morrot, Brochet, and Dubourdieu, 2001).
• Conditions. Distance, lighting, stress, speed, and the presence of a weapon (“weapon focus”) all degrade eyewitness perception.

MemoryListen

Memory lets us retain what we have learned. Without it, we could know nothing beyond the present moment. Psychologists distinguish (following Endel Tulving, 1972):

• Episodic memory: memory of particular events you experienced (“I remember my first day at school”).
• Semantic memory: memory of facts (“I remember that Paris is the capital of France”).
• Procedural memory: memory of how to do things (riding a bicycle), closely related to knowledge-how.

Is memory generative or merely preservative? The traditional view, preservationism (Tyler Burge, “Content Preservation,” 1993), says memory only preserves knowledge and justification acquired earlier; it cannot create new knowledge. Jennifer Lackey (“Memory as a Generative Epistemic Source,” 2005) argued that memory can generate knowledge: for example, if a defeater that blocked your knowledge at the time of learning has since disappeared, you may come to know through memory something you did not know originally.

How reliable is memory? Research since Frederic Bartlett’s Remembering (1932) shows that memory is reconstructive, not reproductive. We do not replay a recording; we rebuild the memory each time from fragments, filling gaps with expectations, knowledge, and later information.

• Bartlett’s “War of the Ghosts” (1932). English participants who retold an unfamiliar Native American folk tale gradually changed it to fit their own cultural expectations, dropping the supernatural elements and rationalizing the plot.
• The misinformation effect. In Elizabeth Loftus and John Palmer’s 1974 study, participants watched a film of a car accident. Those asked “How fast were the cars going when they smashed into each other?” estimated higher speeds than those asked about the cars that “hit” each other, and a week later were more likely to “remember” broken glass, though there was none.
• Implanted memories. In the “lost in the mall” study (Loftus and Pickrell, 1995), a minority of participants came to remember, sometimes in detail, a childhood event of being lost in a shopping mall that had never happened, after family members “confirmed” it.
• Flashbulb memories. People feel extremely confident about where they were when they heard shocking news. Ulric Neisser and Nicole Harsch (1992) asked students the day after the 1986 Challenger explosion how they had heard the news, then asked again about two and a half years later. Many later accounts were substantially different, yet participants were confident in them. Studies after 9/11 (Talarico and Rubin, 2003) found that flashbulb memories decay like ordinary memories while confidence stays high.
• Eyewitness error. Mistaken eyewitness identification has been a factor in roughly 70% of the wrongful convictions in the United States later overturned by DNA evidence, according to the Innocence Project.

An important nuance: the relationship between confidence and accuracy depends on conditions. John Wixted and Gary Wells (2017) showed that an eyewitness’s initial confidence, recorded at a fair, properly administered first identification, is quite diagnostic of accuracy. Confidence expressed later, after feedback, repetition, and rehearsal, is much less so, because confidence inflates over time.

Philosophically, trust in memory raises the same issue as trust in perception: we cannot check memory without relying on memory. Richard Foley (Intellectual Trust in Oneself and Others, 2001) argues that some degree of self-trust in our basic faculties is unavoidable and rational, though it must remain open to correction. Russell’s skeptical five-minute hypothesis, that the world might have been created five minutes ago complete with our memories, is discussed in Chapter 8.

Introspection and self-knowledgeListen

Introspection is the faculty by which we know our own current mental states: thoughts, feelings, sensations, desires. Descartes held that the mind is transparent to itself: we have privileged access to our own minds, and our judgments about them are especially secure.

There is something right about this. If I sincerely say “I have a headache,” you are in no position to correct me. This first-person authority is part of how we use mental concepts.

But psychology has shown that introspection is far less reliable than it feels, especially about the causes of our mental states and actions.

• Telling more than we can know. Richard Nisbett and Timothy Wilson’s classic 1977 paper reported experiments in which people’s choices were influenced by factors they did not mention and denied when asked. In one study, shoppers asked to choose among four identical pairs of stockings displayed in a row tended to pick the one on the right. Asked why, they cited the quality of the stockings; asked directly whether position had influenced them, they denied it. Nisbett and Wilson concluded that when people report the reasons for their behavior, they often use a priori causal theories about what would plausibly cause such behavior, rather than direct introspective access.
• Choice blindness. Petter Johansson, Lars Hall, and colleagues (2005) showed participants pairs of photographs of faces and asked them to choose the more attractive. Through sleight of hand, the experimenters sometimes handed over the photo that had not been chosen. Most participants failed to notice the switch and went on to give reasons for “their choice,” sometimes citing features present only in the photo they had rejected.
• Confabulation. Neuropsychological patients, such as split-brain patients studied by Michael Gazzaniga, readily invent explanations for actions caused by processes they have no access to, without any awareness that they are inventing.
• Even current experience. Eric Schwitzgebel (“The Unreliability of Naive Introspection,” 2008) argues that people are often wrong even about the character of their current conscious experience, for example about how much of the visual field is clear and colored, or whether they think in words.

Timothy Wilson’s Strangers to Ourselves (2002) summarizes the research: much of the mind’s work is inaccessible to consciousness, and our explanations of our own behavior are often constructions.

How do we know our own beliefs? Gareth Evans (The Varieties of Reference, 1982) noted that if someone asks you “Do you think there will be a third world war?”, you answer by thinking about the world, not by looking inward: “in making a self-ascription of belief, one’s eyes are, so to speak, or occasionally literally, directed outward.” This transparency method (developed by Richard Moran, Authority and Estrangement, 2001, and Alex Byrne, Transparency and Self-Knowledge, 2018) explains first-person authority about belief without an inner eye.

Reason and the a prioriListen

Reason gives us a priori knowledge: knowledge that does not depend on experience beyond what is needed to understand the proposition (see A priori and a posteriori). Candidates include:

• Logic: “If p and q, then p.”
• Mathematics: “7 + 5 = 12.”
• Conceptual truths: “All bachelors are unmarried,” “If A is earlier than B and B is earlier than C, then A is earlier than C.”
• Apparent necessary truths about the world: “Nothing can be red and green all over at the same time.”
• Perhaps some moral truths: “Torturing someone purely for fun is wrong.”

Rationalism and empiricism about the a priori

Rationalists hold that there is substantive a priori knowledge, grasped by rational intuition or insight. Laurence BonJour (In Defense of Pure Reason, 1998) argues that without rational insight we could not know even the principles of inference that empiricists rely on, so empiricism undermines itself.

Moderate empiricists (the logical positivists) accept a priori knowledge but say it is all analytic: true by definition or convention, and so saying nothing about the world. Mathematics and logic, on this view, are systems of tautologies.

Radical empiricists (Mill; Quine) deny that there is any a priori knowledge at all. For Quine, even logic and mathematics are part of the web of belief and could in principle be revised in response to experience. Hilary Putnam (“Is Logic Empirical?,” 1968) suggested that quantum mechanics might give us reason to revise classical logic.

The lesson of non-Euclidean geometry

The strongest historical case against overconfident rationalism is geometry. Kant thought Euclidean geometry was synthetic a priori: necessarily true of space and knowable by pure reason. In the 19th century, Nikolai Lobachevsky, János Bolyai, and Bernhard Riemann developed consistent non-Euclidean geometries in which, for example, the angles of a triangle do not add up to 180°. In 1915, Einstein’s general theory of relativity described physical space-time as non-Euclidean, and observations have supported it since. What seemed self-evident about the world turned out to be false.

The standard lesson is a distinction. Pure geometry (what follows from given axioms) is a priori. Applied geometry (which axioms describe physical space) is empirical. More generally: “It’s self-evident” and “It’s inconceivable that it could be otherwise” are claims to examine, not trump cards. What seems self-evident may reflect habits of thought, limited experience, or lack of imagination.

Mathematical knowledge

Mathematical knowledge poses a special puzzle, set out by Paul Benacerraf (“Mathematical Truth,” 1973). The most natural account of mathematical truth treats numbers as abstract objects existing outside space and time (Platonism). But the most natural account of knowledge requires some causal or other connection between knower and known. How can we have any connection with objects outside space and time? The rival positions are Platonism (with an account of non-causal access, such as rational intuition), nominalism or fictionalism (Hartry Field, Science Without Numbers, 1980: there are no numbers, and mathematics is a useful fiction), structuralism (mathematics is about structures, not objects), and formalism (mathematics is the manipulation of symbols by rules). The Quine–Putnam indispensability argument says we should believe in mathematical objects because our best scientific theories cannot do without them.

Moral knowledge

Can we know moral truths, and if so, how?

• Intuitionism (W. D. Ross, The Right and the Good, 1930): basic moral principles, such as prima facie duties of fidelity, non-maleficence, and justice, are self-evident to a mature, reflective mind, like mathematical axioms.
• Reflective equilibrium (Rawls): moral knowledge results from bringing principles and considered judgments into coherence. See Reflective equilibrium.
• Moral perception: we can sometimes see that an act is cruel, as we see that a surface is red.
• Anti-realism and skepticism: there are no objective moral facts to know, or we could not know them. Evolutionary debunking arguments (Sharon Street, “A Darwinian Dilemma for Realist Theories of Value,” 2006; Richard Joyce, The Evolution of Morality, 2006) argue that our moral intuitions were shaped by natural selection for reproductive fitness, not for tracking moral truth, which undermines their reliability if moral realism is true.

This debate matters for everyday argument because almost every practical dispute has a normative premise. Whatever your meta-ethics, moral claims can be defended with reasons, tested against cases, checked for consistency, and examined for bias, and moral arguments can be valid or invalid like any others.

TestimonyListen

Testimony, in the philosopher’s broad sense, is any case in which we come to believe something because someone told us: in speech, writing, a textbook, a map, a news report, a road sign, an encyclopedia, a scientific paper.

Most of what you know comes from testimony. You know your date of birth, the shape of the Earth, the existence of countries you have never visited, the events of history, the findings of science, and the meanings of most words, because others told you. C. A. J. Coady (Testimony: A Philosophical Study, 1992) showed that even our attempts to check testimony usually rely on further testimony. If knowledge by testimony is not genuine knowledge, we know very little.

The central question: what justifies believing what others tell us?

First, what counts as testimony? If someone says “I have a hoarse voice” and you hear that their voice is hoarse, you learn something by perception: the sentence “Smith got the job,” said in the same voice, would have taught you the same thing. Testimony proper is when the content of what is said does the work: you understand what the speaker says and take their word for it (Jennifer Nagel, Knowledge, ch. 6).

Locke: testimony gives only probability

At one extreme, John Locke held that testimony never gives knowledge. For Locke, knowledge requires certainty, and what others tell us is at best highly probable. If you see a man walk across a frozen lake, you know it; if someone tells you she saw it, you may reasonably judge it very likely, but you do not know it (Essay, IV.xv–xvi).

Locke recommended proportioning confidence in testimony to its fit with your own experience and to six further considerations (IV.xv.4):

1. The number of witnesses.
2. Their integrity.
3. Their skill (competence to observe).
4. Their purpose in reporting it (in the case of written testimony, the design of the author).
5. The consistency of the report’s parts and circumstances.
6. Any contrary testimony.

His most famous illustration concerns the King of Siam (IV.xv.5). A Dutch ambassador told the king that in Holland, water became so hard in cold weather that men, and even an elephant, could walk on it. The king replied: “Hitherto I have believed the strange things you have told me, because I look upon you as a sober fair man, but now I am sure you lie.” Given a lifetime of tropical experience, Locke thought the king’s response reasonable. The story is a vivid lesson in how prior plausibility shapes the evaluation of testimony (compare Hume on miracles and Evidence and confirmation), and a warning that a rational prior can still be wrong when experience is narrow.

Locke’s position is radical. It implies that you don’t know where you were born, that Antarctica exists, or that Locke himself ever lived. His reason, that testimony can always be overturned by later contrary reports, seems to prove too much, as Nagel observes: perception and memory can also be undermined later, and Locke still counted them as sources of knowledge. If the later doubt is misleading, it doesn’t show that you never knew.

Reductionism

Reductionism holds that testimonial justification reduces to other sources: perception, memory, and inference. You are justified in believing testimony only if you have independent, non-testimonial reasons to think the speaker is reliable.

David Hume stated a global version in the Enquiry (Section X): our trust in testimony is based on our experience of the “conformity of facts to the reports of witnesses.” We have found testimony generally reliable, so we trust it.

Objections. Coady argued that no individual has observed nearly enough cases to support a global inference about the reliability of testimony. Most of what anyone has checked has been checked against other testimony. And young children learn language and a vast amount of information by trusting what they are told long before they could possibly have evidence of reliability.

Elizabeth Fricker (“Against Gullibility,” 1994) proposed local reductionism: for each particular piece of testimony, the hearer needs positive reasons to trust this speaker on this topic, which can come from monitoring the speaker for signs of insincerity or incompetence. This is more plausible for adults, though critics doubt that we usually have such reasons. The two versions come apart in a simple case: lost in a strange city, you ask a passing stranger for directions. The global reductionist says you can gain knowledge, since you have a standing reason to trust testimony in general; the local reductionist says you have at best a lucky true belief, since you have no specific reason to trust this stranger.

Why might testimony call for more caution than perception? Nagel (Knowledge, ch. 6) suggests a contrast with honeybees. A bee that learns the location of nectar from another bee’s dance can fly there as if it had been there itself, a kind of “cognition by proxy.” Bee signals can be faulty, as eyes can, but bees cannot deliberately deceive. Human informants are free agents with their own purposes, and some of them lie. On the other hand, perhaps we are more careful than we seem. Research on epistemic vigilance (Dan Sperber and colleagues, “Epistemic Vigilance,” 2010; Hugo Mercier, Not Born Yesterday, 2020) suggests that people constantly, and mostly unconsciously, monitor speakers for competence and honesty and check what they are told against what they already believe. Even young children prefer informants who have been accurate in the past.

Anti-reductionism

Anti-reductionism (also called the direct or default view) holds that testimony is a basic source of justification in its own right, like perception and memory. The view is ancient. In the Nyāya Sūtra (c. 2nd century CE), Akṣapāda Gautama treated testimony (śabda) as a distinct means of knowledge, not a form of inference: we do not reason “Lee said Smith got the job; Lee is reliable; so Smith got the job,” but come to know it simply by understanding what a knowledgeable speaker says. The Nyāya tradition added that knowledge can come from anyone who knows and intends to share it, sage or foreigner alike (Nagel, Knowledge, ch. 6). Thomas Reid argued that we have a natural principle of credulity (a disposition to believe what we are told) matched by a principle of veracity (a disposition to tell the truth), both given by nature.

Tyler Burge (“Content Preservation,” 1993) proposed the Acceptance Principle: “A person is entitled to accept as true something that is presented as true and that is intelligible to him, unless there are stronger reasons not to do so.” Testimony is justified by default, and the default can be defeated.

Objection: gullibility. Doesn’t this license believing whatever strangers on the internet say, as long as you have no specific reason to doubt them? Anti-reductionists reply that we usually do have defeaters available, from background knowledge about the source, the implausibility of the claim, or signs of insincerity, and that we are obliged to heed them.

Hybrid and interpersonal views

Jennifer Lackey (Learning from Words, 2008) proposed a dualist view that combines the insights of both sides. Testimonial justification requires both that the speaker’s testimony is in fact reliable (a condition on the speaker’s side, as externalists and anti-reductionists stress) and that the hearer has some positive reasons for accepting it (a condition on the hearer’s side, as reductionists stress), though not necessarily reasons strong enough to justify the belief alone.

Interpersonal or assurance views (Richard Moran, “Getting Told and Being Believed,” 2005; Edward Hinchman, “Telling as Inviting to Trust,” 2005; Paul Faulkner) emphasize that testimony is not just evidence, like a thermometer reading. When someone tells you something, they give you their assurance, take responsibility for its truth, and invite your trust. If they are wrong, you can hold them accountable in a way you cannot hold a thermometer accountable.

Transmission and generation

Does testimony only transmit knowledge, like a pipe carrying water, or can it generate it?

The transmission view says a hearer can come to know p through testimony only if the speaker knows p. Plato’s jury example already points this way (see Plato): jurors persuaded by a lawyer who has no idea whether his client is innocent don’t come to know that he is, even if he is. Lackey describes the transmission view with the image of a bucket brigade: to pass you a full bucket of water, I must have a full bucket; and if I pass it, then, spills aside, yours will be full too. “Spills” include mishearing, and doubts, even unreasonable ones, that stop you from accepting what you’re told.

Jennifer Lackey (Learning from Words, 2008) offered a counterexample, the creationist teacher. Stella is a devoutly religious biology teacher who does not believe in evolution. But she recognizes that the scientific evidence supports it and conscientiously teaches it accurately to her students. The students come to know that humans evolved from earlier species, though Stella neither believes nor knows it. So testimony can generate knowledge the speaker lacks. What matters is that the testimony is a reliable conduit of information, not that the speaker has knowledge.

Groups can fill the bucket together. Nagel (Knowledge, ch. 6) gives the example of a Wikipedia article. The editor who first wrote a particular sentence may have been unsure of a detail, but after the sentence has been checked and corrected by many editors, the article as a whole may be reliable enough to give readers knowledge. How each theory treats this case is revealing:

• A Lockean says readers get only probable opinion.
• A reductionist says readers gain knowledge only if they have reasons to trust Wikipedia, such as the 2005 Nature study comparing its accuracy with Encyclopaedia Britannica (see Epistemic institutions) or their own experience of checking it.
• A direct theorist can say that even a naive twelve-year-old gains knowledge, as long as the site’s quality controls are actually working, though traditional versions would require some individual contributor to know the fact.

Evaluating testimony in practice

Philosophy does not give a simple rule for when to believe what you are told. But historians, judges, journalists, and intelligence analysts have developed practical criteria. When assessing testimony, ask:

1. Competence: Was the source in a position to know? (Did they witness it? Do they have relevant expertise?)
2. Sincerity: Does the source have a motive to deceive? A claim against interest is more credible. Historians call this the criterion of embarrassment: a source reporting something unflattering to itself is less likely to be inventing it.
3. Track record: Has this source been reliable before, on this kind of topic?
4. Independence and corroboration: Is the claim confirmed by sources that are genuinely independent? Ten articles repeating one press release are one source, not ten.
5. Chain length: Is this first-hand, second-hand, or tenth-hand? Information degrades along chains of transmission, as in the children’s game of “telephone.”
6. Specificity and checkability: Vague claims are cheap to make. Specific claims that could be checked are a sign of good faith.
7. Prior plausibility: How likely is the claim, given everything else you know? Hume’s argument about miracles is an application of this criterion. See Philosophical razors and heuristics.
8. Consistency: Is the testimony internally consistent, and consistent with the same source’s other statements?

Hume himself listed signs that should lower our trust: “when the witnesses contradict each other; when they are but few, or of a doubtful character; when they have an interest in what they affirm; when they deliver their testimony with hesitation, or on the contrary, with too violent asseverations.”

A cautionary tale about chains of testimony. For decades, textbooks and articles repeated that the belief that spinach is exceptionally rich in iron (popularized by the cartoon Popeye) arose from a misplaced decimal point in a 19th-century analysis. The story was often told as a lesson in checking sources. But the criminologist Mike Sutton (2010) traced the decimal-point story itself and found no good evidence for it. A myth about the spread of myths had spread by uncritical repetition. The lesson applies to everyone, including authors of critical-thinking guides.

For testimony in a social context, including trust in experts, the novice–expert problem, echo chambers, misinformation, and AI systems as sources, see Chapter 12.

InferenceListen

Inference extends knowledge from the other sources. Deduction preserves truth: if you know the premises and competently deduce the conclusion, you come to know the conclusion. This is the principle of closure (see The closure argument). Induction and abduction go beyond the premises, generating new beliefs that are probable, not certain.

Can you gain knowledge by inference from a false premise? Ted Warfield (“Knowledge from Falsehood,” 2005) gave an example. Before a talk, you count the audience carefully and conclude there are 53 people. You have 100 handouts, so you conclude that you have enough. In fact, you miscounted and there are 52 people. Your conclusion that you have enough handouts is inferred from a false belief, yet it seems to be knowledge. The false premise was “close enough” to the truth to make no difference. This complicates the no false lemmas response to Gettier.

The rest of this guide deals with inference in detail: deductive logic in Chapter 3, induction and probability in Chapter 9, and explanation and causal inference in Chapter 10.

Knowledge by presenceListen

The Illuminationist philosopher Shihāb al-Dīn Suhrawardī (1154–1191) distinguished two kinds of knowledge:

• Acquired or representational knowledge (ʿilm ḥuṣūlī): knowledge through concepts, images, or representations of an object. Almost all knowledge of the external world is of this kind. There is always a gap between the representation and what it represents, so error is possible.
• Knowledge by presence (ʿilm ḥuḍūrī): knowledge in which the object known is itself present to the knower, with no representation in between.

Suhrawardī’s central example is self-knowledge, and his argument is a regress. Suppose I knew myself through a representation (an image or concept of myself). To recognize that representation as a representation of me, I would already have to know myself. So self-knowledge cannot ultimately be representational. The self must be immediately present to itself.

Because knowledge by presence involves no representation, it involves no gap in which error can occur. When I am in pain, the pain itself is present to me, not a picture of it. Later philosophers in the tradition, especially Mullā Ṣadrā (c. 1571–1640), extended the idea to our knowledge of our own mental states and powers, and to mystical knowledge. The 20th-century Iranian philosopher Mehdi Ha’iri Yazdi (The Principles of Epistemology in Islamic Philosophy: Knowledge by Presence, 1992) presented the theory in the vocabulary of analytic philosophy.

The idea has interesting parallels in Western thought:

• Russell’s knowledge by acquaintance (see Chapter 1): direct awareness of sense-data and, Russell sometimes thought, of the self.
• Immunity to error through misidentification (Sydney Shoemaker, “Self-Reference and Self-Awareness,” 1968, building on Wittgenstein’s Blue Book): when I judge “I have a toothache” on the basis of feeling it, I can be wrong about what I feel, but I cannot be wrong about who has it. I cannot think “someone has a toothache, but is it me?”
• Ibn Sīnā’s floating man (see Chapter 2): self-awareness without any sense perception.

Critics ask whether knowledge by presence is really knowledge (which is propositional and can be expressed in judgments) or rather a precondition of knowledge, a kind of awareness that becomes knowledge only when conceptualized. When I judge “I am in pain,” a concept enters, and with it the possibility of error in classification. The psychological research on introspection (see Introspection and self-knowledge) also suggests that our judgments about our own states are far from infallible, even if the states themselves are present to us.

Intuition, emotion, and other candidate sourcesListen

Expert intuition. Is a “gut feeling” a source of knowledge? Daniel Kahneman and the psychologist Gary Klein, who had taken opposite sides on this question (Kahneman studying biases, Klein studying firefighters and nurses whose intuitions saved lives), wrote a joint paper, “Conditions for Intuitive Expertise: A Failure to Disagree” (2009). They agreed that intuition is trustworthy when two conditions hold:

1. The environment is sufficiently regular to be predictable (chess, firefighting, anesthesiology, not long-term stock picking or political forecasting).
2. The person has had prolonged practice with rapid, clear feedback on their judgments.

Where these conditions are absent, confident intuitions are unreliable, and the confidence itself is no guide. This is one of the most practically useful results in the psychology of judgment.

Emotions. Some philosophers (Ronald de Sousa, The Rationality of Emotion, 1987; Christine Tappolet, Emotions, Values, and Agency, 2016) argue that emotions are a kind of perception of value: fear presents something as dangerous, indignation presents something as unjust. On this view, emotions can be sources of evaluative knowledge, and they can be accurate or inaccurate (fitting or unfitting). They can also mislead: fear of flying is out of proportion to the danger. See Appeal to emotion.

Religious experience. William Alston (Perceiving God, 1991) argued that religious experience can be a form of perception of God, and that “Christian mystical practice” is a socially established doxastic practice with the same kind of prima facie standing as sense perception. See Faith, reason, and religious epistemology.

Other pramāṇas. The Nyāya school’s comparison (upamāna) is learning the meaning of a word by being told that the thing it names resembles something familiar, and then recognizing it. The Mīmāṃsā school’s postulation (arthāpatti) is close to inference to the best explanation: “Devadatta is fat but does not eat during the day, so he must eat at night.” And non-apprehension (anupalabdhi) is knowing an absence by failing to perceive something that would be perceived if it were there (“There is no elephant in this room”). This last one is worth remembering. Absence of evidence can be evidence of absence, but only when the evidence would be expected if the thing were present. See Argument from ignorance.

Check your understandingListen

1Explain the argument from illusion and Austin’s objection to it.

Answer

The argument: a straight stick in water looks bent; what we see is bent; the stick is not; so what we see is not the stick but a sense-datum. Austin: we see a straight stick that looks bent; nothing bent is seen. The argument slides from “looks F” to “is an F-looking thing.” He also objected to treating illusions (misperceptions of real objects) as if they were hallucinations.

2What is the “veil of perception” problem?

Answer

If we perceive the world only indirectly, via mental representations, we can never compare the representations with the world itself to check that they match. So it is unclear how perception could give knowledge of the external world. The problem motivates idealism (Berkeley), skepticism, and direct realism (Reid).

3A witness says, “I’m 100% certain; I’ll never forget that face.” Six months later, she identifies the defendant with total confidence. What should a juror consider?

Answer

Whether her initial identification was made promptly, under fair procedures, and with high confidence (which is diagnostic); whether her confidence grew over time through feedback, repetition, or exposure to the defendant’s face (which inflates confidence without improving accuracy); the conditions of the original observation (lighting, distance, stress, duration, weapon focus); and whether the identification is cross-racial, which is known to be less accurate. Vividness and certainty today are poor indicators on their own.

4What did Nisbett and Wilson’s stocking study show?

Answer

People’s choices were influenced by position (they preferred the item on the right), but they didn’t mention position and denied it when asked, citing product quality instead. When people explain their own behavior, they often rely on plausible theories rather than direct introspective access to the actual causes.

5What is the difference between reductionism and anti-reductionism about testimony? Which better describes how a young child learns?

Answer

Reductionism: testimonial justification requires independent (non-testimonial) positive reasons to trust the speaker. Anti-reductionism: testimony is justified by default, absent defeaters. Young children learn by trusting before they could have independent evidence of reliability, which supports anti-reductionism about them, though psychological research also shows that even young children track informants’ past accuracy and prefer reliable ones (so they are not totally undiscriminating).

6Why is the creationist teacher case important?

Answer

It suggests that testimony can generate knowledge rather than merely transmit it. The students come to know that humans evolved, although their teacher does not believe it. What matters is that the testimony is a reliable conduit of information, not that the speaker knows.

7When is a “gut feeling” trustworthy, according to Kahneman and Klein?

Answer

When the environment is regular enough to be predictable and the person has had extensive practice with rapid, clear feedback. An experienced nurse’s sense that a patient is deteriorating may be reliable; a pundit’s gut feeling about next year’s election or the stock market is not.

8What does non-Euclidean geometry teach about claims of self-evidence?

Answer

Kant held that Euclidean geometry was necessarily and knowably true of space. Non-Euclidean geometries proved consistent, and general relativity found physical space-time to be non-Euclidean. Something that seemed self-evident and inconceivable to deny turned out false (as a claim about physical space). “It’s self-evident” should prompt examination, not end discussion.

Further reading

General

• Robert Audi, Epistemology: A Contemporary Introduction (Routledge, 3rd ed. 2011), Part I on the sources.
• Jennifer Nagel, Knowledge: A Very Short Introduction (2014), chs. 6–7.

Perception

• Bertrand Russell, The Problems of Philosophy (1912), chs. 1–4.
• J. L. Austin, Sense and Sensibilia (Oxford University Press, 1962).
• Tim Crane and Craig French, “The Problem of Perception,” Stanford Encyclopedia of Philosophy.
• Christopher Chabris and Daniel Simons, The Invisible Gorilla (Crown, 2010).

Memory

• Elizabeth Loftus and Katherine Ketcham, The Myth of Repressed Memory (St. Martin’s, 1994).
• Daniel Schacter, The Seven Sins of Memory (Houghton Mifflin, 2001; updated ed. 2021).
• Julia Shaw, The Memory Illusion (Random House, 2016).

Introspection

• Timothy Wilson, Strangers to Ourselves (Harvard University Press, 2002).
• Eric Schwitzgebel, Perplexities of Consciousness (MIT Press, 2011).

Reason and the a priori

• Laurence BonJour, In Defense of Pure Reason (Cambridge University Press, 1998).

Testimony

• C. A. J. Coady, Testimony: A Philosophical Study (Oxford University Press, 1992).
• Jennifer Lackey, Learning from Words (Oxford University Press, 2008).
• Jennifer Lackey and Ernest Sosa (eds.), The Epistemology of Testimony (Oxford University Press, 2006).

Knowledge by presence

• Mehdi Ha’iri Yazdi, The Principles of Epistemology in Islamic Philosophy: Knowledge by Presence (SUNY Press, 1992).
• Suhrawardī, The Philosophy of Illumination, translated by John Walbridge and Hossein Ziai (Brigham Young University Press, 1999).