1 What does the verb ‘know’ mean?

Philosophers have spent two and a half millennia failing to define “know” satisfactorily.

1.1 The Tripartite Definition

Plato’s dialogue Theaetetus (c. 369 BCE) is the source of what has become the standard starting point — though the dialogue itself examines the definition of knowledge as true belief with an account (logos) and ends by rejecting it. The tradition kept the proposal Socrates found wanting, polished into justified true belief: to know something, you must (1) believe it, (2) it must be true, and (3) you must have adequate justification for your belief. This is often abbreviated JTB.⁴ (Plato’s Greek distinguishes episteme (knowledge) from doxa (belief or opinion) and aletheia (truth) — the trio JTB is in the business of relating.)

Three conditions, three separate questions:

• Belief: Can you know something you don’t believe? Could you know the answer to a question and still write the wrong answer on a test? (Some philosophers say yes — you might “know” in some implicit sense while consciously believing the opposite.)
• Truth: Can you know something that turns out to be false? If a geography teacher “knows” that the capital of Australia is Sydney, and they are wrong, were they ever knowing, or only believing?
• Justification: What makes a belief justified? Perception? Memory? Testimony? Inference? This is where much of the real work in epistemology happens.

The tripartite definition has the advantage of capturing our intuitions: hunches are not knowledge (no justification), lucky true guesses are not knowledge (no justification), false beliefs are not knowledge (not true).

Plato’s own reason for rejecting the definition lands on the third clause — the logos, the account a knower is supposed to be able to give. Socrates tries three readings of it and none survives. If an account is simply putting your belief into words, then anyone holding a true belief has one already, and the clause adds nothing to belief. If it means listing the elements a thing is made of, the schoolboy sinks it: he writes the first syllable of “Theaetetus” correctly and the first syllable of “Theodorus” wrongly, though he can recite the letters of both — the same account twice, and no knowledge the second time. And if it means naming the mark that distinguishes this thing from every other, the definition eats its own tail: you could not have held a true belief about Theaetetus at all without already grasping what sets him apart, so the clause either repeats the work belief has done or defines knowledge by helping itself to knowledge. The dialogue ends with nothing caught — and the condition Plato could not make work is the one Gettier would break.

1.2 The Gettier Problem

In 1963, Edmund Gettier published a three-page paper — “Is Justified True Belief Knowledge?” — that changed the field overnight.⁵ He gave two counterexamples showing that you could have a justified true belief that is, intuitively, not knowledge.

The classic reconstruction: Smith and Jones are applying for a job. Smith has strong evidence (the boss told him directly) that Jones will get the job. Smith has also counted the coins in Jones’s pocket: ten coins. Smith therefore believes: The man who will get the job has ten coins in his pocket. This belief is justified. But — unbeknownst to Smith — Smith himself gets the job. And, by coincidence, Smith also has ten coins in his pocket. So Smith’s belief is true. And it was justified.

Philosophers since 1963 have tried to add a fourth condition, change what we mean by justification, or abandon JTB. None of the proposed solutions has won consensus. The problem remains open.

What it tells us: Justification can be accidentally disconnected from truth.

flowchart TB
  subgraph BEL["Everything you believe"]
    FAL["False beliefs"]
    subgraph TRU["True beliefs"]
      LUCK["Lucky true belief<br/>true, but unjustified"]
      subgraph JTB["Justified true belief"]
        GET["Gettier cases<br/>justified and true —<br/>but true by accident"]
        KNO["Knowledge"]
      end
    end
  end

Return now to Théo. His belief — Trump will win — was true, and it was held for reasons: a specific theory about polling error and a privately commissioned survey built to test it. On the face of it that looks like justified true belief, the very thing JTB says knowledge is. The trouble is the route from his justification to the truth. A single presidential election is one trial; a method that has never been validated against a run of cases can deliver the right answer without that answer being produced by the method’s grip on reality. Théo’s neighbour-poll might have detected a real, hidden Trump majority — or it might have been one plausible story among several, vindicated by an outcome it did not actually track. If the second description fits, his belief has the Gettier shape exactly: justified, true, and true for reasons disconnected from the justification — right, but lucky.

This is what makes the case sharper than an ordinary wrong prediction. Nobody learns much from a forecaster who was confidently mistaken; we simply say the evidence was misread. The whale was confidently right — and richly paid for it — which is precisely the condition under which we are most tempted to credit someone with knowledge they may never have had. The question below turns on whether being vindicated is the same as having known.

1.3 Beyond JTB

Gettier attacks the definition from the inside: he grants all three conditions and shows they do not add up to knowledge. The moves below do something different. They ask whether the definition has the right shape — whether every belief in the structure can be a justified one, and whether knowing that something is the case is the only kind of knowing there is.

Knowledge, belief, opinion. The Greek Plato uses is instructive: episteme (knowledge), doxa (belief or opinion), aletheia (truth). Opinion can be true or false; knowledge, on JTB, must be true. Compare three sentences:

• “I believe it will rain tomorrow.”
• “I know that water is \(\mathrm{H_2O}\).”
• “In my opinion, the painting is beautiful.”

They feel categorically different — but do they differ in what is claimed, or only in how firmly it is held? Wittgenstein presses the question from the other end. In On Certainty (1969) he points to beliefs so basic that doubting them would dismantle the frame in which doubt makes sense: that the earth existed before you were born, that this is a hand. He calls them hinges — not things we know on evidence, but what our weighing of evidence turns on.

At the foundation of well-founded belief lies belief that is not founded.

— Ludwig Wittgenstein, On Certainty §253⁶

If that is right, JTB cannot be the whole story: a structure of justified beliefs has to rest on something that is not itself justified. Lesson 2 makes exactly this the sceptic’s problem.

Knowledge by acquaintance and knowledge by description. Bertrand Russell drew a distinction in The Problems of Philosophy (1912) that cuts across the JTB debate:⁷

• Knowledge by acquaintance — direct, unmediated contact with something: you know the colour red by seeing it, pain by feeling it.
• Knowledge by description — knowing about something through propositions: you know that Julius Caesar was stabbed on the Ides of March without having been there.

Russell’s point is that almost everything we know of the world is knowledge by description. We are rarely acquainted with the facts themselves, which puts the weight on the chain that carries them to us — testimony, books, memory. Lesson 5 takes up that chain.

Russell’s reversal. The distinction is worth knowing as Russell drew it, and worth knowing that Russell gave it up. Under Wittgenstein’s criticism in 1913 he abandoned the Theory of Knowledge manuscript mid-book, and by “On Propositions” (1919) acquaintance had gone as a primitive epistemic relation.⁸ A course that teaches acquaintance as Russell’s settled view is teaching the 1912 Russell, not the philosopher Russell became.

McDowell’s disjunctivism. The idea that perception puts us in direct contact with the world survives its Russellian form. John McDowell argues that seeing and hallucinating share no “highest common factor”: experience is either a case of the fact making itself manifest or a mere appearance, and in the good case your warrant is the world itself rather than an inner seeming.⁹ Keep it in view for the next lesson — the sceptic’s “it would seem exactly the same from the inside” already assumes the common factor McDowell denies. Lesson 4 returns to the argument in full.

Knowing how and knowing that. Gilbert Ryle separated knowing how — riding a bike, reading a room — from knowing that: Paris is the capital of France.¹⁰ The first is procedural, the second propositional, and JTB was built for the second: belief, truth and justification are all things a proposition can have. The cyclist holds no candidate proposition to justify, yet no one denies that they know how to ride. Either knowing-how reduces to a stock of knowing-that — Ryle’s target, and he thought the reduction fails — or JTB defines one kind of knowing among several. Lesson 6 returns to this with Merleau-Ponty and the body’s own knowledge.

Gettier says the definition is incomplete. These four say something sharper: it may be a definition of propositional knowledge, resting on hinges it cannot justify, in a world where much of what we know we never had to justify at all.

1.4 Questions

Questions 2 and 4 are retrieval checks — answer them from memory before you argue the rest.

1. If you cannot know you are not dreaming right now, can you know anything about the external world?
2. State Gilbert Ryle’s distinction between knowing how (to ride a bike) and knowing that (Paris is the capital of France) — which of the two is procedural and which propositional knowledge?¹¹ Then say which of them JTB was built to define, and what that costs it with the other.
3. Can a computer know something? What would it lack, if anything?
4. State the structure of a Gettier case: which of JTB’s three conditions are satisfied, and where does luck enter? Then say what the whale’s neighbour-poll shares with that structure — and whether that makes the Gettier problem devastating, or a quirky edge case.

2 Am I a brain in a vat?

In 1641, René Descartes sat by a fire and decided to doubt everything he could doubt. His method of systematic doubt — scepticism wielded as a tool, not held as a conclusion — produced one of the most influential arguments in the history of philosophy.

2.1 The Method of Doubt

Descartes begins Meditations on First Philosophy by noticing that his senses have deceived him before. Things look different sizes depending on their distance. Straight sticks appear bent in water. If the senses can deceive sometimes, can he trust them at all?

He escalates. Perhaps he is dreaming right now. Everything he experiences might be a dream. You’ve had dreams that seemed real. How do you know this isn’t one?

He escalates further. What if an all-powerful evil demon (malin génie) is deceiving him about everything — not just his senses, but even mathematics? Could 2+2=5 be a deception planted by this demon?

“I will suppose therefore that not God, who is supremely good and the source of truth, but rather some malicious demon of the utmost power and cunning has employed all his energies in order to deceive me.” — René Descartes, Meditations on First Philosophy, Meditation I¹⁴

This is not paranoia. It is a philosophical method: if a belief could possibly be false under any scenario, however far-fetched, it cannot serve as a foundation for knowledge.

2.2 The Brain in a Vat

The contemporary version of Descartes’ evil demon is the brain-in-a-vat scenario, articulated by Hilary Putnam in Reason, Truth and History (1981).¹⁵ Imagine that you have always been a disembodied brain in a laboratory, kept alive in a vat of nutrients, with all your sensory experiences generated by a supercomputer. Everything you have ever seen, touched, felt, and thought has been a simulation. How would you know?

The scenario is logically coherent. There is no internal contradiction in the idea. If it is coherent, then for all you know, it might be true. And if you can’t rule it out, how can you claim to know anything about the external world?

This is external world scepticism: the view that we cannot have knowledge of the world beyond our own minds. Descartes himself didn’t accept this as the final word. He thought he could escape it. The escape route he took — proving God’s existence to guarantee the reliability of his faculties — is not widely accepted today, though Alvin Plantinga’s reformed epistemology defends a version of the move, arguing that without a divine guarantee the reliability of our cognitive faculties is itself unsupported.¹⁶ Whether one accepts that reply or not, the Cartesian move has serious twentieth-century defenders.

2.3 The Cogito

At the bottom of the doubt, Descartes finds something he cannot doubt:

“I think, therefore I am.” (Je pense, donc je suis) — René Descartes, Discourse on the Method, Part IV¹⁷

The Latin cogito, ergo sum is from Descartes’ Principles of Philosophy (1644, Part I, §7), not the Meditations. In the Meditations on First Philosophy (the later, more rigorous work), the formulation is different: “I am, I exist” — ego sum, ego existo — which Descartes says “is necessarily true whenever it is put forward by me or conceived in my mind” (Meditation II).

The act of doubting itself proves there is a doubter. Even if the demon is deceiving me, there must be a me being deceived. The Arup employee sat in a meeting in which every other face was a fabrication, and was, on Descartes’ picture, the cogito’s residue made vivid: a deepfake can counterfeit every other participant on the call, but not the one it is aimed at deceiving — whatever else was in doubt, the deception was happening to someone.

Descartes treats this as the unshakeable foundation: I exist as a thinking thing.

But the cogito raises its own problems:

• Does “I think” already smuggle in the assumption of an “I”? Lichtenberg argued that all Descartes can know is that there is thinking, not that I am doing it.¹⁸
• If the cogito is the only certainty, it is thin. What can you build on “I exist as a thinking thing”?
• Hume later argued that when he introspected to find the “I”, he never caught anything — only a bundle of perceptions. There is no unified self.¹⁹

2.4 Responses to Radical Scepticism

Philosophers have not simply accepted that we know nothing — this would be a strange response from people whose entire professional existence depends on the possibility of knowing things. Several responses follow. The textbook habit is to list them as a buffet; this obscures that they are doing very different things. Some try to refute the sceptic on the sceptic’s own terms (Moore, Putnam’s externalist argument). Some try to change the subject — to show, with Wittgenstein, that the sceptic’s question is not the question it appeared to be. Some accept and deflate the sceptic’s conclusion (the pragmatists). These are not interchangeable moves, and the sceptic’s grip depends on theoretical commitments that can themselves be put in question.²⁰

Moore’s common sense (G.E. Moore, “Proof of an External World”, 1939): Moore held up his hands and declared, “Here is one hand, and here is another”; external things, he concluded, therefore exist.²¹ This seems ludicrous as a philosophical argument, but Moore’s point was that the certainty of the external world is more certain than any philosophical argument against it.

Externalism (Hilary Putnam): Putnam — who introduced the brain-in-a-vat scenario — also argued that if you were a brain in a vat, your word “vat” wouldn’t even refer to real vats (since all your experience of “vat” comes from the simulation). This is a subtle semantic argument that the scenario may be self-undermining.

Wittgenstein on hinges: In On Certainty, Wittgenstein argues that some beliefs function as hinge propositions — they are not things we test against evidence, but the framework within which testing happens at all. The proposition that the earth has existed for many years past is not a hypothesis I entertain; it is a hinge. Doubting it isn’t possible inside a normal form of life.²² As Wittgenstein puts it: “At the foundation of well-founded belief lies belief that is not founded” (On Certainty, §253).²³ The Arup employee proceeded on a hinge — that the colleagues he could see and hear were his colleagues — and the fraud succeeded not by defeating that certainty in argument but by exploiting it: you cannot conduct a video meeting while treating each familiar face as a hypothesis awaiting proof, and the deepfake turned exactly that un-testability into a weapon.

Zhuangzi’s butterfly dream (Zhuangzi, Book II, “Discussion on Making All Things Equal,” ~4th c. BCE): “Once Zhuang Zhou dreamt he was a butterfly, a butterfly flitting and fluttering around, happy with himself and doing as he pleased. He didn’t know he was Zhuang Zhou. Suddenly he woke up and there he was, solid and unmistakable Zhuang Zhou. But he didn’t know if he was Zhuang Zhou who had dreamt he was a butterfly, or a butterfly dreaming he was Zhuang Zhou. Between Zhuang Zhou and a butterfly there must be some distinction! This is called the Transformation of Things.”²⁴ Zhuangzi raises the same scenario Descartes will raise two thousand years later but draws a different conclusion: not “doubt everything until something certain remains,” but “the demand for a vantage-point above the dreaming and the waking is itself the mistake.” On the Zhuangzian reading the sceptic and Moore are both committed to a perspective from which the question can be settled — and the genuine philosophical move is to notice that no such perspective is available.

Pragmatism: William James and John Dewey argued that the question “am I really perceiving the external world?” is meaningless if it makes no practical difference. The sceptical hypothesis, by design, makes no predictions different from ordinary realism.²⁵

flowchart LR
  DOU["The sceptical doubt:<br/>nothing inside experience<br/>rules out the demon,<br/>the dream, the vat"]
  DOU --> R["Refute<br/>the doubt is false"]
  DOU --> S["Dissolve<br/>the doubt has no<br/>coherent target"]
  DOU --> F["Deflate<br/>the doubt is idle"]
  R --> R2["Moore's two hands<br/>Putnam's semantic argument"]
  S --> S2["Wittgenstein's hinges<br/>Zhuangzi's butterfly"]
  F --> F2["James and Dewey:<br/>it predicts nothing<br/>ordinary realism does not"]

2.5 What is Left to Know?

Descartes’ inquiry doesn’t leave us with nothing. It leaves us with a precise map of what we can and cannot justify:

• The existence of our own mental states (certainty)
• Mathematical and logical truths (highly certain, but vulnerable to demon scenarios)
• The existence and nature of the external world (not directly justified from the inside)

The limits of Cartesian doubt mark the boundary between knowledge that survives the demon scenario (one’s own mental states, at minimum; perhaps mathematics; perhaps the cogito) and knowledge that does not (the external world, other minds, the past as remembered).

2.6 Questions

Questions 1 and 3 are retrieval checks — answer them from memory before you argue the rest.

1. Could you ever prove — not just assume — that you are not a brain in a vat? Two things to state from the lesson before you answer: why no test run from inside the stream of experience can settle it, and what Putnam’s semantic argument claims about the sentence “I am a brain in a vat.”
2. Is Moore right that common sense certainties should override philosophical arguments? Or does that just refuse to engage with the question?
3. Descartes thought the cogito was the one thing immune to doubt. State the two challenges the lesson put to it: Lichtenberg’s objection about the “I”, and what Hume reports finding when he introspects. What assumptions does the cogito turn out to rely on?
4. If we cannot know whether we’re in a simulation, should that change how we live? Or is the question practically meaningless?

3 What is truth?

“What is truth?” — Pontius Pilate asked the question and didn’t wait for the answer. He may have been wiser than he looked. Everyone wants to know the truth. Politicians say they’re speaking it. Scientists claim to discover it. Courts try to establish it. The major philosophical accounts of what truth is, however, have continued to disagree — and each captures something the others have to concede.

3.1 The Correspondence Theory

The oldest and most intuitive theory: a statement is true if and only if it corresponds to reality. To say “it is raining” is true just in case it is, in fact, raining.

Aristotle stated this with admirable clarity:

“To say of what is that it is not, or of what is not that it is, is false, while to say of what is that it is, or of what is not that it is not, is true.” — Aristotle, Metaphysics, Book IV, Chapter 7²⁹

Truth is about the world fitting the words.

The problems:

• What are the “truth-bearers”? Sentences? Propositions? Thoughts? In different languages, different sentences express the same truth — so the truth-maker must be something language-independent.
• What is the “correspondence relation” exactly? It can’t be resemblance (the word “cat” doesn’t look like a cat). Correspondence turns out to be difficult to cash out precisely.
• How does correspondence work for abstract truths? “7 is prime” corresponds to what, exactly, in physical reality?

3.2 The Coherence Theory

Rather than asking whether a belief corresponds to external reality, the coherence theory says: a belief is true if it coheres with a system of beliefs. Truth is internal consistency within a web.

This sounds strange, but it has a serious motivation: we never have direct access to reality independent of our beliefs. All our evidence for the world comes through our perceptual, conceptual, and theoretical frameworks. There is no “view from nowhere.” So the only available criterion of truth is whether beliefs hang together.

F.H. Bradley and later Bernard Bosanquet (British Idealists, late 19th century) developed this view.³⁰ It is also reflected in Neurath’s “ship” metaphor: we are always rebuilding the ship of knowledge while at sea. We cannot put every plank on dry land at once.³¹

The problem: Two incompatible and internally coherent belief systems could both be “true” by this criterion.

3.3 The Pragmatic Theory

William James (1907) proposed a radical alternative: a belief is true if it works — if believing it is useful, if it “pays” in terms of practical experience.

“‘The true,’ to put it very briefly, is only the expedient in the way of our thinking, just as ‘the right’ is only the expedient in the way of our behaving.” — William James, Pragmatism, Lecture VI³²

This has a democratic, anti-elitist feel: truth is not some mystical correspondence with a Platonic realm, but a practical instrument for navigating experience.

The problem: It seems possible for a false belief to be useful. Believing you are invincible might help you win a fight (briefly). And some truths seem useless or actively harmful in the short term.

The earlier Cardiff Giant hoax of 1869 — a planted gypsum statue mistaken for a petrified biblical giant³³ — is the textbook precursor: a false belief that paid (admission fees, religious controversy, newspaper coverage) until it didn’t, with the same logical shape Nikola would later display on a stock exchange. Pragmatism’s defenders reply that the working-out was the falsification, and that “pays by the way” never meant “is currently profitable”; pragmatism’s critics reply that any account that needs that much qualification has handed the work back to correspondence.

flowchart LR
  T["A belief is true<br/>when it…"]
  T --> C["…matches how<br/>things are<br/><b>Correspondence</b><br/>Aristotle, Tarski"]
  T --> H["…hangs together with<br/>the rest of what you hold<br/><b>Coherence</b><br/>Bradley, Neurath"]
  T --> P["…pays its way<br/>in experience<br/><b>Pragmatic</b><br/>James, Dewey"]
  C --> C2["You cannot step outside<br/>belief to inspect<br/>the match"]
  H --> H2["A consistent fiction<br/>coheres perfectly"]
  P --> P2["Useful falsehoods work —<br/>the Nikola One<br/>paid for four years"]

3.4 Questions

Questions 2 and 4 are retrieval checks — answer them from memory before you argue the rest.

1. Can there be more than one “true” account of an event? (Multiple witnesses, different but equally accurate perspectives?) Is this coherentism, or just multiple truths?
2. Name the three theories of truth this lesson set out, and give the one-line test each applies to a claim. Which tests did “the Nikola One drives” pass for four years, and which did it fail the whole time — and is James’s “pays by the way” then just usefulness misnamed?
3. Does it matter, for practical purposes, which theory of truth is correct? Or is this a purely academic dispute?
4. State the Liar’s structure: what follows if “This sentence is false” is true, and what follows if it is false? Then state what Tarski’s separation of object-language from meta-language does to block it. Does the block resolve the paradox or quarantine it — and does that mean logic has limits? (Preview of the Logic unit on Gödel.)

4 Can we trust our senses?

Look at a stick standing in a glass of water. It bends at the surface. Philosophers and psychologists have spent two and a half millennia asking what to make of that. Perception clearly can go wrong; what is at stake here is what its going wrong implies about the kind of access perception gives to the world at all.

4.1 The Müller-Lyer Illusion

Look at two lines of equal length, one ending in inward-pointing arrowheads (<–>) and one in outward-pointing arrowheads (>–<). The second looks longer. Measure them: they are identical. Now look again. It still looks longer. This is the Müller-Lyer illusion, described by Franz Carl Müller-Lyer in 1889.³⁹

Perceptual systems do not record passively; they interpret, using context, angles, and prior assumptions to build our experience. The brain fills in and edits constantly.

4.2 Russell’s Table

These cases — disagreement that survives knowing the truth — make a metaphysical question pressing. Bertrand Russell opens The Problems of Philosophy (1912) by asking you to look at a table. You might say the table is brown. But the colour you see depends on the lighting. The shape looks different from different angles. The texture changes depending on your distance from it.

“It becomes evident that the real table, if there is one, is not the same as what we immediately experience by sight or touch or hearing.” — Bertrand Russell, The Problems of Philosophy, Chapter 1⁴⁰

Russell distinguishes between sense-data (what we directly experience — patches of colour, sounds, sensations) and physical objects (the supposed real things causing those experiences); on his view we only ever perceive sense-data.

Russell’s distinction creates the problem of the external world: if all I ever directly experience is my own sense-data, how do I know there is a world of physical objects behind it? How do I know the table persists when no one is looking?

George Berkeley (1685–1753) drew the radical conclusion: esse est percipi — to be is to be perceived. There are no physical objects independent of perception; there is only God and minds and their ideas.⁴³ Samuel Johnson, walking with Boswell in 1763, kicked a large stone and said “I refute it thus.” Berkeley’s defenders reply that the kick produces a sensation of pain — which Berkeley never denied — but that nothing in the kick proves there is anything behind the sensation. Berkeley’s critics reply that mind-independent objects are the simplest hypothesis explaining the regularity, intersubjectivity, and predictability of perception (why do unwatched apples rot the way they do? why do other observers report the same readings I do, even when their interests would have inclined them to report otherwise?). Berkeley’s own reply to that — that God’s perception sustains the world in our absence — is the move on which the whole position stands or falls.

4.3 Locke’s Primary and Secondary Qualities

John Locke (An Essay Concerning Human Understanding, 1689) offered a more moderate response. He distinguished:⁴⁴

• Primary qualities: features of objects that are genuinely in the objects themselves — size, shape, number, motion, solidity. These ideas in our minds resemble how things really are.
• Secondary qualities: features that exist in the object only as powers to produce sensations in us — colour, sound, taste, smell, temperature. The tomato’s redness is not in it the way its shape is. Redness is what the tomato’s surface reflectance does to our visual system.

Modern science reflects this split: physics describes the world in terms of mass, charge, and position rather than in terms of colour, taste, or beauty.

flowchart LR
  OBJ["The object"]
  OBJ -- "genuinely in it" --> P["<b>Primary qualities</b><br/>size · shape · number<br/>motion · solidity<br/><br/>your idea resembles<br/>how things are"]
  OBJ -- "only a power to<br/>produce sensations" --> S["<b>Secondary qualities</b><br/>colour · sound · taste<br/>smell · temperature<br/><br/>the redness is in you,<br/>not in the tomato"]

Webb’s images are a clean Lockean case: colour is treated openly as a secondary quality — a code an instrument assigns to primary differences (wavelength, energy) that no eye can register. Locke’s distinction predicts that the colour is “in us”; what it does not so comfortably predict is that the encoding the telescope performs deliberately is one the eye performs unwittingly, for every colour anyone has ever seen.

The problem with Locke: Berkeley immediately objected that we have no reason to trust primary qualities either. If colour is in the mind, what makes shape any different? The “resemblance” between our idea of shape and the shape itself cannot be verified, since the only access we have to “the shape itself” is through more ideas of shape.

4.4 The Reliability of Perception

Contemporary philosophy of perception divides into several camps. Each picture handles Webb’s colours differently and pays a different price.

• Naive realism (direct realism): we directly perceive physical objects, not intermediaries. The stick in water is straight; we just misjudge it.
• Representationalism: we perceive internal mental representations of the world. The question is whether those representations are reliable.
• Disjunctivism: genuine perception (of a real object) and hallucination are fundamentally different kinds of mental states, even if they feel the same from the inside.

flowchart LR
  Q["What stands between<br/>you and the object?"]
  Q --> N["<b>Naive realism</b><br/>Nothing. You perceive the<br/>object itself; the stick is<br/>straight and you misjudge it"]
  Q --> R["<b>Representationalism</b><br/>A mental representation.<br/>The question becomes<br/>whether it is reliable"]
  Q --> D["<b>Disjunctivism</b><br/>Wrong question. Seeing and<br/>hallucinating are different<br/>kinds of state — no<br/>highest common factor"]

The trichotomy is a useful first map, but it can mislead. Disjunctivism is not a “third way” between naive realism and representationalism in the manner of a compromise; it is a structural disagreement about what kind of state perceiving even is.

In “Criteria, Defeasibility, and Knowledge” (1982), McDowell argues that the sceptic’s argument trades on a hidden assumption — that perception and hallucination share a “highest common factor” — and once that assumption goes, the argument loses its grip. Which view is right turns in part on how the sceptical problem has been set up in the first place.⁴⁵

4.5 Does Attention Constitute Its Object?

The lesson has so far treated the scene before your eyes as given, and asked what kind of contact you have with it. One pressure remains: whether the scene is given at all. In Simons and Chabris’s celebrated experiment (1999), viewers asked to count basketball passes were shown a video in which a person in a gorilla suit walks into the middle of the game, thumps its chest, and walks off; roughly half of them never saw it.⁴⁶ The phenomenon is called inattentional blindness, and the name carries a whole theory inside it: the scene was there, complete, gorilla included, and attention is a narrow gate that let only some of it through to you. Call this the filter picture. It is the picture built into the ordinary metaphors — attention “caught,” “paid,” “grabbed” — and into the psychology of limited bandwidth: the world arrives ready-made, and attending selects which parts of it you get.

A line of thinkers — an art critic, a phenomenologist, a philosopher of perception — holds that the filter picture has the process backwards in one precise respect: attention does not select from a scene that is already assembled; it does part of the assembling.

John Berger opens Ways of Seeing (1972) — a book about paintings, not laboratories — by denying that seeing is mechanical registration: “We only see what we look at. To look is an act of choice.” What the choice yields is not a selection but a situation: “We never look at just one thing; we are always looking at the relation between things and ourselves.” And the paragraph ends on the strongest word available — vision is “continually holding things in a circle around itself, constituting what is present to us as we are.”⁴⁷ Constituting, not filtering: on Berger’s account, what is present to you — the scene itself, not merely your report of it — is partly your looking’s work.

Maurice Merleau-Ponty had made the same denial philosophically exact a generation earlier. In Phenomenology of Perception (1945) he weighs the searchlight analogy — the filter picture’s own favourite metaphor — and rejects it in a clause: it “presupposes given objects on to which the beam plays,” whereas the function that directs us at the world, “before bringing objects to our sight or knowledge, makes them exist in a more intimate sense, for us.” His name for that function is the intentional arc: the life of consciousness “is subtended by an ‘intentional arc’ which projects round about us our past, our future, our human setting” — so that what shows up for you is already shaped by what you have done, what you can do, and what you care about.⁴⁸ (His companion claim — that it is the knowing body that carries most of this — is taken up in the intuition lesson.)

Bence Nanay restates the claim in the vocabulary of current philosophy of perception by distinguishing ways of exercising attention. Looking for a cab, your attention is fixated: one feature — yellow — scanned across many objects. Defusing a bomb, it is fixated on many features of one object. In aesthetic experience it is open-ended: distributed across many features of one object “but not with a specific aim or goal in mind.” The distinction is doing constitutive work — “Attending to a relevant feature can completely transform your experience.” His example is Bruegel’s Fall of Icarus: a quaint ploughing scene until your attention finds the drowning legs in the corner of the canvas, and never quite that painting again.⁴⁹ Two viewers before one picture, one scanning for the subject, one dwelling on the whole, are not — on Nanay’s account — receiving different quantities of a single given experience. They are having different experiences: the exercise of attention is part of what the experience is.

The dispute changes what expertise in looking amounts to. A follow-up to the gorilla experiment asked radiologists to hunt for lung nodules in chest CT scans into which a gorilla — many times a nodule’s size — had been matted; a large majority of the experts never reported it, and eye-tracking showed that most of those who missed it looked straight at it.⁵⁰ On the filter picture this is an embarrassment: the best-trained filters in the hospital passed over the largest object in the scan. On the constitutive picture it is what training means: years of practice have built a way of looking in which lung-things show up and gorillas do not — the very way of looking that finds a nodule no novice could see. Expertise in looking is then Ryle’s knowing how⁵¹ operating inside perception itself: skill deciding, before any belief is formed, what there is to have beliefs about. Russell’s knowledge by acquaintance (margin note above) was supposed to be the innocent, unearned floor of knowing; if the constitutive claim is right, acquaintance can be trained — and the floor is built. Put plainly: on this picture, even “just looking” is a skill, and what you see is partly what you have learned to see.

One case sits awkwardly for both sides. In February 2015 a snapshot of a striped dress divided the internet: most viewers saw it as blue and black — which the dress in fact was — while a large minority saw white and gold, and the split ran through households, classrooms, and vision labs alike. The favoured explanation is that the photograph strands the visual system’s normally invisible assumption about the ambient light: viewers whose systems discount bluish daylight see white and gold; those who discount warm artificial light see blue and black.⁵² The filter picture struggles here, because nothing was filtered — both groups attended to the same pixels with full concentration and were given different scenes. But the constitutive side inherits its own embarrassment, because nothing was chosen either. No act of looking, no effort of will, moves a viewer from one dress to the other; as with the Müller-Lyer lines, the construction happens upstream of choice. If attention helps make the scene, the Dress shows the making running deeper than attention — Berger’s “act of choice” is at most the top layer of an assembly the chooser never sees.

What the disagreement makes visible is that the lesson’s question has been double all along. On the filter picture, “can we trust our senses?” asks whether enough of a ready-made world gets through the gate. On the constitutive picture, a prior question opens underneath it: how much of what arrives was assembled in the arriving — by attention, by training, by the assumptions the Dress exposed. The missing gorilla was, on the first picture, a failure of bandwidth; on the second, it is a fact about what your world contained for that minute. Each of the three camps above — naive realist, representationalist, disjunctivist — still owes an answer to whichever picture of attention it adopts, and the questions below press the choice.

4.6 Questions

Questions 1 and 2 are retrieval checks — answer them from memory before you argue the rest.

1. The Müller-Lyer lines still look unequal after you have measured them. State what that persistence shows about the relationship between perception and belief: which of the two updates the other, and which does not?
2. State Locke’s sorting: which qualities are primary, which secondary, and what does he claim only of the primary ones? Then state Berkeley’s attack — why, on his argument, does the resemblance claim fail? Is the distinction stable?
3. Does it matter for daily life whether we are directly perceiving the world or representing it? Or is this a distinction without a practical difference?
4. If different species perceive the world very differently (dogs smell what we can’t; bats echolocate), which perception is “correct”? Is there a fact of the matter about what the world “really” looks like?
5. The gorilla experiment is usually summarised as “we miss what we do not attend to.” Rewrite that moral twice — once assuming attention filters a ready-made scene, once assuming attention helps constitute the scene. Which rewrite better fits the radiologists who looked straight at the gorilla and did not see it — and which better fits the Dress, where no shift of attention changes anything?

5 Why should I believe what others tell me?

Take an inventory. You “know” that Antarctica exists, that Caesar crossed the Rubicon, that Sirius is a star, that your birthday is the date written on your passport, that the medication on the chemist’s shelf is what the label says it is. Almost none of this knowledge is yours by direct acquaintance. You have it because somebody told you — a teacher, a parent, a textbook, a stranger in a uniform, an anonymous Wikipedia editor, a journalist you have never met. Strip out everything you accepted on someone else’s word, and what survives would not get you through breakfast. So is testimony — being told that p — a genuine source of knowledge in its own right? Or is it only a derivative source, one that has to be cashed back into perception and inference before it counts? That dispute runs straight through the philosophy of how anyone gets through their day — and, below, through an institution that took a machine’s word over hundreds of people’s.

5.1 Testimony as a Source

Jennifer Nagel observes that “many of our prized possessions come to us second-hand.”⁵⁵ The technical term is testimony: someone tells you something, in speech or writing, and the content of what they say plays a special role in what you take from the exchange. (You overhear a friend say “I have a hoarse voice” and hear that the voice is hoarse — that is perception. You read in a newspaper that there has been an earthquake in Turkey — that is testimony.) Testimony so understood is the load-bearing source for almost everything we know about history, geography, science we did not personally do, and the lives of everyone we have not personally observed. Robert Audi puts the structural point bluntly: “If perception, memory, consciousness, and reason are our primary individual sources of knowledge and justification, testimony from others is our primary social source of them.”⁵⁶ The philosophical question is whether testimony is another basic source — like perception — or whether it is at best a derivative source, parasitic on the others.

5.2 Reductionism: The Hume Position

The reductionist holds that there is no special epistemic credit attached to being told. When you accept what someone says, you must (in principle) be able to back the acceptance with independent evidence about the speaker’s reliability — past track-record, internal coherence, plausibility against your own observation. Testimony reduces to perception, memory, and inference; it adds nothing of its own. The classical reductionist text is Hume’s Enquiry, §X “Of Miracles,” where he is forced to defend the position in the hardest case — testimony to a violation of natural law — but states the underlying epistemology as a general claim:

“A wise man, therefore, proportions his belief to the evidence. […] our assurance in any argument [from testimony] is derived from no other principle than our observation of the veracity of human testimony, and of the usual conformity of facts to the reports of witnesses.” — David Hume, Enquiry Concerning Human Understanding §X.4–5⁵⁷

Strong consequence (Hume’s main target): “no testimony is sufficient to establish a miracle, unless the testimony be of such a kind, that its falsehood would be more miraculous, than the fact, which it endeavours to establish.”⁵⁸ Locke had taken a similar but more general line a century earlier: testimony, however well-supported, falls short of knowledge because it is at best probable, never certain. On the Lockean account, you have only “highly probable opinion” about most of the things you would ordinarily say you know on testimony.⁵⁹

5.3 Anti-Reductionism: The Reid Position

The anti-reductionist holds that testimony is a basic source — like perception — operating with a default of acceptance, not a default of suspicion. We do not, and could not, derive trust in others’ words from prior independent evidence; we begin with credulity, and reasons are required to reject a piece of testimony, not to accept it. The locus classicus is Thomas Reid’s “principle of credulity”:

“The wise and beneficent Author of nature, who intended that we should be social creatures, and that we should receive the greatest and most important part of our knowledge by the information of others, hath, for these purposes, implanted in our natures two principles that tally with each other. […] Another original principle implanted in us by the Supreme Being, is a disposition to confide in the veracity of others, and to believe what they tell us. […] we shall, for want of a more proper name, call this the principle of credulity. It is unlimited in children, until they meet with instances of deceit and falsehood: and it retains a very considerable degree of strength through life.” — Thomas Reid, An Inquiry into the Human Mind, Chapter VI, Section 24⁶⁰

The most influential modern argument for the Reidian position is C. A. J. Coady’s: pure reductionism is self-undermining. To assemble the case for trusting testimony in general, one would need a vast independent track-record — but most of one’s evidence about that track-record is itself testimonial (history books, reports of others’ observations, experimental results one has not personally re-run). The reductionist programme cannot get off the ground without already trusting the source it claims to reduce.⁶¹ The anti-reductionist conclusion is that testimony is on epistemic par with perception, and that the burden of proof falls on whoever wants to reject a particular piece of it.

5.4 Testimonial Injustice

Both Hume and Reid assume the listener applies the same default across speakers. Miranda Fricker (Epistemic Injustice, 2007) showed this is empirically false: hearers systematically discount or inflate testimony on the basis of speakers’ identities, and testimonial injustice names a structural pathology in the credibility economy that any account of testimony has to face. Whichever side of the Hume/Reid dispute one takes, the phenomenon is real — for the reductionist, a discrimination in how the inferential checklist is applied; for the anti-reductionist, a corruption of the default itself. Because testimonial injustice operates at the level of the credibility economy and shapes who counts as a knower, its full treatment — alongside its companion category, Fricker’s hermeneutical injustice — comes in the shared-knowledge unit.

flowchart LR
  T["Someone tells you<br/>that <i>p</i>"]
  T --> HU["<b>Hume — reductionism</b><br/>No special credit for<br/>being told. Back it with<br/>independent evidence of<br/>the speaker's track record"]
  T --> RE["<b>Reid — anti-reductionism</b><br/>Testimony is a basic source,<br/>like perception. Acceptance<br/>is the default; reasons are<br/>needed to <i>reject</i>"]
  HU --> CO["Coady: pure reductionism<br/>is self-undermining —<br/>the track record is itself<br/>known by testimony"]
  HU --> FR["Fricker: neither default<br/>is applied evenly.<br/><b>Testimonial injustice</b><br/>discounts speakers<br/>for who they are"]
  RE --> FR

5.5 Questions

Questions 1 and 2 are retrieval checks — answer them from memory before you argue the rest.

1. Hume’s miracle argument in §X is reductionism applied to the hardest case. State its structure from memory: what does the wise man proportion belief to, where does our assurance in testimony derive from, and what must be true of a testimony’s falsehood before it can establish a miracle (§X.13)? Then decide: does the argument require general reductionism about testimony, or is it consistent with Reid’s credulity principle plus a specific defeater for “miraculous” reports?
2. State Coady’s self-undermining objection from memory: what would the reductionist need to assemble, and why can it not be assembled without trusting the very source it is meant to vet? Then judge: is this fatal, or can a local reductionism (case-by-case checking) avoid it?
3. Suppose Reid is right and we have a default propensity to believe what we are told. Does this principle apply equally to your parents, to a stranger asking for directions, to an anonymous Wikipedia editor, to an advertisement, and to a chatbot? If not, on what grounds do you draw the lines — and is that drawing-of-lines just reductionism with extra steps?
4. Fricker’s testimonial injustice presupposes that there is a “right” amount of credibility owed to a speaker, against which under-crediting is a wrong. Where does that owed amount come from? Is it Hume’s evidence-proportional measure or Reid’s default-acceptance, and does Fricker’s argument work better on one of the two?

6 What is intuition — and can it count as knowledge?

You see someone’s face and immediately know they are angry. You sit down to a maths problem and — before you’ve done any calculation — you “feel” the answer is around 500. You’re at a dinner party and something about the other guest makes you uneasy, though you couldn’t say why. These are all called “intuitions” — though already the noun is doing too much work, since the three cases involve quite different cognitive processes. The chapter that follows treats them under one heading because they share the phenomenology of immediacy, but stay alert to the lump: Kant’s a priori, Klein’s expert pattern-recognition under feedback, and Kahneman’s System 1 are all called “intuition” and are not the same thing.

The accounts that follow split into two heads. The first treats intuition as a cognitive kind — a way the mind reaches what is true. The second treats moral intuition specifically — the felt sense of right and wrong. The question closes by pressing both heads against Karikó’s expert intuition and the 18th-century slaveholder’s moral intuition.

flowchart LR
  I["Intuition"]
  I --> C["As a <b>cognitive kind</b>:<br/>a way the mind reaches<br/>what is true"]
  I --> M["As <b>moral intuition</b>:<br/>the felt sense of<br/>right and wrong"]
  C --> C1["Kant — the <i>a priori</i>:<br/>the framework any<br/>experience must already have"]
  C --> C2["Kahneman — System 1:<br/>fast pattern-matching<br/>on trained exposure"]
  C --> C3["Poincaré — the hunch<br/>that arrives on the<br/>step of the omnibus"]
  C --> C4["Merleau-Ponty — the body<br/>schema: knowing that was<br/>never propositional"]
  M --> M1["Rawls — reflective<br/>equilibrium: intuitions<br/>count as evidence"]
  M --> M2["de Waal — evolutionary<br/>roots: empathy, fairness,<br/>reciprocity, the thrown<br/>cucumber"]

6.1 Kant and A Priori Knowledge

Immanuel Kant made one of the most important distinctions in epistemology: between knowledge that comes from experience (a posteriori knowledge) and knowledge that is independent of experience (a priori knowledge).⁶⁵

“\(2 + 2 = 4\)” is a priori: you don’t need to go count things to know it. But Kant argued that some a priori knowledge is synthetic — it adds information about the world, not just unpacking definitions. The laws of Euclidean geometry were his central example. These, Kant thought, were conditions that the human mind imposes on experience, rather than reading off it.

The deeper point: Kant believed the mind is not a blank slate that receives impressions passively. It actively structures experience using categories (causation, substance) and forms of intuition (space and time), which on his view are not learned from experience but are the conditions any experience has to satisfy in the first place.

Kant is not being mystical. He is making a specific philosophical claim: some knowledge does not need empirical justification because it constitutes the framework within which empirical justification operates. Whether Kant is right about the details is contested. Whether the category of a priori knowledge is real is also contested, though less widely: Quine’s “Two Dogmas of Empiricism” (1951) argues that even apparently a priori truths are simply the most entrenched beliefs in our web of belief, and could in principle be revised under sufficient pressure.⁶⁶

6.2 Expert Intuition: Kahneman’s System 1

Daniel Kahneman (Thinking, Fast and Slow, 2011) distinguishes between:⁶⁷

• System 1: fast, automatic, intuitive, unconscious processing. It recognises patterns instantly.
• System 2: slow, deliberate, analytical, conscious reasoning.

Kahneman’s point: System 1 is often right, but for specific reasons. A chess grandmaster who “sees” the right move after a glance is not guessing. She is pattern-matching against tens of thousands of hours of encoded experience. Naturalistic Decision Making research (Gary Klein) shows that expert firefighters, nurses, and pilots routinely make excellent rapid decisions that they cannot fully articulate.⁶⁸ The info-box case is the awkward one for this picture: Karikó’s conviction that modified mRNA could be made into medicine was held for three decades in a domain where the prompt feedback Kahneman’s reliable intuitions require was missing — the verdict delayed by almost the span of a career. Whether a certainty sustained that long, with its correction that far off, is knowledge or merely obstinacy the biology eventually happened to reward is what the question below turns on.

But System 1 is also the source of cognitive biases: the availability heuristic, anchoring, the representativeness heuristic. The conditions under which intuition runs reliably (a regular domain, prompt feedback, accumulated trained exposure) and the conditions under which it does not (novel situations, delayed feedback, statistically misleading patterns) are themselves part of what Kahneman’s research has tried to specify. Whether the intuitions that turn up in ethics, philosophy, or everyday social life sit on the reliable or the unreliable side of that line is precisely what is at stake when someone says “I just know.”

6.3 Intuition in Mathematics and Science

Many mathematicians describe their work as guided by aesthetic intuition — a sense of which conjectures are worth pursuing, which approaches feel elegant. Henri Poincaré, recounting the work that became the theory of Fuchsian functions, described a sleepless night in which the unconscious seemed to do the sorting itself:

“The ideas rose in crowds; I felt them collide until pairs interlocked, so to speak, making a stable combination.” — Henri Poincaré, Science and Method (1908), Book I, Chapter III (“Mathematical Discovery”)⁶⁹

Days later came the famous moment: boarding an omnibus at Coutances, mid-conversation about something else entirely, the decisive idea arrived at the instant his foot touched the step — unconnected, as he tells it, to any thought that had been preparing the way. Physicists describe similar experiences. Einstein said that the general theory of relativity emerged from a “thought experiment” — intuitive visualisation — before the mathematics was worked out.

That intuition can be a source of knowledge in mathematics — a domain otherwise defined by rigorous proof — is one of the harder facts to fit into any tidy account of justification.

6.4 Embodied Knowing: Merleau-Ponty

The cognitive-kind accounts above — Kant’s a priori, Kahneman’s System 1, the mathematician’s hunch — all treat intuition as something that happens in the mind, even if quickly or unconsciously. Maurice Merleau-Ponty (Phenomenology of Perception, 1945) challenged this picture at its root: knowing is not primarily a mental act. It is a bodily achievement.⁷⁰

A skilled cyclist does not think about balance — their body maintains it. A violinist does not consciously calculate finger positions — their hands know where to go. Merleau-Ponty’s term is the body schema: a pre-reflective, non-propositional knowledge that the body carries, that cannot be adequately captured as a set of beliefs or rules. When a blind person navigates with a cane, the cane becomes an extension of their body schema — they feel the world through it directly, not by consciously inferring from the pressure in their hand.

This matters for epistemology: it suggests that much of our knowledge of the world is not propositional (not a matter of holding beliefs that are true or false) but practical and embodied — a kind of knowing that cannot be fully translated into the JTB framework. The question “is it justified?” doesn’t quite apply to the cyclist’s balance. Whether this counts as a deficiency in the cyclist’s knowledge or as a limit in the JTB framework is itself contested. Defenders of JTB will say procedural skill is simply not knowledge in the relevant sense — it is ability or competence, distinct from knowledge that; Merleau-Ponty’s heirs will reply that any framework which excludes the bulk of competent action from “knowledge” has shrunk the concept until it no longer fits the human case.

6.5 Moral Intuitions: Rawls’ Reflective Equilibrium

In A Theory of Justice (1971), John Rawls described a method for moral reasoning he called reflective equilibrium (§9).⁷¹ The idea: we start with our pre-theoretical moral intuitions — the strong sense that torturing children for fun is wrong, that people should not be punished for crimes they didn’t commit. These intuitions are data. We then construct general principles that accommodate as many of them as possible. Where the principles and intuitions conflict, we revise — sometimes the principle, sometimes the intuition.

“We can think of the original position as a device of representation.” — John Rawls, Political Liberalism, Lecture I (“Fundamental Ideas”)⁷²

Rawls is treating moral intuitions as evidence. Not infallible evidence — intuitions can be culturally biased, historically contingent, or the product of prejudice. But evidence nonetheless, on his view, in a sense that does not collapse into “feelings to be explained away.”

The Trolley Problem (see Ethics unit) works precisely because our intuitions give incompatible verdicts: we feel pulling the lever is permissible, but pushing a fat man off a bridge to stop the trolley feels like murder, even though the arithmetic of consequences is identical.

6.6 The Evolutionary Roots of Moral Intuition

Kahneman’s System 1 is rapid and pattern-based, but where do the patterns come from? Primatologist Frans de Waal (Primates and Philosophers, 2006; The Age of Empathy, 2009) argues that the roots of human moral intuition go deeper than culture or rational reflection — they are evolutionary:

Capuchin monkeys will refuse a reward (a cucumber slice) if they observe another capuchin receiving a better reward (a grape) for the same task. The refusal is dramatic: they throw the cucumber at the researcher. This is not calculated indignation. It is a visceral response to perceived unfairness — a functional sense of equity in a non-human primate.⁷³

Chimpanzees console distressed companions, share food, reconcile after conflicts, and punish free-riders in cooperative tasks. De Waal’s conclusion is that the building blocks of human morality — empathy, fairness, reciprocity, conflict resolution — are not unique to humans, and that they evolved in social species under selective pressure for cooperation.

This has a consequence for how we understand moral intuitions — though it leaves entirely open whether it has any consequence for how we should trust them. They are not arbitrary cultural deposits, and they are not purely rational constructions. They are the elaboration, in a language-using, culturally complex species, of responses that evolution shaped over millions of years because social living required them. When you feel that pushing someone off a bridge to save five people is wrong in a way that pulling a lever is not, you are responding from a deep structure that predates culture. Whether that deep structure justifies the response, however — or merely explains it — is a separate question, and one that the descriptive evolutionary account cannot settle on its own. (To slide from “evolution shaped this response” to “this response is correct” is to commit the naturalistic fallacy: the descriptive premise does not entail the normative conclusion.)

What the evolutionary account does explain is why moral intuitions feel so compelling, and why philosophical theories that ignore them entirely tend to produce conclusions that nobody can actually live by — without thereby showing those conclusions to be wrong, only showing why they will be unliveable.

6.7 Questions

Questions 3 and 5 are retrieval checks — answer them from memory before you argue the rest.

1. Is intuition a form of implicit knowledge, or just a feeling that may or may not track truth?
2. Rawls treats moral intuitions as evidence. But intuitions about who counts as a person, or who deserves rights, have been profoundly wrong historically. How do we know which intuitions to trust?
3. Name the conditions under which Kahneman and Klein say expert intuition runs reliably — and the conditions under which it does not. Which side of that line does Karikó’s thirty-year conviction sit on? And does the reliable side mean intuition is ultimately reducible to pattern-matching from experience — or is there something else going on?
4. Kant argues that some knowledge is a priori — built into the structure of the mind. Is this the same as intuition, or something different?
5. State Merleau-Ponty’s term for the knowledge the body carries, and what the blind person’s cane shows about it. Is that the same kind of knowledge as knowing that Paris is the capital of France — or a genuinely different kind? Does the JTB framework apply to it?
6. De Waal’s capuchin monkeys refuse unfair treatment. Does this show that a sense of fairness is “natural,” or does it just show that capuchins have learned what behaviour gets them rewards?

7 How does shared knowledge differ from personal knowledge?

For roughly a decade, one of the most cited findings in behavioural science held that a tiny change to a form — asking people to sign an honesty declaration at the top, before filling it in, rather than at the bottom — made them measurably more honest. Insurers redesigned paperwork around it; governments ran trials on it; it was taught to executives. Then the data turned out to be fabricated. The case forces a definitional question: does “shared knowledge” mean knowledge held in common among those competent to evaluate it (the narrow reading), or knowledge functioning as part of a community’s working understanding of itself (the broad reading)? On the narrow reading the fabrication was never knowledge at all, and the competent eventually expelled it; on the broad reading it functioned as shared knowledge — cited, taught, applied — for ten years before anyone caught it. The two readings give different verdicts, and the rest of the unit’s questions — how personal knowledge becomes shared, what makes collective knowledge reliable, who counts as a knower — sit on top of that split.

7.1 Popper’s Falsifiability and the Mark of Shared Knowledge

Karl Popper (The Logic of Scientific Discovery, 1934; Conjectures and Refutations, 1963) proposed that what distinguishes science from non-science is falsifiability: a claim is scientific if and only if it specifies what evidence would count against it. A theory that can explain any possible observation is not scientific — it is unfalsifiable.⁸⁰

Einstein’s general theory of relativity predicted that starlight would bend around the sun — a precise, quantitative, risky prediction. Eddington’s 1919 eclipse observations confirmed it.⁸¹ The theory could have been disconfirmed.

Popper’s criterion is a criterion for shared knowledge: falsifiable claims are in principle checkable by anyone, anywhere, with the right equipment, and so are public, repeatable, and open to criticism in a way that protects them from the private foibles of any single knower.

By contrast: “Everything happens for a reason” or “God loves you” may be deeply meaningful as personal knowledge, but they are unfalsifiable — no observation could disconfirm them. This does not by itself make them false or worthless; it does mean they are not the kind of claim Popper’s machinery was designed to adjudicate.

7.2 The Wikipedia Problem

Wikipedia is enormous and openly editable by anyone. This raises a genuine epistemological puzzle:

• A 2005 study in Nature found Wikipedia science articles contained on average 4 errors per article to Britannica’s 3 — closer than most expected, though not a clean Wikipedia victory.⁸²
• But Wikipedia is unreliable in specific ways: recent events, niche topics, and contested political subjects suffer from systemic bias and edit wars.
• The reliability of any Wikipedia article depends on how many expert eyes have reviewed it, how contested the topic is, and how resistant the community is to motivated editing.

The question is not “is Wikipedia reliable?” but “what makes any source of shared knowledge reliable?” Reliability in any such source seems to depend on processes — peer review, replication, adversarial testing, transparent methods, institutional accountability — rather than on the authority or sincerity of any individual contributor.

7.3 Social Epistemology and Epistemic Injustice

Miranda Fricker’s Epistemic Injustice (2007) names two structural pathologies in how knowledge moves through communities. The first, testimonial injustice, sits inside any default of credibility: hearers withhold credibility on the basis of identity prejudice. Her central case is Tom Robinson, the defendant in Harper Lee’s To Kill a Mockingbird: a Black man in a 1935 Alabama courtroom gives a truthful account of events, and the all-white jury disbelieves him — not because his account is implausible but because racial prejudice deflates his credibility in their minds. He suffers a double wrong — the original injury and the epistemic wrong of not being credited as a source of testimony about his own life. The structural point is that testimony does not merely succeed or fail piece by piece; it operates inside a credibility economy in which some speakers are systematically discounted and others systematically overweighted.⁸³

The second, hermeneutical injustice, is not about any particular speaker’s credibility but about the shared conceptual resources by which any speaker can articulate experience at all. Before the term “sexual harassment” entered common vocabulary, women who experienced it had difficulty even naming what was being done to them. The gap was not in their experience but in the shared interpretive apparatus — and the socially powerful had disproportionately shaped that apparatus. Hermeneutical injustice is therefore a defect in the collective knowledge-making infrastructure, not in any individual transaction.

The epistemological consequence: who counts as a knower, whose testimony is taken seriously, and which experiences have names available for them are not personal matters but political ones, with Fricker’s two categories naming two layers — case-by-case credibility and the underlying vocabulary — at which that politics enters the shape of what gets known in common.

7.4 When the Machinery Works: A Case Study in Successful Shared Knowledge

The failures are vivid; the successes are taken for granted. Here is a success — and on the narrow reading of shared knowledge, this is what success looks like.

In March 2004 and June 2005, the South Korean scientist Hwang Woo-suk published two papers in Science claiming to have produced first a cloned human blastocyst and then eleven patient-specific embryonic stem-cell lines through somatic cell nuclear transfer — a landmark result that would have revolutionised regenerative medicine.⁸⁴ The papers received enormous international attention; Hwang was treated as a national hero in South Korea. Within a year, everything had collapsed. Suspicions were raised first by junior members of his own team, who reported anomalies in the data — and the coercive sourcing of human eggs from female lab members — to the Korean public broadcaster MBC’s investigative programme PD Su-cheop, which aired its initial exposé in November 2005. Other researchers attempted replication and could not produce his results. Science commissioned an independent investigation; Seoul National University conducted its own. On 10 January 2006 the SNU committee concluded that all eleven cell lines were fabricated; two days later, on 12 January, Science unconditionally retracted both papers.⁸⁵ Hwang was subsequently convicted of embezzlement and bioethics violations.

What is notable here is not the fraud — fraud happens. What is notable is the speed and mechanism of correction in this case. The system that detected the fraud is the one Popper described: transparent methods (published data that could be scrutinised), replication attempts by independent researchers, adversarial peer review operating across institutional and national lines, and a formal retraction process. No authority figure declared the papers fraudulent; the community of practice identified the specific discrepancies, applied the falsification criterion, and revised the knowledge base. From the first broadcast suspicions in November 2005 to formal retraction took less than two months. Many fraud cases run slower: Schön in physics, Stapel in social psychology, and Macchiarini in surgery each took years from suspicion to retraction, and some fraud is presumably never detected at all.

The contrast with the tobacco industry’s decades-long campaign to deny that cigarettes cause lung cancer can be read in two ways. On one reading, Hwang shows the epistemic machinery working as it is supposed to, and tobacco shows what happens when motivated actors target the social infrastructure that machinery depends on. On a different reading, Hwang is a rare and unusually fast case (the journalists, the junior whistleblowers, the SNU committee all functioned simultaneously) — closer to an exhibit of best-case conditions than to a base rate against which tobacco is the failure mode. The two readings agree on what conditions correction depends on: public data, independent replication, no institutional interest in protecting the result, and professional norms strong enough to survive the retraction of a hero.

7.5 When Shared Knowledge Breaks Down: A Case Study

Students now alive have lived through one of the most dramatic crises of shared knowledge since the Enlightenment. Consider two cases:

COVID-19 and scientific authority (2020–2022): Epidemiologists reached rapid consensus on a number of key facts — transmission routes, vaccine efficacy, the role of prior immunity. These were falsifiable claims subjected to enormous scrutiny. And yet a substantial subset of the public in high-income countries rejected some of these claims. The empirical literature on what exactly drove vaccine refusal is more mixed than the satisfying “trust collapse” story suggests: partisan identity, prior health-system experiences, social-network effects, and individualised risk perception all contributed, and the relative weights vary by country and cohort. The narrower phenomenon — that epistemically well-supported claims sometimes fail to be taken up by parts of the public among whom the relevant institutions had become objects of organised distrust — is real, even if “institutional trust collapsed” is too broad a description. The claims were falsifiable; the mechanism for adjudicating them did exist. Shared knowledge in the broad sense (functioning as common public understanding) requires functioning institutions to make and communicate the claim, and populations willing to engage with those institutions, on top of the narrow Popperian conditions.

Climate science and manufactured doubt (1980s–present): Naomi Oreskes and Erik Conway (Merchants of Doubt, 2010) documented a deliberate strategy, used first by the tobacco industry and later by fossil fuel interests, of manufacturing apparent scientific controversy where genuine consensus existed. The strategy was not to produce better evidence; it was to create the impression of a debate, exploiting the journalistic norm of “balance” to give fringe views equal airtime with scientific consensus.

Popper’s criterion tells us what distinguishes a good scientific claim from a bad one. It does not tell us how to defend shared knowledge when well-funded actors are motivated to corrupt the epistemic environment, and the question of how communities produce, evaluate, and transmit knowledge under those pressures is what social epistemology has had to take up.

7.6 Standpoint Epistemology

Standpoint epistemology pushes the broad reading: who counts as a knower and which positions yield evidence are not questions Popper’s machinery alone can settle. Sandra Harding and other feminist epistemologists developed standpoint theory: knowledge is always situated.⁸⁷ Knowers occupy particular social positions (gender, class, race, history), and those positions can make certain things more visible or less visible.

The provocative claim is that the standpoint of the marginalized can provide epistemic advantages in certain domains, not as a consolation prize but as a description of where some kinds of evidence can and cannot be gathered from. A worker on a factory floor knows things about how the factory operates that the CEO does not — things the CEO cannot learn from the quarterly report. A patient knows things about their own illness that the doctor cannot access from the outside. An enslaved person understands the plantation economy in ways the plantation owner structurally cannot. The view is not that all standpoints are equally valid for all questions; it is that standpoint is a variable in knowledge production that the detached-universal-observer picture has tended to suppress.

The objection: does standpoint theory commit itself to relativism about truth? If knowledge depends on standpoint, do different standpoints generate incompatible “truths”? Harding’s response is that strong objectivity requires taking standpoints seriously as variables rather than eliminating the knower from the picture, and that pretending the knower has no position can itself produce concealed partiality. Naturalist critics (Susan Haack, Philip Kitcher, broadly Quinean philosophers of science) reply that situatedness is one variable among many that ordinary scientific method already tracks and discounts, and that Harding treats one failure-mode (concealed partiality) as if it were the only alternative to her programme.⁸⁸

flowchart LR
  P["<b>Personal knowledge</b><br/>one knower's belief,<br/>experience, skill"]
  P --> M["<b>The machinery</b><br/>falsifiable claims ·<br/>transparent methods ·<br/>peer review · replication ·<br/>adversarial testing"]
  M --> S["<b>Shared knowledge</b><br/>public, repeatable,<br/>open to correction"]
  S --> W["Works — Hwang's cloned<br/>stem-cell lines: fraud<br/>found and retracted<br/>within months"]
  S --> B1["Breaks — manufactured<br/>doubt outruns correction.<br/>Brandolini's Law"]
  S --> B2["Breaks upstream — who<br/>counts as a knower, and<br/>which experiences have<br/>names, is settled before<br/>the machinery starts.<br/>Fricker · standpoint theory"]

7.7 Questions

Questions 1 and 3 are retrieval checks — answer them from memory before you argue the rest.

1. State Popper’s falsifiability criterion from memory: what must a claim specify to count as scientific, and what made Einstein’s light-bending prediction the model case? Then press the criterion: is mathematics falsifiable? Is history? Are all forms of shared knowledge scientific?
2. Wikipedia’s reliability depends on crowd-sourced editing and adversarial correction. Is this fundamentally different from how academic peer review works — or the same thing with different social mechanisms?
3. On 11 October 1991, Anita Hill — a law professor formerly Clarence Thomas’s assistant at the EEOC — testified before the all-male, all-white Senate Judiciary Committee during Thomas’s Supreme Court confirmation. She described specific episodes: Thomas pressing her to date him after she declined; describing pornographic films in detail; asking, “Who has put pubic hair on my Coke?” Senators publicly suggested she was a “scorned woman” or fabricating for political gain; a second corroborating witness (Angela Wright) was subpoenaed but not called. Thomas was confirmed 52–48. Subsequent reporting (Mayer and Abramson, Strange Justice, 1994) corroborated much of Hill’s account; in 2018 several senators publicly revisited the 1991 committee’s conduct.⁸⁹ State Fricker’s test for a distinctively epistemic wrong: what must happen to a speaker’s credibility, and on what grounds, for her to be wronged as a knower — over and above the moral wrong of being unfairly disbelieved? Then apply the test: was the wrong done to Hill only the moral one, or the epistemic one on top of it? Defend your answer.
4. If a person’s social position gives them epistemically privileged access to certain truths, does that mean the social sciences should deliberately seek out marginalized perspectives? What are the limits of this idea?

8 What is a knower?

Knowledge requires a knower. But who — or what — is the knower? A disembodied thinking thing, as Descartes suggested? A bundle of impressions, as Hume argued? A socially situated embodied person? An artificial system? The question runs through epistemology, philosophy of mind, and personal identity, and the major answers continue to disagree.

8.1 Descartes’ Ego

After stripping everything away through radical doubt, Descartes arrives at the cogito: he exists as a res cogitans — a thinking thing. This is a substance, a subject. What kind of thing is it?

Descartes is a substance dualist: the mind is a non-physical substance, distinct from the body (res extensa). This creates the mind-body problem: how does an immaterial mind interact with a physical body? (Descartes’ answer — they interact in the pineal gland — is not taken seriously, and the interaction problem has never been fully solved.)

The Cartesian knower is, on this picture, a unified self, transparent to itself, essentially non-physical, and numerically the same through time. Each of those four claims has been seriously challenged in the centuries since.

8.2 Hume’s Bundle Theory

Hume looked inward and reported what he found:

“For my part, when I enter most intimately into what I call myself, I always stumble on some particular perception or other, of heat or cold, light or shade, love or hatred, pain or pleasure. I never can catch myself at any time without a perception, and can never observe any thing but the perception.” — David Hume, A Treatise of Human Nature, Book I, Part IV, Section VI⁹²

There is no self beyond the stream of perceptions, on Hume’s account. The self is a “bundle or collection of different perceptions, which succeed each other with inconceivable rapidity,” and what we ordinarily call the “self” is the impression of continuity laid over a flux that contains no persisting thing.

If the knower is not a stable, unified thing, who is accumulating knowledge over time — and, in the dementia case, who holds the authority a life-or-death decision needs? The advance-wish case makes this concrete: by Hume’s lights there is no persisting self in the woman who formed the wish, and none in the woman who must now be the subject of it — only successive bundles — so the question of which “self” the decision belongs to has no further fact beneath it to settle who is right.

8.3 Locke on Personal Identity and Memory

Consider Leonard Shelby, the protagonist of Christopher Nolan’s Memento (2000). Leonard suffers from anterograde amnesia — he cannot form new long-term memories. Everything since the injury slips away within minutes; scene after scene, he surfaces with no recollection of how he got there. He navigates the world through photographs, tattoos, and Polaroids with handwritten notes. He is, in some important sense, the same person from day to day — same body, same pre-injury memories, same name. But is he the same knower? Again and again he must reconstruct his epistemic situation from external records he cannot remember making. Knowledge that ordinary knowers accumulate over time — who to trust, what has already been tried, what conclusions have been drawn — he must reinvent or read from a note.

John Locke (An Essay Concerning Human Understanding, Book II, Chapter 27) gave a different answer: personal identity consists in continuity of consciousness, specifically memory.⁹³ You are the same person who was punished at school because you remember being punished. You are not the same person as Socrates because you remember nothing of being Socrates.

Derek Parfit (Reasons and Persons, 1984) pushed this to its limit with the teleporter case: you step into a teleporter, your body is destroyed and an exact replica — atom-for-atom — is assembled on Mars with all your memories. Is the person on Mars you? If yes: what survived the destruction of your body? If no: when exactly did you cease to exist? Parfit’s conclusion is that personal identity is not what matters. What matters — for ethics, for prudential reasoning — is psychological continuity, with the question of whether that continuity adds up to strict identity left as a further dispute.⁹⁴

Locke’s verdict on the dementia case is sharp, and it cuts against the use the case wants to make of it. With psychological continuity broken, the competent woman who formed the wish and the confused woman in the bed are, by Locke’s own criterion, not the same person — so the advance wish is an instruction left by someone who, strictly, is no longer here. The criterion built to honour continuity ends by dissolving the very authority an advance directive was meant to carry.

8.4 The Situated Knower

Feminist epistemologists and philosophers of science argued that the Cartesian picture of the knower as a disembodied, universal reasoner is not only philosophically wrong but politically loaded. The argument is not that Descartes happened to be a French Catholic gentleman of the 17th century — that biographical fact, on its own, no more impeaches the cogito than Newton’s Anglicanism impeaches the Principia. The argument is that the very content of his picture — the knower as Nobody in Particular, suspended in pure reason — does covert epistemic work: it treats the assumptions characteristic of his social position (which questions are worth asking, which experiences count as standard, what counts as a neutral starting point) as the universal default. The “view from nowhere,” on this critique, hides which somewhere it has been built from.

Donna Haraway (Situated Knowledges, 1988) argued for partial perspective: knowledge claims that acknowledge their situatedness are more honest and, in certain respects, more objective than claims that pretend to a “god’s-eye view” they cannot have.⁹⁵

Sandra Harding’s strong objectivity requires that the social position of the knower be included as a variable in the evaluation of knowledge, not suppressed.⁹⁶ A scientist’s class, gender, and cultural assumptions affect what hypotheses they consider, what they observe, and how they interpret results. On her account, suppressing those variables in the name of objectivity is itself an epistemic move that can go wrong.

8.5 Can Machines Know?

Artificial intelligence has made the question urgent: what is it for anything to be a knower? Large language models can pass medical licensing exams, write legal briefs, and produce plausible philosophical essays — including plausible philosophical essays about whether large language models can know things.

The standard responses:

• No — they lack understanding (John Searle’s Chinese Room argument, 1980): a system that manipulates symbols according to rules doesn’t understand the symbols, any more than someone who follows rules for manipulating Chinese characters understands Chinese.⁹⁷ Searle’s Chinese Room has been replied to in nearly as many ways as there are philosophers, but the intuition it pumps remains powerful.
• Yes, or approaching it — functionalists argue that if a system behaves in all the ways a knower behaves, the label applies. What more could “knowing” require? If you say “but there’s nothing it’s like to be the system,” you are invoking subjective experience as the criterion — which immediately raises the question of how you verify that criterion in other humans.
• It’s the wrong question — the concept of “knowing” was developed for human beings in human social contexts. Applying it to machines may require revision of the concept rather than a verdict either way.

flowchart LR
  Q["Who — or what —<br/>is the knower?"]
  Q --> D["<b>A thinking substance</b><br/>Descartes: unified,<br/>transparent to itself,<br/>non-physical, the same<br/>through time"]
  Q --> H["<b>A bundle of perceptions</b><br/>Hume: he looks inward<br/>and never catches a self,<br/>only the stream"]
  Q --> L["<b>A chain of memory</b><br/>Locke: identity is<br/>continuity of consciousness<br/>— Parfit's teleporter<br/>asks what survived"]
  Q --> S["<b>A situated body</b><br/>Haraway, Harding: partial<br/>perspective, and no<br/>view from nowhere"]
  Q --> M["<b>An artificial system?</b><br/>Searle: symbol-shuffling<br/>is not understanding.<br/>Functionalism: what more<br/>could knowing require?"]

8.6 Questions

Questions 1 and 4 are retrieval checks — answer them from memory before you argue the rest.

1. State what Hume reports when he enters most intimately into what he calls himself: what does he always stumble on, and what does he never catch? What, then, is the self on the bundle account — and can there be knowledge without a stable knower?
2. Locke grounds personal identity in memory. But memory is notoriously unreliable. Does this mean our sense of who we are is built on an unreliable foundation?
3. Parfit argues personal identity is not what matters — psychological continuity is. Does this view change how you think about self-interest, or about why death is bad?
4. The Cartesian knower is disembodied. State Clark and Chalmers’s “extended mind” thesis⁹⁸: on their argument, when does an external prop — a notebook, a phone, Leonard Shelby’s Polaroids and tattoos — count as part of the knower rather than a tool the knower consults? What counts as part of you?
5. Should a sufficiently sophisticated AI be considered a knower? What exactly does it lack?

8.7 The Horizon File, Reopened

At the Post Office question you signed a default: either the wise man proportions belief to the evidence, and an audited record outweighs a bare denial case by case — or testimony runs on a default of acceptance that the Post Office withdrew, wholesale, from a class of speaker it had decided not to credit. You ruled at the counter, in 2010, before any bug was public. The lessons since have been quietly pressing on that ruling.

Take out what you wrote and test it against three pressures it did not face at the time:

• Fricker’s credibility economy. The shared-knowledge lesson gave the machinery its full name: Tom Robinson truthful and disbelieved in a 1935 courtroom, Anita Hill discounted before a 1991 committee — hearers systematically deflating testimony by the speaker’s identity. Does your default handle a hearer who is not neutral, or did it quietly assume one? Whichever side you signed, say where identity-driven deflation enters your account: as a checklist misapplied, or as a default corrupted.
• The Gino inversion. The honesty-study case ran the Horizon pattern in reverse: a celebrated professor’s word was credited for a decade while the correctors were unpaid outsiders who ended up sued. If your default sorts the subpostmasters’ case, does it also sort this one — where the over-credited speaker was high-status and human, and the numbers were on the side of the doubters?
• The machine as knower. The knower lesson asked what it would take for anything to be a knower at all — and the Post Office had already granted Horizon the standing of one. On your account of testimony, can a machine’s record testify? If it cannot, say what exactly the courts were admitting into evidence — and what your answer does to the standing of every audited system a court will meet after Horizon.

Re-sign or revise — in writing, a few sentences either way. A revision is not a defeat; an unexamined signature is. Name what the intervening lessons changed, or say why they changed nothing.

9 Media

Novels, films, and artworks that illuminate the questions above:

• Plato, The Allegory of the Cave (in Republic, Book VII) — Not a novel, but the most powerful thought experiment in the history of epistemology. Prisoners in a cave mistake shadows for reality. One escapes and sees the sun. What is the relationship between education and liberation from illusion?
• Jorge Luis Borges, “Tlön, Uqbar, Orbis Tertius” (in Ficciones, 1944) — A secret society invents an encyclopedia of an imaginary world; the invented world begins to replace the real one. A meditation on reality, language, and collective knowledge-making.
• Philip K. Dick, Do Androids Dream of Electric Sheep? (1968) — What distinguishes genuine empathy from its simulation? What is the status of artificial minds as knowers? The basis for Blade Runner.
• Christopher Nolan, Memento (2000) — A man with anterograde amnesia (no new long-term memories) investigates his wife’s murder. Parfit’s thought experiments on personal identity and memory made visceral.
• The Wachowskis, The Matrix (1999) — Brain-in-a-vat scepticism as blockbuster film. Legitimately useful for discussing Descartes, despite being pulp cinema.
• Ingmar Bergman (dir.), Persona (1966) — A nurse and her mute patient gradually exchange identities. Bergman’s most concentrated treatment of personal identity as a fluid, performed, and partly fictional construction. The 80-minute running time is itself an argument: the disintegration of the self is shown faster than any theory could describe it.
• Kazuo Ishiguro, Never Let Me Go (2005) — Clone children raised for organ donation. Questions of what it means to be a knower, a person, a self — obliquely and devastatingly rendered.

10 Bibliography

11 Notes