Contextualism, Invariantism, and Pragmatic Encroachment

It is Friday afternoon. You and your spouse drive past the bank intending to deposit your pay cheques, and the queues are long. You suggest coming back tomorrow. Your spouse asks whether the bank is open on Saturdays. You say you know it is, since you were there two Saturdays ago.

Now change one thing. A large cheque is about to bounce if the money is not in the account by Monday, and your spouse points out that banks do change their hours. You were there two Saturdays ago. Do you know the bank will be open tomorrow? Most people say that in this version you do not, and that you should go in and ask.

Keith DeRose's bank cases, published in 1992, turn on a puzzle. Your evidence is identical in the two versions. The bank's hours are identical. Your belief is equally true, equally well formed, equally reliable. And yet the verdict on whether you know appears to change. Something other than the traditional ingredients of knowledge is making a difference, and the question is what.

Three answers

Each position in this debate keeps two of three attractive claims and gives up the third.

The first claim is that the intuitions in the bank cases are correct: the sentence "I know the bank will be open" is true in the first case and false in the second. The second claim is that the sentence means the same thing in both cases, so that its truth depends only on facts about the subject. The third claim is that knowledge depends only on truth-relevant factors, such as evidence and reliability, and not on how much the subject has riding on the question.

Give up the second claim and you are a contextualist. Give up the third and you are a subject-sensitive invariantist. Give up the first and you are a classical invariantist who owes an explanation of why the intuitions mislead.

Contextualism

The model is the semantics of gradable and context-sensitive words. Whether "Amsterdam is flat" is true depends on what standard of flatness the conversation has adopted; a physicist concerned with surfaces at the micron scale and a cyclist planning a route are not contradicting each other when one says it is not flat and the other says it is. Contextualists hold that "knows" behaves similarly. There is a scale of epistemic positions, and conversational context determines how high on the scale a subject must stand for the word to apply.

Three versions were developed in the 1990s, and they differ in what moves the standards.

Stewart Cohen's version ties standards to the attributor's practical concerns and to what alternatives are salient. In his airport case, Mary and John are about to fly via Chicago and have an important meeting there. They overhear Smith, who has checked his itinerary, say he knows the plane stops in Chicago. Given what hangs on it for them, they reasonably say that Smith does not know, since itineraries contain misprints. Cohen argues that both Smith and the couple are speaking truly, because they are speaking in contexts with different standards.

DeRose's version ties standards to how far into modal space a subject's belief must be sensitive. In ordinary contexts it suffices that one would not believe falsely in nearby worlds. When a sceptical hypothesis is raised, the range of worlds that matter expands to include it, and a belief insensitive to it no longer counts as knowledge in that context.

David Lewis's version, from his 1996 paper "Elusive Knowledge", is the most radical. To know is to have evidence that eliminates every possibility in which the proposition is false, "psst", as Lewis says, except those we are properly ignoring. Which possibilities are properly ignored is governed by rules, and one of them, the Rule of Attention, says that a possibility that has been attended to is not being ignored. Lewis draws the conclusion explicitly: epistemology destroys knowledge. The moment one considers the demon hypothesis in the seminar room, one ceases to know that one has hands, and one recovers the knowledge on leaving.

The response to scepticism

Contextualism's main selling point is its treatment of the closure-based sceptical argument from the intermediate tier. The argument runs: if I know I have hands, I know I am not a handless brain in a vat; I do not know I am not a brain in a vat; so I do not know I have hands.

On the contextualist account, each premise is true at the standards in force when the argument is considered. Raising the sceptical hypothesis raises the standards, and at high standards the argument is sound: I do not know I have hands, relative to that context. But this does not threaten my ordinary claims to knowledge, which are made at ordinary standards and are true at those standards. Closure holds within each context. What the sceptic exploits is the shift in context the argument itself produces.

The package is attractive. It keeps closure, explains why scepticism is compelling in the seminar room, and explains why the compulsion evaporates outside it. It gives the sceptic a victory that costs ordinary knowledge nothing.

The problems for contextualism

Semantic blindness

Stephen Schiffer's 1996 objection is that if "knows" is context-sensitive, competent speakers are strangely unaware of it. When someone says "Amsterdam is flat" and someone else says "Amsterdam is not flat", both parties can be brought to see that they mean different standards and are not disagreeing. But the person moved by the sceptical argument does not think "I do not know I have hands, in the high-standards sense, though I know it in the ordinary sense". They think they have discovered that they do not know, full stop. Contextualism must attribute to competent speakers a systematic blindness to the semantics of a common word.

Disagreement and retraction

If Cohen's couple and Smith are speaking in different contexts, their claims do not conflict, any more than two people saying "I am hungry" contradict each other. But the couple do seem to be disagreeing with Smith. They seem to be saying that he is wrong. Contextualism has to explain the appearance of disagreement as an illusion, and the same pressure appears with retraction. Having been persuaded by the sceptic, people often say "I was wrong, I did not know", rather than "What I said was true in my earlier context".

The linguistic evidence

Jason Stanley argued in 2004 that "knows" lacks the linguistic features of the context-sensitive expressions it is supposed to resemble. Gradable adjectives such as "flat" and "tall" accept degree modifiers: very flat, flatter than, rather tall. "Knows" does not, at least not in the same way; "very knows" is ungrammatical and "knows better than" means something quite different. If "knows" were context-sensitive in the way contextualists claim, it should behave like the words it is modelled on, and it does not.

Contextualists have replied by finding other models, including quantifier domain restriction, where "every bottle is empty" is true in the kitchen despite the world's many full bottles. This model does not require gradability, but it brings its own problems, since restriction is typically easy to make explicit and "knows" resists such explicit qualification.

Invariantism

Classical invariantism holds that the standards for knowledge are fixed and that they depend only on truth-relevant factors. There are two classical options.

Sceptical invariantism, defended by Peter Unger in 1975, sets the standards very high. Knowledge requires certainty, and almost nothing is known. Our ordinary knowledge attributions are false but useful approximations, in the way "the room is empty" is false when there is dust in it. Almost no one holds this now, but it has the virtue of taking seriously the idea that the sceptic's standards are the real ones.

Moderate invariantism sets the standards at an ordinary level. You know the bank will be open in both cases. Your reluctance to say so in the high-stakes case is explained pragmatically: it would be misleading to assert that you know, since the assertion would suggest you have checked in the way the situation demands, or it would imply that further inquiry is unwarranted. Patrick Rysiew and Jessica Brown have developed versions of this strategy. The difficulty is that the intuition in the high-stakes case seems to concern whether the subject knows, not whether it would be appropriate to say so, and a pragmatic explanation of a semantic intuition is always vulnerable to the charge that it is ad hoc.

Subject-sensitive invariantism

John Hawthorne in 2004 and Jason Stanley in 2005 proposed that we keep the intuitions and keep the single meaning of "knows", and give up the purity of knowledge instead. Whether you know that the bank will be open depends on whether you are in the high-stakes case, because knowledge is connected to action. Jeremy Fantl and Matthew McGrath developed the principle that underwrites this.

The first premise is the knowledge-action principle, and it has independent support. We criticise people for acting on what they do not know ("you shouldn't have gone ahead, you didn't know the bridge would hold"), and we defend people by citing their knowledge ("I knew the bank would be open, so I went home"). If knowledge licenses action, and whether an action is licensed depends on what is at stake, then knowledge must depend on what is at stake.

The costs of subject-sensitivity

The view's problems are the mirror image of contextualism's. Because knowledge now depends on the subject's situation rather than the attributor's, the view generates strange temporal and modal claims.

Consider someone whose stakes rise over time. On Friday morning the cheque is small; by Friday afternoon a larger sum has arrived that will bounce. The subject's evidence has not changed. Subject-sensitive invariantism says that they knew in the morning and do not know in the afternoon. It follows that they lost knowledge without losing evidence, without any change in their belief, and without any change in the world except in their bank account.

Or consider counterfactuals. "If the stakes had been higher, I would not have known the bank was open" sounds bizarre, as does "I know it, but if more depended on it, I wouldn't". These sentences are true on the view, and they sound false.

A further problem concerns ignorant high stakes. Suppose the cheque will bounce but you do not know this. Your stakes are high, but you have no idea. Does the view say you do not know the bank will be open? If it does, knowledge depends on facts the subject is unaware of in a way that seems disconnected from anything epistemic. If it does not, and only perceived stakes matter, the view has become a kind of psychologism about stakes, and it is unclear why a belief about stakes should matter to knowledge when the stakes themselves do not.

Where this leaves closure and scepticism

The three positions take different routes around the sceptical argument.

The contextualist grants the sceptic's conclusion at the sceptic's standards and denies that it touches ordinary knowledge claims. The moderate invariantist denies the second premise outright: you do know you are not a brain in a vat, at ordinary standards, since that is the only standard there is, and the appearance of ignorance is a pragmatic effect. The subject-sensitive invariantist can say that in the seminar room, where nothing practical hangs on the question, the stakes are low and you know; the sceptic's argument trades on treating the question as if it were practically fraught when it is not.

None of these is obviously the right route, and the debate has become a test case for a methodological question the next article takes up directly. If knowledge resists analysis into belief plus further conditions, as sixty years of Gettier literature suggests, perhaps the search for the right conditions was the wrong search. Perhaps knowledge is the primitive, and the things we tried to build it from should be understood in its terms instead.

Next

The proposal that knowledge should be treated as fundamental rather than analysed, and Timothy Williamson's argument that no non-trivial mental condition is fully transparent to its subject.