Knowledge-First Epistemology and Anti-Luminosity

For most of the twentieth century, epistemology had a fixed order of explanation. Belief came first, as a mental state. Truth was a matter of how the world is. Justification was a property of beliefs, explicable in terms of evidence, reasons, or reliable processes. Knowledge was the complex thing built from these: a belief that is true, justified, and whatever fourth condition fixes Gettier. The project was to find the recipe.

Timothy Williamson's Knowledge and Its Limits, published in 2000, proposed reversing the order. Knowledge is not a compound of belief and other conditions. It is a basic mental state in its own right, and the other epistemic notions, including evidence and justification, should be explained in terms of it. The book has been the most discussed work in epistemology since, and its programme is now called knowledge-first.

Why stop analysing

The first argument is inductive. Since 1963, philosophers have proposed dozens of fourth conditions, and each has met counterexamples. The intermediate tier followed several of these through: no false lemmas, defeasibility, causal theories, reliabilism, safety, sensitivity. Each fixes some cases and breaks others. Williamson's suggestion is that the pattern is not an accident. Most interesting concepts resist analysis into necessary and sufficient conditions; attempts to analyse causation, meaning, or personhood have the same history. The persistence of failure is evidence that there is no analysis to find.

This does not mean nothing can be said about knowledge. Williamson offers a characterisation: knowledge is the most general factive mental state. But this is not a reductive analysis. It tells you where knowledge sits among other states without building it from them.

The second argument concerns the metaphysics. If knowledge were belief plus some external conditions, then it would be a hybrid: part mental, part a matter of how the world happens to be. Williamson denies this. Knowing is as much a mental state as believing is.

Factive mental states

Language marks a family of attitudes that entail their content. You can believe that it is raining when it is not, but you cannot see that it is raining when it is not; at most you can seem to see it. You can think you remember locking the door, but if you did not lock it, you do not remember locking it. Williamson calls the verbs that express these attitudes factive stative attitude verbs, and argues that all of them entail knowing. If you see that it is raining, you know it is raining. If you regret that you missed the train, you know you missed it. Knowledge is the most general state in the family, the one entailed by every other member.

The traditional picture had to say that seeing that p is a composite of something inner, a visual experience, and something outer, the fact that p. Williamson's objection is that nothing forces this decomposition. The fact that a state has conditions involving the external world does not make it less mental; the content of a belief also depends on the external world, as Putnam's Twin Earth cases showed, and no one concludes that belief is not a mental state.

One consequence is that a subject who knows and a subject who is deceived, even if they are exactly alike internally, are in different mental states. Internalists will find this unacceptable. The next article turns to a view, epistemological disjunctivism, that develops the same idea from a different starting point.

Evidence is knowledge

The thesis looks surprising at first, but Williamson argues it follows from how evidence functions.

Evidence is used in inference to the best explanation, in probabilistic reasoning, and in ruling out hypotheses. In each role, the evidence must be propositional: we explain that the footprints are there, we conditionalise on the proposition that the test came back positive, we rule out hypotheses inconsistent with what we observed. Evidence must also be true, or at least it seems perverse to count a false proposition among your evidence. If you believe the defendant was in Leeds on the night in question, and he was not, then the proposition that he was in Leeds is not part of your evidence, though your belief that he was might be.

If evidence is propositional and true, the question is which true propositions count. Not every true proposition is part of your evidence, only those you are in some way in touch with. Williamson's claim is that knowledge is the relevant form of contact. A true proposition you merely believe, perhaps for bad reasons, should not be something you can treat as settled in your reasoning; a true proposition you know can be.

The thesis has sharp consequences. It implies that two subjects who are internally alike can have different evidence: the ordinary perceiver who knows there is a table and the brain in a vat who does not. This is exactly what internalists deny, and the next article considers the new evil demon problem that expresses their resistance. It also implies that one can fail to know what one's evidence is, since one can fail to know what one knows. That leads to the anti-luminosity argument.

The knowledge norm of assertion

Williamson argues that assertion is a speech act governed by a single constitutive rule, in the way that a game is defined by its rules, and that the rule is: assert only what you know.

The evidence is conversational. When someone asserts something, it is appropriate to challenge them with "How do you know?" or "Do you know that?", which presuppose that they should know. It is not appropriate to challenge with "Do you believe that?" or "Is that justified?" in the same way.

Moore's paradox provides a second strand. The sentence "It is raining, but I don't know that it is raining" is not a contradiction, since it can be true. But it is absurd to assert. The knowledge norm explains why: in asserting the first conjunct you represent yourself as knowing it, and in asserting the second you deny that you know it.

Lottery assertions provide a third. You hold one ticket in a fair lottery with a million tickets. The chance your ticket loses is 0.999999. Still, it seems improper to assert "Your ticket has lost" before the draw, and if someone says it and turns out to be right, we think they spoke out of turn. A norm based on high probability or justified belief cannot explain the impropriety, since the probability is as high as you like. The knowledge norm can: you do not know your ticket has lost, however probable it is. Lottery propositions will return later in the tier, in the article on the lottery and preface paradoxes, where they create trouble for any view that equates belief with high confidence.

Rival norms have been proposed. Jennifer Lackey argues that reasonable belief suffices, pointing to cases in which it seems permissible to assert what one does not know: a doctor telling a patient a diagnosis based on strong but not decisive evidence. Matthew Weiner defends a truth norm. John Turri has run experiments suggesting that people evaluate assertions by the knowledge standard. The debate continues, but the knowledge norm remains the best-supported single rule.

The anti-luminosity argument

Traditional internalism holds that the conditions that matter for justification are accessible to the subject by reflection. The intermediate tier called this access internalism. Some internalists went further: certain mental conditions, such as being in pain or seeming to see red, are transparent. If you are in them, you can always know you are in them.

Williamson's argument targets this assumption. It uses a slow process in which a condition gradually ceases to obtain.

You wake up at dawn feeling freezing. Over the course of the morning you warm up gradually, until by noon you feel hot. Take a series of moments, $t_0, t_1, \ldots, t_n$, each a millisecond apart. At $t_0$ you feel cold; at $t_n$ you do not. Between adjacent moments, the difference in how you feel is imperceptibly small.

The key premise is the second one. It is the safety condition from the intermediate tier: if you know, you could not easily have been wrong. Your confidence that you feel cold at $t_i$ is almost exactly what it will be at $t_{i+1}$, since the change in how you feel is imperceptible. If at $t_{i+1}$ you did not feel cold, then at $t_i$ your belief would be one you could very easily have held falsely. So it would not be knowledge.

The argument generalises to any condition that can obtain to varying degrees and fade gradually. It does not show that we never know our own mental states. It shows only that there are cases near the boundary where the condition obtains and we are not in a position to know it. Every non-trivial condition has such cases.

What it costs internalism

Access internalism said that justification depends only on conditions to which the subject has reflective access, so that in principle one can always tell whether one is justified. The anti-luminosity argument shows that no condition meets this standard, since whatever condition justification is supposed to depend on, there will be borderline cases where it obtains undetectably.

The internalist has several replies. One is to deny the margin-for-error premise, arguing that for certain states, especially phenomenal states, the belief is constituted by the experience itself and so cannot be unsafe. Another is to concede that no condition is luminous but hold that internalism never needed luminosity, only that justification supervene on internal states. The second reply abandons the motivation that made internalism attractive, which was that one could always tell whether one was being rational.

Objections to knowledge-first

A second objection concerns the evidence thesis. If evidence is knowledge, then a subject who is systematically deceived has less evidence than an ordinary perceiver, even though they are internally identical. But it seems that the deceived subject is just as rational as the ordinary perceiver. If the demon's victim believes there is a table on the basis of a table-like experience, they seem to be doing exactly what they ought to do. Knowledge-first seems to have trouble saying so.

Williamson's response is to distinguish being justified from being blameless. The demon victim's beliefs are blameless, even excusable, but not justified, since they are not appropriately related to knowledge. Critics find this unsatisfying: the victim is not merely blameless but reasoning well, and their evidence is exactly as good as ours from their point of view. That objection is the new evil demon problem, and it is the subject of the next article.

Next

The new evil demon problem as a challenge to externalism, and epistemological disjunctivism, which claims that perceptual reasons can be both factive and reflectively accessible.