Conclusion

The hangman/surprise-examination/prediction paradox is solved. It is not solved by denying knowledge closure (although knowledge closure is false). It is not solved by denying KK or denying that knowing p implies other iterated knowing attitudes (although these are false). It is not solved by misleading evidence causing the students to lose knowledge because students cannot lose knowledge this way. It is solved by showing that a tacit assumption (what is being said to the students/prisoner is informative) is overlooked and that inferences by contradiction are invalid if assumptions are left out. The phenomenology of the surprise-exam paradox is explored to explain why this solution has been missed. Crucial is that in many cases the students/prisoner know(s) there will be a surprise exam/execution because of an inference from what the teacher/judge meant to say, and not directly by the literal application of what he did say.

The (Complex) Structure of the Meaning of “Know(s)”

Why “know(s)” cant be defined is explained: it is not because of Gettier puzzles but because of the failure of parity reasoning. The latter shows that there is a case where S knows p and S does not know q, but where S has as much of a reason to believe p as he does to believe q. Whether “know(s)” should be retooled to have different semantic properties is explored. It is shown that the word needs to be criterion-transcendent, factive, fallible, and vague in order to be a word we can use at all, and in particular to be useful for proposition transfers between agents or between oneself at earlier times and later times, where justification is not transferred. The social roles of “know(s)” are explored next. The word has many roles that correspond to its syntactic and semantic properties.

Usage Traps in the Language of Iterated Knowledge Attributions

How we easily slip between metacognitive thought and assertion and ground-floor thought and assertion is illustrated; how, as a result, we easily confuse the two is also illustrated. “Do you know the time?” is often speaker-meant merely as a request for information as opposed to what it literally conveys, a question about the auditor’s knowledge state. Distinctions between having concepts, grasping one’s own concepts, and metacognizing one’s propositional attitudes (in various ways) are distinguished. Why it is so easy to confuse being aware of being in pain and being in pain is explained; that it seems it isn’t possible to be in pain without being aware of it illustrates metacognitive confusions. Similarly, kinds of justifications are distinguished that are often confused, ones that involve metacognition and ones that don’t. How “level confusions” bedevil philosophical arguments is illustrated.

Knowledge Attributions to Minimal Epistemic Agents

Widespread usage data for “know,” “learn,” “see,” “remember,” and other epistemic words are given. Such words are routinely and literally applied to animals ranging from sophisticated elephants and orcas to insects like ants and honeybees to nonconscious mechanisms like driverless cars, drones, and thermostats. Further, “S knows p” places no constraints on the agent S vis-à-vis the concepts exhibited in p: Rover need not have the concept of “cat” to know that a cat is trapped above him in a tree. How this data shows that a knowing agent need not know much, need not be self-conscious of what she knows, and need not be conscious or capable of metacognition is described. The modularity of agential knowledge is characterized: two agents may have the same sensory evidence for p, and yet one can know p while the other doesn’t because of other aspects of their methods for establishing p.

Usage Challenges to Fallibilism

A definition of fallibility shows that agents are fallible about necessary truths. It is shown that fallibility of agents implies a denial of parity reasoning. Moorean paradoxes appear to undercut fallibility, but they are due entirely to the factivity of “know.” Kripke’s dogmatism paradox is explained: the key is recognizing that knowledge fallibility applies to the knowledge that all evidence against something one knows is misleading. That we do not know we will lose a lottery is denied. Fallibility shows this. And that people argue over this also indicates this. Knowledge closure fails because of fallibility; so does aggregation of assumptions. Vagueness shows why debates about whether we know outcomes of lotteries before winning tickets are drawn are irresolvable. Irrational penny reasoning is analyzed; it applies to nonfactive attitudes such as being really really sure. Preface paradoxes are explained. That it is sometimes rational to believe contradictory propositions is explained.

Confidence, Belief, and Knowledge; The Vagueness of “Know(s)”

Knowledge does not require confidence. An agent may know without confidence because of misleading evidence or for other reasons. An agent may not believe what she knows. Misleading evidence never causes agents to lose knowledge. The vagueness of an expression may be visible to speakers or invisible. In the case of “bald,” it is visible; it is not visible for “know.” This is because knowledge standards are invisible. Vagueness is analyzed as being epistemic in the sense that our ignorance of whether a word applies in a case places no metaphysical constraints on the facts. Agential standards for evidence are also tri-scoped and application-indeterminate. There are cases where such standards determine no answer, knows or not; and there are cases where it is indeterminate whether, or not, standards determine an answer. Because Timothy Williamson’s argument against KK presupposes that knowledge requires confidence, his argument fails.

Iterated and Ground-Floor Cognition, KK and K¬K Arguments and Empirical Studies

A distinction between Cartesian knowers (who are capable of all forms of metacognition) and ground-floor cognizers are drawn. B.B., a virtual ground-floor cognizer, is extensively described: what it knows, what it doesn’t know, and what concepts can be attributed to it. The fragmented nature of iterated cognitions is described. That deduction need not require metacognitions of any sort is described: in successfully deducing q from p, an agent need not recognize or appreciate that she is using propositions, that she is using a rule (modus ponens), or that she is justified. A psychological study of deduction is described, and how it fails to illustrate metacognition is illustrated. The apparent ineffability of metacognition in nonhuman animals is discussed. A single anecdotal case of metacognition in chimpanzees is given, and an implicit knowledge generalization is attributed to the animals on the basis of this case. The use of Morgan’s canon is rejected.

The Variability of Know(s)-that Judgments

The usage evidence—various scenarios that realistically depict where and when we attribute knowledge to ourselves and others—shows that all the alternatives (epistemic contextualism, subject-sensitive invariantism, knowledge relativism) to intellectual invariantism fail. They fail for several reasons: When cases are compared, speaker-hearers tend to retract one or the other conflicting knowledge claim; the intuitions elicited by various cases don’t consistently satisfy any particular position; the situations under which speaker-hearers retract knowledge claims under pressure seem to support an invariantist position. Nevertheless, no standard invariantist position seems supported by the usage data because speaker-hearers do seem to shift because of differences either in the interests of the agents to whom knowledge is attributed, for example, oneself, or because of other apparently non-epistemic reasons. Attempts to use pragmatic tools, such as implicatures, to handle the apparent shifts in knowledge standards are shown to fail as well.

Knowledge and Knowing that p; “Knowledge” and “Knowing that p”

The usage-fact differences between uses of “S knows p,” and “knowledge,” a mass noun, are described. “Knowledge” isn’t factive (a lot of a person’s knowledge can be wrong), and it is contextually sensitive in just the way that mass terms usually are. The knowledge someone (a child) has can be a lot whereas the same knowledge of someone else (an adult) is not a lot. A lot of the knowledge that someone has can be wrong and yet still be knowledge. “Knowledge” is thus open to grading in a way that “know” isn’t. Peter can know more than Sam. This is a use of a verb form derived from the noun “knowledge.” But this is not true when Sam knows p. Sam cannot know p better than Peter knows p. If “know” is contextually sensitive, it’s shown that the word isn’t so in a way like any other contextually sensitive verb.

Assertion Norms

Assertion is a phenomenological category—that is, assertions are experienced as such by speaker-hearers. Speech-act phenomenology is distinguished from semantic perception. We not only experience speech acts, we experience the words and sentences we utter as distinct objects with properties different from those of the speech acts. Using this distinction, evidence against agential-state assertion norms, such as a sincere-belief norm, a knowledge norm, or a warrant norm, etc., is given. Anonymous assertions or shapes resembling inscriptions produced by accident are experienced as assertions and as possessing meaning even when they are recognized to be products of sheer accidents and in reality without utterers. Spokespersons for companies, actors in advertisements for products, cartoon characters (that don’t exist), and flakes who can’t be trusted are all experienced nevertheless as asserting, and what they assert as assertions. The common-ground expectation view is supported. Compatibly with this, Moorean remarks are often naturally utterable.