CAPEレクチャー:Kevin Kelly講演会

日時：2013年12月26日（木）16:30～18:00
場所：京都大学文学部校舎１階会議室

Abstract:  The question “what is knowable?” serves as the point of entry into the systems of Plato, Descartes, Locke, Hume, and Kant.  Traditionally, the focus was on the scope and prospects of scientific inquiry.  For example, Kant held that such apparently empirical questions such as “the universe had a beginning in time” and “matter is infinitely divisible” are un-knowable because they out-run all possible experience.  Contemporary possible worlds models for “s knows that p” merely assign an unanalyzed, static knowledge state to the agent, so they cannot begin to model Kant’s concerns.  Instead, the focus has shifted to arcane examples of un-knowability based on self-reference, such as the Moore sentence: “p and I do not know that p”.  In this talk, I propose a new, learning semantics for knowledge in which the knowing agent is modeled explicitly as a learner who modifies her sentential beliefs according to a concrete, computable learning method that receives raw, non-linguistic inputs from an unstructured world-in-itself.  Kant is vindicated in the sense that  the answers to his questions are un-learnable.  But learnability is not closed under logical consequence.  Insistence upon deductive closure for knowability motivates a concept of knowing weaker than having learned, in the sense that one would merely have to avoid error eventually, rather than find the truth eventually, if p were false.  Then Kant’s examples are knowable but not learnable, so Kant was too pessimistic.  The distinction between learnability and knowability connects the deductive closure of knowability with traditional philosophical discussions of serendipity, Duhem’s problem, and the logic of discovery.  Finally, the proposed semantics explains how you can know your own Moore sentence (although you wouldn’t want to) so standard epistemic logic is unsound even on its own examples.