Jumbly Grindrod

In this paper, I outline and investigate the notion of computational beliefs: beliefs formed on the basis of a deliverance from a machine learning algorithm. Given the increased usage of such algorithms through smart devices, such beliefs are becoming increasingly common in everyday life. First, I argue that such beliefs can be successful by outlining particular examples that possess epistemic properties taken to be indicative of successful beliefs. I then outline how computational beliefs are best understood as a form of what Sosa describes as instrumental beliefs. As such, it may be thought that computational beliefs hold little epistemological interest insofar as they are reducible to more basic epistemic processes that are already the focus of epistemological attention. However, in the final section, I argue that computational beliefs hold epistemological importance insofar as they have specific epistemically normative repercussions i.e. they give rise to epistemic responsibility gaps.

In this paper, I will consider the repercussions that epistemic contextualism has on capturing the distinctive value of knowledge. I will argue that the way that contextualist views capture the value of knowledge depends on the depth of the contextualism involved. To do so, I distinguish between superficial and deep contextualism, and I show how the latter is forced to contextualist epistemic value in a way the former is not. However, I then argue that if the superficial contextualist view does not contextualise epistemic value, it would nevertheless fail to properly capture the value of knowledge. If epistemic contextualism is true, then epistemic value should be contextualised.

The ‘question under discussion’ (or ‘QUD’) framework is a pragmatic framework that draws on work in the semantics of questions to provide an appealing account of a range of pragmatic phenomena, including the use of prosodic focus in English and restrictions on acceptable discourse moves (Roberts 1996). More recently, however, a number of proposals have attempted to use the framework to help to settle issues at the semantics/pragmatics boundary, fixing the truth-conditions of what is said by a speaker (which many theorists take to be a semantic matter). In this discussion piece, we suggest that this kind of putative extension of the work to be done by the QUD framework is illegitimate, as the framework ultimately seems to depend on a prior grip on semantic content. To see this, we first outline the QUD framework and then raise our concern.

This thesis is concerned with the linguistic plausibility of epistemic contextualism. Epistemic contextualism can be (roughly) characterised as the view that the truth conditions of knowledge attributions are sensitive to the context of utterance. As such, it is a linguistic claim that is usually defended on the basis of certain context-shifting experiments and is then usually integrated into a semantic theory that captures this context-sensitivity. The linguistic challenge for epistemic contextualism is to be successfully integrated into our best account of linguistic communication. I will argue that as of yet there is no satisfactory solution to the linguistic challenge and that this should lead us to re-envision the view at its most basic level. First, I analyse the intuitive basis of epistemic contextualism. The contextualist claims regarding ordinary linguistic intuitions in the context-shifting experiments have recently been challenged experimentally. In response to this, I present new experimental evidence for a contextual effect on our linguistic intuitions using context-shifting experiments with third-person knowledge attributions. In doing so, I defend the intuitive basis of epistemic contextualism. I then analyse the plausibility of capturing this contextual effect in a range of linguistic proposals. I argue that we do not have sufficient reason to reject an implicature-based explanation of the phenomenon. I do, however, present defeasible evidence against such an approach. I then analyse the most plausible semantic contextualist account in the literature – Schaffer & Szabó’s epistemic comparativism. However, I find their proposal to be unmotivated and problematic. I then analyse the plausibility of capturing the context-sensitivity of knowledge attributions via a pragmatic enrichment account of what is said. I do so by analysing Schoubye & Stokke’s minimalist account of what is said and Peter Ludlow’s dynamic lexicon view. However, I find both approaches to be problematic. With this in mind, epistemic contextualism finds itself at something of an impasse: there is experimental evidence for a contextual effect on our linguistic intuitions regarding knowledge attributions but as of yet there is no good explanation of this effect available. Rather than using this as a point at which to defend a new semantic or pragmatic proposal, I instead raise a fundamental concern regarding the contextualist project. I argue that there is a fundamental lack of clarity as to what the contextualist aims to achieve. I distinguish between two possible approaches to epistemic contextualism that differ in their aims – one more ambitious and one more modest. Once this distinction is drawn, I argue that there is a severe explanatory lacuna in the ambitious approach, and no promising way of filling it. Given this, the epistemic contextualist should re-envision some of their basic commitments along more modest lines. This proves crucial regarding the linguistic challenge, as modest contextualism is able to bypass the linguistic challenge altogether by remaining neutral on a range of particular linguistic issues. Modest epistemic contextualism is defended in the final chapter.

In the past few years there has been a turn towards evaluating the empirical foundation of epistemic contextualism using formal (rather than armchair) experimental methods. By-and-large, the results of these experiments have not supported the original motivation for epistemic contextualism. That is partly because experiments have only uncovered effects of changing context on knowledge ascriptions in limited experimental circumstances (when contrast is present, for example), and partly because existing experiments have not been designed to distinguish between contextualism and one of its main competing theories, subject-sensitive invariantism. In this paper, we discuss how a particular, "third-person", experimental design is needed to provide evidence that would support contextualism over subject-sensitive invariantism. In spite of the theoretical significance of third-person knowledge ascriptions for debates surrounding contextualism, no formal experiments evaluating such ascriptions that assess contextualist claims have previously been conducted. In this paper, we conduct an experiment specifically designed to examine that central gap in contextualism’s empirical foundation. The results of our experiment provide crucial support for epistemic contextualism over subject-sensitive invariantism.