Christoph Kelp

Assertion is the central vehicle for the sharing of knowledge. Whether knowledge is shared successfully often depends on the quality of assertions: good assertions lead to successful knowledge sharing, while bad ones don't. In Sharing Knowledge, Christoph Kelp and Mona Simion investigate the relation between knowledge sharing and assertion, and develop an account of what it is to assert well. More specifically, they argue that the function of assertion is to share knowledge with others. It is this function that supports a central norm of assertion according to which a good assertion is one that has the disposition to generate knowledge in others. The book uses this functionalist approach to motivate further norms of assertion on both the speaker and the hearer side and investigates ramifications of this view for other questions about assertion.

Two important philosophical questions about assertion concern its nature and normativity. This article defends the optimism about the constitutive norm account of assertion and sets out a constitutivity thesis that is much more modest than that proposed by Timothy Williamson. It starts by looking at the extant objections to Williamson’s Knowledge Account of Assertion and argues that they fail to hit their target in virtue of imposing implausible conditions on engaging in norm-constituted activities. Second, it makes a similar proposal and shows how it does better than the competition. It suggests that Knowledge Norm of Assertion is not constitutive of the speech act of assertion in the same way in which rules of games are constitutive, and thus KAA comes out as too strong. The final section embarks on a rescue mission on behalf of KAA; it puts forth a weaker, functionalist constitutivity thesis. On this view, KNA is etiologically constitutively associated with the speech act of assertion, in virtue of its function of generating knowledge in hearers.

Can we engineer conceptual change? While a positive answer to this question would be exciting news for philosophy, there has been a growing number of pessimistic voices in the literature. This paper resists this trend. Its central aim is to argue not only that conceptual engineering is possible but also that it is not even distinctively hard. In order to achieve this, we will develop a novel approach to conceptual engineering that has two key components. First, it proposes a reorientation of the conceptual engineering project away from fixing conceptual defects and towards bringing about conceptual innovation. Second, it offers a new account of when conceptual engineering is successful in terms of etiological functions. We then turn to the reasons that have motivated various forms of pessimism about conceptual engineering and show that, on our novel approach, none of them stands up to scrutiny.

Suppose you could change people’s way of thinking about the world at its very roots, by changing the very concepts by means of which they think. Suppose, further, that this would make the world a better place; that would be quite something. Conceptual engineering is concerned with this remarkable kind of feat. This book is a comprehensive and systematic study of the nature and normativity of conceptual engineering. The study is comprehensive in that it deals with all the central questions in the field: concerning the very possibility of conceptual engineering, its success conditions, the epistemic and moral permissibility of conceptual engineering, the nature of conceptual ignorance and conceptual understanding, duties to engage in conceptual engineering, as well as with political aspects of conceptual engineering, such as engineering social kind concepts, the nature of hermeneutical injustice, and conceptual disinformation. The study is systematic in that it answers these questions in a unified manner, whereby the accounts put forth centre on conceptual functions. We develop a functionalist approach to conceptual engineering that is wholeheartedly optimistic, in that it takes conceptual engineering to be something that we can and should do. In fact, it is argued, it is not even distinctively difficult to do it.

In ‘Trustworthy Artificial Intelligence’, we develop a novel account of how it is that AI can be trustworthy and what it takes for an AI to be trustworthy. In this paper, we respond to a suite of recent comments on this account, due to J. Adam Carter, Dong-yong Choi, Rune Nyrup, and Fei Song. We would like to thank all four for their thoughtful engagement with our work, as well as the Asian Journal of Philosophy for publishing the symposium on our paper. The game plan for the paper is as follows. We will first briefly rehearse the account and then respond to comments in turn.

This paper develops an account of trustworthy AI. Its central idea is that whether AIs are trustworthy is a matter of whether they live up to their function-based obligations. We argue that this account serves to advance the literature in a couple of important ways. First, it serves to provide a rationale for why a range of properties that are widely assumed in the scientific literature, as well as in policy, to be required of trustworthy AI, such as safety, justice, and explainability, are properties (often) instantiated by trustworthy AI. Second, we connect the discussion on trustworthy AI in policy, industry, and the sciences with the philosophical discussion of trustworthiness. We argue that extant accounts of trustworthiness in the philosophy literature cannot make proper sense of trustworthy AI and that our account compares favourably with its competitors on this front.

According to the achievement account (AA) of the value of knowledge, knowledge is finally valuable because it is a species of a finally valuable genus, achievement. The achievement account is said to solve Pritchard's tertiary value problem (TVP), the problem of showing that knowledge enjoys a different kind of value than mere true belief. This paper argues, first, that AA fails to solve TVP, and, second, that Pritchard's motivations for TVP are inadequate. They do, however, motivate a weaker value problem, one that can be solved by showing that knowledge is weakly superior to mere true belief. Finally, third, we argue that knowledge does enjoy this form of weak superiority. Since this argument rests on AA's key species-genus claim, AA can retain the upper hand in the debate.

The phenomenon of defeat is hot in epistemology. However, surprisingly little attention has been paid to defeat in the semantics of aesthetic discourse and aesthetic epistemology. We think that this is a lack that needs supplying. Here, we argue for a conditional claim: if epistemic defeat about aesthetic matters—what we will call, for convenience, aesthetic defeat—exists, this gives us (pro tanto) reason to worry about several views in the semantics of aesthetic discourse—to wit, contextualism and relativism—and one major player in the epistemology of aesthetic testimony—i.e. pessimism about the capacity of aesthetic testimony to generate knowledge. None of these can straightforwardly accommodate aesthetic defeat. The alternative is to endorse scepticism about aesthetic defeat; however, there is reason to believe the latter move is highly implausible. We conclude that the theorist of aesthetic discourse is faced with a choice between either being a realist about the semantics and an optimist about the epistemology, or coming up with a thorough defence of defeat scepticism.

In ‘Trustworthy Artificial Intelligence’, we develop a novel account of how it is that AI can be trustworthy and what it takes for an AI to be trustworthy. In this paper, we respond to a suite of recent comments on this account, due to J. Adam Carter, Dong-yong Choi, Rune Nyrup, and Fei Song. We would like to thank all four for their thoughtful engagement with our work, as well as the Asian Journal of Philosophy for publishing the symposium on our paper. The game plan for the paper is as follows. We will first briefly rehearse the account and then respond to comments in turn.

This paper develops an account of trustworthy AI. Its central idea is that whether AIs are trustworthy is a matter of whether they live up to their function-based obligations. We argue that this account serves to advance the literature in a couple of important ways. First, it serves to provide a rationale for why a range of properties that are widely assumed in the scientific literature, as well as in policy, to be required of trustworthy AI, such as safety, justice, and explainability, are properties (often) instantiated by trustworthy AI. Second, we connect the discussion on trustworthy AI in policy, industry, and the sciences with the philosophical discussion of trustworthiness. We argue that extant accounts of trustworthiness in the philosophy literature cannot make proper sense of trustworthy AI and that our account compares favourably with its competitors on this front.

This paper develops a novel, bifocal account of trustworthiness according to which both trustworthinesssimpliciter(as in ‘Ann is trustworthy’) and trustworthiness tophi(as in ‘Ann is trustworthy when it comes to keeping your secrets’) are analysed in terms of dispositions to fulfil one's obligations. We also offer a systematic account of the relation between the two types of trustworthiness, an account of degrees of trustworthiness and comparative trustworthiness, as well as a view of permissible trustworthiness attribution.

While fallibilism has been the dominant view in epistemology in recent times, the field has witnessed the rise of a new form of infallibilism. In a recent book, Jessica Brown has taken on the task of mounting a systematic defence of fallibilism against this new infallibilism. She argues that new infallibilism incurs several problematic commitments that fallibilism can avoid. In addition, the key data points that infallibilists have adduced in support of their view can be accommodated by fallibilism, giving fallibilism the upper hand. This paper develops a rejoinder on behalf of new infallibilism. It explores ways in which new infallibilists can avoid the problematic commitments Brown identifies and provides reason to think that some of the data points supporting new infallibilism cannot be as readily handled by fallibilism as Brown would have us think.

This paper highlights some connections between work on truth approximation and work in social epistemology, in particular work on peer disagreement. In some of the literature on truth approximation, questions have been addressed concerning the efficiency of research strategies for approximating the truth. So far, social aspects of research strategies have not received any attention in this context. Recent findings in the field of opinion dynamics suggest that this is a mistake. How scientists exchange and take into account information about each others’ beliefs may greatly influence the accuracy and speed with which the scientific community as a whole approximates the truth. On the other hand, social epistemologists concerned with peer disagreement have so far neglected the question of how practices of responding to disagreements with peers fare with respect to the goal of approximating the truth. Again, work on opinion dynamics shows that this may be a mistake, and that how we ought to respond to disagreements with our peers may depend on the specific purposes of our investigations.

Traditionally, epistemologists have held that only truth-related factors matter in the question of whether a subject can be said to know a proposition. Various philosophers have recently departed from this doctrine by claiming that the answer to this question also depends on practical concerns. They take this move to be warranted by the fact that people’s knowledge attributions appear sensitive to contextual variation, in particular variation due to differing stakes. This paper proposes an alternative explanation of the aforementioned fact, one that allows us to stick to the orthodoxy. The alternative applies the conceptual spaces approach to the concept of knowledge. With knowledge conceived of spatially, the variability in knowledge attributions follows from recent work on identity, according to which our standards for judging things (including concepts) to be identical are context-dependent. On the proposal to be made, it depends on what is at stake in a context whether it is worth distinguishing between knowing and being at least close to knowing.

According to a much discussed argument, reliabilism is defective for making knowledge too easy to come by. In a recent paper, Weisberg aims to show that this argument relies on a type of reasoning that is rejectable on independent grounds. We argue that the blanket rejection that Weisberg recommends of this type of reasoning is both unwarranted and unwelcome. Drawing on an older discussion in the philosophy of science, we show that placing some relatively modest restrictions on the said type of reasoning suffices to block the anti-reliabilist argument.