Brian Ball

In recent years, there have been growing calls to operationalize artificial intelligence (AI) ethics - to move from theory to practice, or (as one group of authors has put it) ‘from what to how’ (Morley et al. 2020). In this paper, we propose a novel account of what ethical AI practice might look like, which we call the 3D model, named for its recognition, within the overall AI design cycle, of the three stages of design, development, and deployment. This model aims to embed ethics throughout this cycle, offering questions that should be addressed at each stage. We articulate the benefits of this approach to ethical AI practice: that it is pro-ethical and value-aware, amenable to implementation, it embeds ethics at every stage of the development process, it embeds a culture and language of ethics in organizations and provides clear decision points. Our model is not a panacea, of course, and we accordingly provide an indication of the context in which the implementation of our model might be most effective in ensuring ethical AI practice.

This paper explores what computational methodologies can tell us about philosophical education, particularly in the context of artificial intelligence (AI) ethics. Taking the readings on our AI ethics and responsible AI syllabi as a corpus of AI ethics literature, we conduct an analysis of the content of these courses through a variety of methods: word frequency analysis, term frequency–inverse document frequency (TF–IDF) scoring, document vectorization via SciBERT, clustering via k-means, and topic modelling using latent Dirichlet allocation (LDA). We reflect on the findings of these analyses, and more broadly on what computational approaches can offer to the practice of philosophical education. Finally, we compare our approach to previous computational approaches in philosophy, and more broadly in the digital humanities. This project offers a proof of concept for how contemporary natural-language processing techniques can be used to support philosophical pedagogy: not only to reflect critically on what we teach, but to discover new materials, explore conceptual gaps, and make our courses more accessible to students from a range of disciplinary backgrounds.

Embodied semi/autonomous systems (e.g., robots) have tremendous potential to improve the human experience, especially if they can be developed from tools and/or semi/autonomous interactive agents to become true collaborative teammates. But what exactly is required to achieve such autonomy in human-robot teams? Our interdisciplinary investigation in this paper is both empirical and conceptual. We argue that autonomous teams require an interdependence between teammates sharing an ultimate goal, and able to flexibly adopt and embrace intermediate goals in a manner that (empirically) requires both (social) intelligence (aka theory of mind) and (a limited form of) authenticity. This is (conceptually) compatible with robotic team members being autonomous only in an attenuated sense that bears a certain family resemblance to autonomy in humans.

The two volumes on Wittgenstein and AI aim to trace and suggest Wittgensteinian influences in some of the most cutting-edge areas of research in Artificial Intelligence (such as Computation, Machine Learning, Natural Language Processing and the use of automation in legal settings). The collection is driven by an essentially interdisciplinary approach, featuring articles from philosophers, computer scientists and legal scholars, writing on a broad range of topics in AI. The chapters across these two volumes are grouped into two sets of themes: Mind and Language and Value and Governance. These themes correspond to two major areas of research in the philosophical study of AI: the computational mind and the ethics of artificial intelligence. These volumes make a significant and unprecedented contribution to the question of what Wittgenstein’s philosophy can offer to the ever-growing field of AI. It aims to provide insight for both philosophers and non-philosophers alike, offering reflection on the significance of Wittgenstein’s work for AI, and on the implications of advancements in AI technology for Wittgenstein’s philosophy and philosophy influenced by Wittgenstein. VOLUME II: Wittgenstein and AI (Volume II): Value and Governance. This volume includes chapters on ethical AI, rules in AI, rules and the law, human-AI interaction, the moral implications of robotics and the status of AI art.

In this thesis, Semantics, Meta-Semantics, and Ontology, I provide a critique of the method of truth in metaphysics. Davidson has suggested that we can determine the metaphysical nature and structure of reality through semantic investigations. By contrast, I argue that it is not semantics, but meta-semantics, which reveals the metaphysically necessary and sufficient truth conditions of our claims. As a consequence I reject the Quinean criterion of ontological commitment. In Part I, chapter 1, I argue that the metaphysically primary truth bearers are not propositions, but rather concrete representations, either beliefs or sentences. I show, in chapter 2, that we can give sense to a truth predicate applying to sentences, given a truth operator and quantification into sentence position. I argue that this strategy does not commit us to the existence of propositions serving as truth bearers. In Part II I argue that although we must assign semantic values to sentences and/or predicates, the meaningfulness of these expressions is not thereby explained. In chapter 3 I articulate Davidson’s problem of predication and his solution, but argue that he was wrong to attribute this solution to Tarski. In chapter 4 I examine the semantics of modal languages; I conclude that although they require semantic values for predicates and/or sentences we should be instrumentalists about these theories. In Part III I consider the relationship between truth and existence. In chapter 5, I defend Pluralism about truth: in some domains of discourse, I claim, semantic reference plays a merely instrumental role in explaining truth. In chapter 6, I show that Hume’s Principle, which is committed by the Quinean criterion to the existence of numbers, can be true even though numbers do not exist. In doing so, I appeal to meta-semantic and diachronic considerations. In the conclusion I compare my views on ontology and commitment to Jody Azzouni’s; and in the appendix I suggest how one might pursue diachronic linguistics.

The three main approaches to the metaphysics of intentionality can arguably be subjected to analysis in terms of grammatical point of view: the approach of the (internalist) phenomenal intentionality programme (plus productivism about linguistic content) may be regarded as first-personal; interpretationism, perhaps, as second-personal; and (reductive externalist) causal information theories (including teleosemantics) as third-personal. After making this plausible, the current paper focusses on the role of the interpreter (if any) in interpretationism. It argues that, despite some considerations from the publicity of meaning potentially suggesting the contrary, radical interpretation is not subject to epistemic constraint; nor should the interpretationist appeal to the idiosyncratic interests of actual interpreters, thereby rendering the approach irremediably relativistic. Instead, an appeal to the pure form of interestedness is all that is involved; this supports a methodologically non-reductive outlook on intentionality.

The central thesis of this paper is that fake news and related phenomena serve as defeaters for knowledge transmission via journalistic channels. This explains how they pose a threat to democracy; and it points the way to determining how to address this threat. Democracy is both intrinsically and instrumentally good provided the electorate has knowledge (however partial and distributed) of the common good and the means of achieving it. Since journalism provides such knowledge, those who value democracy have a reason to protect it. Hostile agents, however, can undermine both the effectiveness of democratic decision-making and faith in democracy itself, by deliberately promulgating fake news and hyper-partisan views; moreover, these effects can come about unintentionally on social media. I conclude that we may need to change, not just the way we process information online, but also the informational environment in which we operate.

In this paper I propose that game-playing situations may be regarded as involving agents standing in the playing relation to abstract games with essential rules; and I inquire whether linguistic activity can be so regarded as well. Timothy Williamson (2000) has argued that it can, and that assertion in particular should be thought of as a move in such a game, governed by the rule assert only what you know; but Ishani Maitra (2011) argues against the possibility of such an analysis on the basis of what I take to be a refinement of the thought that we follow no such strong, truth-entailing rule. I defend the Williamsonian proposal by rejecting accounts of rule-following that would allow the argument to succeed. Arguably, playing a game does require that agents follow its rules; but agents do, in the relevant (weaker) sense, follow a strong rule of assertion. And even if this account of rule-following should turn out to be mistaken, I suggest that it is possible to play a game while flagrantly violating its rules. In section 1, I introduce the game playing analysis, Williamson’s application of it to linguistic activity through his knowledge rule for assertion, and what I call ‘The Rule-Following Argument’ against his view. In section 2, I consider two analyses of rule-following, one of which (from intention, not conformity) might be taken to vindicate that argument; but in section 3, I suggest that the adoption of a third (sensitivity) analysis, or an alternative (rule-governance) account of game play, suffice to undermine it. In section 4, I respond to an objection; and in section 5, I engage directly with Maitra’s concerns about flagrant rule-violations. I briefly conclude, in section 6, that, despite arguments to the contrary, linguistic activity can be regarded as involving game play.

There is a deluge of AI-assisted decision-making systems, where our data serve as proxy to our actions, suggested by AI. The closer we investigate our data (raw input, or their learned representations, or the suggested actions), we begin to discover “bugs”. Outside of their test, controlled environments, AI systems may encounter situations investigated primarily by those in other disciplines, but experts in those fields are typically excluded from the design process and are only invited to attest to the ethical features of the resulting system or to comment on demonstrations of intelligence and aspects of craftmanship after the fact. This communicative impasse must be overcome. Our idea is that philosophical and engineering considerations interact and can be fruitfully combined in the AI design process from the very beginning. We embody this idea in the role of a philosopher engineer. We discuss the role of philosopher engineers in the three main design stages of an AI system: deployment management (what is the system’s intended use, in what environment?); objective setting (what should the system be trained to do, and how?); and training (what model should be used, and why?). We then exemplify the need for philosopher engineers with an illustrative example, investigating how the future decisions of an AI-based hiring system can be fairer than those contained in the biased input data on which it is trained; and we briefly sketch the kind of interdisciplinary education that we envision will help to bring about better AI.

What role, if any, should centered possible worlds play in characterizing the attitudes? Lewis :513–543, 1979) argued that, in order to account for the phenomena of self-location :474–497, 1977, Noûs 13:3–21, 1979), the contents of the attitudes should be taken to be centered propositions. Stalnaker Assertion: New philosophical essays, Oxford University Press, Oxford, 2011, Context, Oxford University Press, Oxford, 2014), however, has argued that while centered worlds are needed to characterize e.g. belief states, the contents of such states should be understood as ordinary, uncentered propositions. But Stalnaker does not, as is common, provide a semantics of attitude ascriptions based on the models he develops of the attitudinal states themselves. This paper begins to explore the prospects for doing so. It argues that a simple but well-motivated approach does not yield the principles of knowledge and belief Stalnaker endorses; and that a modification which does brings with it worries of its own surrounding communication and learnability. A technical appendix contains novel and pertinent results in doxastic/epistemic logic.

This book presents 12 original essays on historical and contemporary philosophical discussions of judgment. The central issues explored in this volume can be separated into two groups namely, those concerning the act and object of judgment. What kind of act is judgment? How is it related to a range of other mental acts, states, and dispositions? Where and how does assertive force enter in? Is there a distinct category of negative judgments, or are these simply judgments whose objects are negative? Concerning the object of judgment: How many objects are there of a given judgment? One, as on the dual relation theory of Frege and Moore? Or many as in Russell's later multiple relation theory? If there is a single object, is it a proposition? And if so, is it a force-neutral, abstract entity that might equally figure as the object of a range of intentional attitudes? Or is it somehow constitutively tied to the act itself? These and related questions are approached from a variety of historical and contemporary perspectives. This book sheds new light on current controversies by drawing on the details of the distinct intellectual contexts in which previous philosophers' positions about the nature of judgment were formulated. In turn, new directions in present-day research promise to raise novel interpretive prospects and challenges in the history of philosophy. istorical and contemporary perspectives. This book sheds new light on current controversies by drawing on the details of the distinct intellectual contexts in which previous philosophers' positions about the nature of judgment were formulated. In turn, new directions in present-day research promise to raise novel interpretive prospects and challenges in the history of philosophy.

Safety-based theories of knowledge face a difficulty surrounding necessary truths: no subject could have easily falsely believed such a proposition. Failing to predict that ill-grounded beliefs in such propositions do not constitute knowledge, standard safety theories are therefore less informative than desired. Some have suggested that the subjects at issue could easily have believed some related false proposition; but they have given no indication as to what makes a proposition related. I suggest a solution to this problem: a belief is safe iff its subject could not easily have believed a false answer to the same question.

In this paper, I outline and defend a novel approach to alethic pluralism, the thesis that truth has more than one metaphysical nature: where truth is, in part, explained by reference, it is relational in character and can be regarded as consisting in correspondence; but where instead truth does not depend upon reference it is not relational and involves only coherence. In the process, I articulate a clear sense in which truth may or may not depend upon reference: this involves distinguishing semantic denotation from pragmatic speaker reference and claiming that there may or may not exist a metasemantic connection between these two notions. Finally, I argue that reference is not in general inscrutable—that this metasemantic connection does exist in the case of our ordinary discourse about present macroscopic concrete objects—but that it is in pure mathematics, where reference cannot be secured, and which therefore plays no role in accounting for truth. In this manner, alethic pluralism is upheld.

Carey has argued that there is a system of core numerical cognition – the analog magnitude system – in which cardinal numbers are explicitly represented in iconic format. While the existence of this system is beyond doubt, this paper aims to show that its representations cannot have the combination of features attributed to them by Carey. According to the argument from abstractness, the representation of the cardinal number of a collection of individuals as such requires the representation of individuals as such, and this in turn requires non-iconic format, from which it is concluded that the explicit representation of the cardinal number of some individuals requires non-iconic representational format. In support of the first premise, an account is given of what approximate cardinal numbers might be, and in support of the second, a direct argument is articulated and defended. Finally, in response to an objection, a second argument for the central thesis is provided. While the discussion is couched in the terms of Carey’s work, the considerations it adduces are perfectly general, and the conclusion should therefore be taken into consideration by all those aiming to characterize the AM system.

In this paper I compare Timothy Williamson's knowledge rule of assertion with Ishani Maitra and Brian Weatherson's action rule. The paper is in two parts. In the first part I present and respond to Maitra and Weatherson's master argument against the knowledge rule. I argue that while its second premise, to the effect that an action X can be the thing to do though one is in no position to know that it is, is true, its first premise is not: the data do not support the claim that whenever X is the thing for one to do, one is in a position to assert that it is. In the second part I consider Maitra and Weatherson's alternative hypotheses, arguing that they do not provide a better explanation of the linguistic data. I conclude, in particular, that the knowledge rule is preferable to the action rule.

Stewart Cohen’s New Evil Demon argument raises familiar and widely discussed concerns for reliabilist accounts of epistemic justification. A now standard response to this argument, initiated by Alvin Goldman and Ernest Sosa, involves distinguishing different notions of justification. Juan Comesaña has recently and prominently claimed that his Indexical Reliabilism (IR) offers a novel solution in this tradition. We argue, however, that Comesaña’s proposal suffers serious difficulties from the perspective of the philosophy of language. More specifically, we show that the two readings of sentences involving the word ‘justified’ which are required for Comesaña’s solution to the problem are not recoverable within the two-dimensional framework of Robert Stalnaker to which he appeals. We then consider, and reject, an attempt to overcome this difficulty by appeal to a complication of the theory involving counterfactuals, and conclude the paper by sketching our own preferred solution to Cohen’s New Evil Demon.

Certain puzzling cases have been discussed in the literature recently which appear to support the thought that knowledge can be obtained by way of deduction from a falsehood; moreover, these cases put pressure, prima facie, on the thesis of counter closure for knowledge. We argue that the cases do not involve knowledge from falsehood; despite appearances, the false beliefs in the cases in question are causally, and therefore epistemologically, incidental, and knowledge is achieved despite falsehood. We also show that the principle of counter closure, and the concomitant denial of knowledge from falsehood, is well motivated by considerations in epistemological theory--in particular, by the view that knowledge is first in the epistemological order of things.