In a recent paper, Erasmus et al. (2021) defend the idea that the ambiguity of the term “explanation” in explainable AI (XAI) can be solved by adopting any of four different extant accounts of explanation in the philosophy of science: the Deductive Nomological, Inductive Statistical, Causal Mechanical, and New Mechanist models. In this chapter, I show that the authors’ claim that these accounts can be applied to deep neural networks as they would to any natural phenomenon is mistaken. I also (...) provide a more general argument as to why the notion of explainability as it is currently used in the XAI literature bears little resemblance to the traditional concept of scientific explanation. It would be more fruitful to use the label “understandable AI” to avoid the confusion that surrounds the goal and purposes of XAI. In the second half of the chapter, I argue for a pragmatic conception of understanding that is better suited to play the central role attributed to explanation in XAI. Following De Regt (2017) and Kuorikoski and Ylikoski (2015), the conditions of satisfaction for understanding an ML system are fleshed out in terms of an agent’s success in using the system. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory strategies will (...) lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

Explainable AI (xAI) methods are important for establishing trust in using black-box models. However, recent criticism has mounted against current xAI methods that they disagree, are necessarily false, and can be manipulated, which has started to undermine the deployment of black-box models. Rudin (2019) goes so far as to say that we should stop using black-box models altogether in high-stakes cases because xAI explanations ‘must be wrong’. However, strict fidelity to the truth is historically not a desideratum in science. Idealizations (...) – the intentional distortions introduced to scientific theories and models – are commonplace in the natural sciences and are seen as a successful scientific tool. Thus, it is not falsehood qua falsehood that is the issue. In this paper, I outline the need for xAI research to engage in idealization evaluation. Drawing on the use of idealizations in the natural sciences and philosophy of science, I introduce a novel framework for evaluating whether xAI methods engage in successful idealizations or deceptive explanations (SIDEs). SIDEs evaluates whether the limitations of xAI methods, and the distortions that they introduce, can be part of a successful idealization or are indeed deceptive distortions as critics suggest. I discuss the role that existing research can play in idealization evaluation and where innovation is necessary. Through a qualitative analysis we find that leading feature importance methods and counterfactual explanations are subject to idealization failure and suggest remedies for ameliorating idealization failure. (shrink)

In the recent literature, several authors have argued that the capacity to track factive mental states plays a central role in explaining our ability to understand and predict people’s behavior (Nagel 2013; Nagel 2017; Phillips & Norby 2019; Phillips et al. 2020; Westra & Nagel 2021). The topic of this chapter is whether this capacity also enters into an explanation of our ability to track skilled and intentional actions.

Frege begins his discussion of factives in ‘On Sense and Reference’ with an example of a purported contrafactive, that is, a verb that entails, or presupposes, the falsity of the complement sentence. But the verb he cites, ‘wähnen’, is now obsolete, and native speakers are sceptical about whether it really was a contrafactive. Despite the profusion of factive verbs, there are no clear examples of contrafactive propositional attitude verbs in English, French or German. This paper attempts to give an explanation (...) of this, and to use this to shed light on the behaviour of factives more generally. The suggestion is that factive propositional attitude verbs take facts, not propositions, as the referents of their complement sentences; and that as there are no contra-facts, there can be no contra-factives. This claim is also used to help explain Timothy Williamson's observation that there is no stative propositional attitude factive that requires only belief. Various conclusions are drawn within a broadly ‘knowledge first’ approach. (shrink)

As systems based on opaque Artificial Intelligence (AI) continue to flourish in diverse real-world applications, understanding these black box models has become paramount. In response, Explainable AI (XAI) has emerged as a field of research with practical and ethical benefits across various domains. This paper not only highlights the advancements in XAI and its application in real-world scenarios but also addresses the ongoing challenges within XAI, emphasizing the need for broader perspectives and collaborative efforts. We bring together experts from diverse (...) fields to identify open problems, striving to synchronize research agendas and accelerate XAI in practical applications. By fostering collaborative discussion and interdisciplinary cooperation, we aim to propel XAI forward, contributing to its continued success. Our goal is to put forward a comprehensive proposal for advancing XAI. To achieve this goal, we present a manifesto of 27 open problems categorized into nine categories. These challenges encapsulate the complexities and nuances of XAI and offer a road map for future research. For each problem, we provide promising research directions in the hope of harnessing the collective intelligence of interested stakeholders. (shrink)

I evaluate non-factive or truth-insensitive accounts of the ordinary concepts used to evaluate beliefs, evidence, assertions, and decisions. Recent findings show that these accounts are mistaken. I propose three hypotheses regarding how philosophers defending these accounts got things so wrong. I also consider one potential consequence for the discipline.

Famously, the Church–Fitch paradox of knowability is a deductive argument from the thesis that all truths are knowable to the conclusion that all truths are known. In this argument, knowability is analyzed in terms of having the possibility to know. Several philosophers have objected to this analysis, because it turns knowability into a nonfactive notion. In addition, they claim that, if the knowability thesis is reformulated with the help of factive concepts of knowability, then omniscience can be avoided. In this (...) article we will look closer at two proposals along these lines :557–568, 1985; Fuhrmann in Synthese 191:1627–1648, 2014a), because there are formal models available for each. It will be argued that, even though the problem of omniscience can be averted, the problem of possible or potential omniscience cannot: there is an accessible state at which all truths are known. Furthermore, it will be argued that possible or potential omniscience is a price that is too high to pay. Others who have proposed to solve the paradox with the help of a factive concept of knowability should take note :53–73, 2010; Spencer in Mind 126:466–497, 2017). (shrink)

In “The Myth of Factive Verbs” (Hazlett 2010), I had four closely related goals. The first (pp. 497-99, p. 522) was to criticize appeals to ordinary language in epistemology. The second (p. 499) was to criticize the argument that truth is a necessary condition on knowledge because “knows” is factive. The third (pp. 507-19) – which was the intended means of achieving the first two – was to defend a semantics for “knows” on which <S knows p> can be true (...) even if p isn’t true. The fourth (Ibid.) – which seemed necessary for the success of the third – was to defend a pragmatic account of the fact that utterances of <S knows p> typically imply p, on which the implication in those cases is down to conversational implicature. In this paper I’ll go after these goals again, with an emphasis on the second. Our topic will be whether the factivity of “knows” (whatever this amounts to) supports the truth condition on knowledge. A new goal will be to defend my argument against some criticisms from John Turri (2011) and Savas Tsohatzidis (forthcoming). We’ll first look at the truth condition (§1) and factive presupposition (§§2 – 3), before turning to replies to Turri and Tsohatzidis (§§4 – 7). (shrink)

In this paper I give a novel argument for the view that epistemic normative reasons (or evidence) need not be facts. I first argue that the nature of normative reasons is uniform, such that our positions about the factivity of reasons should agree across normative realms –– whether epistemic, moral, practical, or otherwise. With that in mind, I proceed in a somewhat indirect way. I argue that if practical motivating reasons are not factive, then practical normative reasons are not (...) factive. If it is possible to act rationally in the light of a falsehood, as I will say, then some good reasons must be falsehoods. The implication of this argument is perhaps surprising: for one to firmly establish the view that epistemic normative reasons are factive, one must discredit the view that practical motivating reasons are not factive. (shrink)

In epistemology and philosophy of science, there has been substantial debate about truth’s relation to understanding. “Non-factivists” hold that radical departures from the truth are not always barriers to understanding; “quasi-factivists” demur. The most discussed example concerns scientists’ use of idealizations in certain derivations of the ideal gas law from statistical mechanics. Yet, these discussions have suffered from confusions about the relevant science, as well as conceptual confusions. Addressing this example, we shall argue that the ideal gas law is best (...) interpreted as favoring non-factivism about understanding, but only after delving a bit deeper into the statistical mechanics that has informed these arguments and stating more precisely what non-factivism entails. Along the way, we indicate where earlier discussions have gone astray, and highlight how a naturalistic approach furnishes more nuanced normative theses about the interaction of rationality, understanding, and epistemic value. (shrink)

This is a reply to Howard Sankey’s comment (“Factivity or Grounds? Comment on Mizrahi”) on my paper, “You Can’t Handle the Truth: Knowledge = Epistemic Certainty,” in which I present an argument from the factivity of knowledge for the conclusion that knowledge is epistemic certainty. While Sankey is right that factivity does not entail epistemic certainty, the factivity of knowledge does entail that knowledge is epistemic certainty.

How radical is the idea that reasons are factive? Some philosophers consider it a dramatic departure from orthodoxy, with surprising implications about the bearing of the external world on what credences it’s reasonable to have, what beliefs are epistemically appropriate, and what actions are rational. I deny these implications. In the cases where external matters imply differences in factive states, there will inevitably be important weaker factive states in common. For example, someone who knows it is raining has many factive (...) states in common with someone who has a Gettiered belief that it is raining, or one who falsely but justifiably believes that it is raining. The factive reasons denied to subjects in Gettier cases or skeptical scenarios are in an important sense redundant; appropriate belief or action supervenes on internal states, even if reasons must be factive (and even if appropriate belief and action supervenes on reasons). The degree to which the strategy is applicable depends substantively on epistemic assumptions about basic or foundational knowledge. I argue that even given a significantly externalist approach to the latter internalist intuitions about rational credence, belief, and action can be vindicated. (shrink)

The main question addressed in this paper is whether some false sentences can constitute evidence for the truth of other propositions. In this paper it is argued that there are good reasons to suspect that at least some false propositions can constitute evidence for the truth of certain other contingent propositions. The paper also introduces a novel condition concerning propositions that constitute evidence that explains a ubiquitous evidential practice and it contains a defense of a particular condition concerning the possession (...) of evidence. The core position adopted here then is that false propositions that are approximately true reports of measurements can constitute evidence for the truth of other propositions. So, it will be argued that evidence is only quasi-factive in this very specific sense. (shrink)

Previous research in Explainable Artificial Intelligence (XAI) suggests that a main aim of explainability approaches is to satisfy specific interests, goals, expectations, needs, and demands regarding artificial systems (we call these “stakeholders' desiderata”) in a variety of contexts. However, the literature on XAI is vast, spreads out across multiple largely disconnected disciplines, and it often remains unclear how explainability approaches are supposed to achieve the goal of satisfying stakeholders' desiderata. This paper discusses the main classes of stakeholders calling for explainability (...) of artificial systems and reviews their desiderata. We provide a model that explicitly spells out the main concepts and relations necessary to consider and investigate when evaluating, adjusting, choosing, and developing explainability approaches that aim to satisfy stakeholders' desiderata. This model can serve researchers from the variety of different disciplines involved in XAI as a common ground. It emphasizes where there is interdisciplinary potential in the evaluation and the development of explainability approaches. (shrink)

Fitch's Paradox and the Paradox of the Knower both make use of the Factivity Principle. The latter also makes use of a second principle, namely the Knowledge-of-Factivity Principle. Both the principle of factivity and the knowledge thereof have been the subject of various discussions, often in conjunction with a third principle known as Closure. In this paper, we examine the well-known Surprise Examination paradox considering both the principles on which this paradox rests and some formal characterisations of (...) the surprise notion, crucial in this paradox. Standard formalizations of the Surprise Examination paradox in modal logic do not seem, at first glance, to depend on either factivity or knowledge-of-factivity, but we will argue that both factivity and knowledge-of-factivity play a key implicit role in the paradox. Namely, they are implicitly, perhaps unintentionally, used in order to simplify the definition of surprise. We analyze modal logical formalizations of three versions of the paradox concluding that the Surprise Examination paradox is the result of two flaws: the assumption of knowledge-of-factivity, and the over-simplification of the definition of "surprise" accordingly. By fixing these two flaws, the Surprise Examination paradox vanishes. (shrink)

It’s a cornerstone of epistemology that knowledge requires truth – that is, that knowledge is factive. Allan Hazlett boldly challenges orthodoxy by arguing thatthe ordinary concept of knowledge is not factive. On this basis Hazlett further argues that epistemologists shouldn’t concern themselves with the ordinary concept of knowledge, or knowledge ascriptions and related linguistic phenomena. I argue that either Hazlett is wrong about the ordinary concept of knowledge, or he’s right in a way that leaves epistemologists to carry on exactly (...) as they have, paying attention to much the same things they always did. (shrink)

In this paper we argue that standard, indexical contextualism about "knows" is unable to account for the factivity of this epistemic expression.

Antirealists who hold the knowability thesis, namely that all truths are knowable, have been put on the defensive by the Church-Fitch paradox of knowability. Rejecting the non-factivity of the concept of knowability used in that paradox, Edgington has adopted a factive notion of knowability, according to which only actual truths are knowable. She has used this new notion to reformulate the knowability thesis. The result has been argued to be immune against the Church-Fitch paradox, but it has encountered several (...) other triviality objections. Schlöder in a forthcoming paper defends the general approach taken by Edgington, but amends it to save it in turn from the triviality objections. In this paper I will argue, first, that Schlöder's justification for the factivity of his version of the concept of knowability is vulnerable to criticism, but I will also offer an improved justification that is in the same spirit as his. To the extent that some philosophers are right about our intuitive concept of knowability being a factive one, it is important to explore factive concepts of knowability that are made formally precise. I will subsequently argue that Schlöder's version of the knowability thesis overgenerates knowledge or, in other words, it leads to attributions of knowledge where there is ignorance. This fits a general pattern for the research programme initiated by Edgington. This paper also contains preliminary investigations into the internal and logical structure of lines of inquiries, which raise interesting research questions. (shrink)

Perceptions guide our actions and provide us with evidence of the world around us. Illusions and hallucinations can mislead us: they may prompt as to act in ways that do not mesh with the world around us and they may lead us to form false beliefs about that world. The capacity view provides an account of evidence that does justice to these two facts. It shows in virtue of what illusions and hallucinations mislead us and prompt us to act. Moreover, (...) it shows in virtue of what we are in a better epistemic position when we perceive than when we hallucination. In this paper, I develop the capacity view, that is, the view that perceptual experience has epistemic force in virtue of the epistemic and metaphysical primacy of the perceptual capacities employed in perception. By grounding the epistemic force of experience in facts about the metaphysical structure of experience, the capacity view is not only an externalist view, but moreover a naturalistic view of the epistemology of perceptual experience. So it is an externalist and naturalistic alternative to reliabilism. I discuss the repercussions of this view for the justification of beliefs and the epistemic transparency of mental states, as well as, familiar problem cases. (shrink)

Suppose that A explains B. Do A and B need to be true? Provided that we have metaphysical explanation in mind, orthodoxy answers “yes:” metaphysical explanation is factive. This article introduces and defends a non-factive notion of metaphysical explanation. I argue that we need a non-factive notion of explanation in order to make sense of explanationist arguments where we motivate a view by claiming that it offers better explanations than its competitors. After presenting and rejecting some initially plausible rivals, I (...) account for non-factive metaphysical explanation by drawing on existing applications of structural equation models to metaphysical grounding. (shrink)

This paper defends and develops the capacity view against insightful critiques from Matt McGrath, Adam Pautz, and Ram Neta. In response to Matt McGrath, I show why capacities are essential and cannot simply be replaced with representational content. I argue moreover, that the asymmetry between the employment of perceptual capacities in the good and the bad case is sufficient to account for the epistemic force of perceptual states yielded by the employment of such capacities. In response to Adam Pautz, I (...) show why a perceiver’s belief is better justified than the belief of someone who suffers a subjectively indistinguishable hallucination. I show, moreover, why the capacity view is compatible with standard Bayesian principles and how it accounts for degrees of justification. In response to Ram Neta, I discuss the relationship between evidence and rational confidence, as well as the notion of evidence in light of an externalism about perceptual content. (shrink)

The thesis that knowledge is a factive mental state plays a central role in knowledge-first epistemology, but accepting this thesis requires also accepting an unusually severe version of externalism about the mind. On this strong attitude externalism, whether S is in the mental state of knowledge can and often will rapidly change in virtue of changes in external states of reality with which S has no causal contact. It is commonly thought that this externalism requirement originates in the factivity (...) of knowledge. However, despite a number of recent defenses of non-factive accounts of knowledge, epistemology has yet to consider whether a non-factive approach might produce a version of the mental state thesis that can avoid strong externalism. Here I do just that, exploring how three different proposals for weakening factivity might be adapted to theories of knowledge as a non-factive mental state. Contrary to what we might expect, however, none of these proposals are compatible with anything close to attitude internalism about knowledge—or even a substantially weaker externalism. All told, the widespread view that wraps up the severe externalism required for knowledge to be a mental state in factivity is mistaken. Knowledge’s external-world connection runs far deeper than the factivity constraint. (shrink)

In recent discussions in this journal, Moti Mizrahi defends the claim that knowledge equals epistemic certainty. Howard Sankey finds Mizrahi’s argument to be problematic, since, as he reads it, this would entail that justification must guarantee truth. In this article, I suggest that an account of the normativity of justification is able to bridge the gap between Mizrahi’s proposal and Sankey’s objections.

The concept of knowledge can be modelled in epistemic modal logic and, if modelled by using a standard modal operator, it is subject to the problem of logical omniscience. The classical solution to this problem is to distinguish between implicit and explicit knowledge and to construe the knowledge operator as capturing the concept of implicit knowledge. In addition, since a proposition is said to be implicitly known just in case it is derivable from the set of propositions that are explicitly (...) known by using a certain set of logical rules, the concept of implicit knowledge is definable on the basis of the concept of explicit knowledge. In any case, both implicit and explicit knowledge are typically characterized as factive, i.e. such that it is always the case that what is known is also true. The aim of the present paper is twofold: first, we will develop a dynamic system of explicit intersubjective knowledge that allows us to introduce the operator of implicit knowledge by definition; secondly, we will show that it is not possible to hold together the following two theses: (1) the concept of implicit knowledge is definable along the lines indicated above and (2) the concept of implicit knowledge is factive. (shrink)

This paper is a response to Baumann's comments on "Can Knowledge Really be Non-fative?" In this paper Baumann's suggestions for how those who deny the factivty of knowledge might deal with the argument from inconsistency and explosion are addressed.

It is widely thought that knowledge is factive – only truths can be known. However, this view has been recently challenged. One challenge appeals to approximate truths. Wesley Buckwalter and John Turri argue that false-but-approximately-true propositions can be known. They provide experimental findings to show that their view enjoys intuitive support. In addition, they argue that we should reject the factive account of knowledge to avoid widespread skepticism. A second challenge, advanced by Nenad Popovic, appeals to multidimensional geometry to build (...) a case where it seems intuitive that a person knows p even though p is false. In addition, Popovic argues that we should reject the factive account of knowledge because most of us would not become widespread skeptics if we discovered that ordinary objects in our world are actually four-dimensional. In this paper, we defend the factive account of knowledge against these arguments by challenging the intuitive appeal of the cases and arguing that there is no real threat of widespread skepticism for the factive account of knowledge. (shrink)

There are three main accounts of scientific progress: 1) the epistemic account, according to which an episode in science constitutes progress when there is an increase in knowledge; 2) the semantic account, according to which progress is made when the number of truths increases; 3) the problem-solving account, according to which progress is made when the number of problems that we are able to solve increases. Each of these accounts has received several criticisms in the last decades. Nevertheless, some authors (...) think that the epistemic account is to be preferred if one takes a realist stance. Recently, Dellsén proposed the noetic account, according to which an episode in science constitutes progress when scientists achieve increased understanding of a phenomenon. Dellsén claims that the noetic account is a more adequate realist account of scientific progress than the epistemic account. This paper aims precisely at assessing whether the noetic account is a more adequate realist account of progress than the epistemic account. (shrink)

This paper shows that the knowability paradox isn’t a paradox because the derivation of the paradox is faulty. This is explained by showing that the K operator employed in generating the paradox is used equivocally and when the equivocation is eliminated the derivation fails.

Philosophers are divided over moral internalism, the claim that moral judgement entails some motivation to comply with that judgement. Against moral internalism, externalists defend the conceptual coherence of scenarios in which an individual makes genuine moral judgements but is entirely unmoved by them. This is amoralist skepticism and these scenarios can be called amoralist scenarios. While the coherence of amoralist scenarios is disputed, philosophers seem to agree that the coherence of amoralist scenarios is not affected by whether the amoralist is (...) described as having moral knowledge or mere belief. But recent experimental research challenges this assumption. When evaluating amoralist scenarios, people’s intuitions lean towards externalism when the amoralist is described as knowing that X is morally wrong, whereas people’s intuitions lean towards internalism when the amoralist is described as believing that X is morally wrong. Call this the factivity effect. In this paper, I argue that the factivity effect is unlikely to be explained as an experimental artifact and that as a consequence, the traditional dispute over moral internalism and amoralist skepticism may need a major overhaul. The results of three studies testing the factivity effect provide support for this thesis. Implications of these results for the traditional debate over moral internalism are discussed. (shrink)

This paper delves into one particular topic within this Buddhist theory of cognition. I examine a single argument by Phywa pa Chos kyi seṅ ge (1109–1169) contained within his famous epistemology text, the Tshad ma yid kyi mun sel, drawing out the philosophical implications that this argument has on his theory of cognition and his account of ontological dependence. I make the case that Phywa pa’s argument fails to explain adequately the nature of the relation between certain cognitive episodes and (...) the contents of those episodes. In addition, I will show that Phywa pa is forced to accept an arguably dubious version of externalism about mental content. (shrink)

How is human social intelligence engaged in the course of ordinary conversation? Standard models of conversation hold that language production and comprehension are guided by constant, rapid inferences about what other agents have in mind. However, the idea that mindreading is a pervasive feature of conversation is challenged by a large body of evidence suggesting that mental state attribution is slow and taxing, at least when it deals with propositional attitudes such as beliefs. Belief attributions involve contents that are decoupled (...) from our own primary representation of reality; handling these contents has come to be seen as the signature of full-blown human mindreading. However, mindreading in cooperative communication does not necessarily demand decoupling. We argue for a theoretical and empirical turn towards “factive” forms of mentalizing here. In factive mentalizing, we monitor what others do or do not know, without generating decoupled representations. We propose a model of the representational, cognitive, and interactive components of factive mentalizing, a model that aims to explain efficient real-time monitoring of epistemic states in conversation. After laying out this account, we articulate a more limited set of conversational functions for nonfactive forms of mentalizing, including contexts of meta-linguistic repair, deception, and argumentation. We conclude with suggestions for further research into the roles played by factive versus nonfactive forms of mentalizing in conversation. (shrink)

In the General Psychopathology Jaspers famously draws a distinction between the understandable and explainable. Meaningful connections between psychic events, he argues, can only be understood empathetically and cannot be explained causally. The idea behind this distinction, according to some interpreters at least, seems to be that psychic events do not fall under any general causal rules whereas ordinary events do fall under such rules. Also Jaspers distinguishes empathetic understanding of the connection between two psychic events from a mere interpretation of (...) it, which may turn out to be false. Hence, understanding seems to be able to give us the truth about the connection and is factive as well as self-evident in nature. Contemporary epistemologists, such as Linda Zagzebsky, Duncan Pritchard, and Jonathan Kvanvig, for example, distinguish three varieties: propositional, objectual or holistic, and atomistic understanding. They do not agree on factivity and transparency of understanding. What then is the difference between their views and that of Jaspers? This essay compares recent epistemological views of understanding with those of Jaspers and critiques his claims about empathetic understanding as being both factive and self-evident or transparent; to show that empathetic understanding of connections between psychic events needs a public criterion for its individuation. (shrink)

For synergistic interactions between humans and artificial intelligence (AI) systems, AI outputs often need to be explainable to people. Explainable AI (XAI) systems are commonly tested in human user studies. However, whether XAI researchers consider potential cultural differences in human explanatory needs remains unexplored. We highlight psychological research that found significant differences in human explanations between many people from Western, commonly individualist countries and people from non-Western, often collectivist countries. We argue that XAI research currently overlooks these variations and that (...) many popular XAI designs implicitly and problematically assume that Western explanatory needs are shared cross-culturally. Additionally, we systematically reviewed over 200 XAI user studies and found that most studies did not consider relevant cultural variations, sampled only Western populations, but drew conclusions about human-XAI interactions more generally. We also analyzed over 30 literature reviews of XAI studies. Most reviews did not mention cultural differences in explanatory needs or flag overly broad cross-cultural extrapolations of XAI user study results. Combined, our analyses provide evidence of a cultural bias toward Western populations in XAI research, highlighting an important knowledge gap regarding how culturally diverse users may respond to widely used XAI systems that future work can and should address. (shrink)

I will argue that cases of massive deception, such as New Evil Demon cases, as well as one-off cases of local deception present challenges to views according to which epistemic reasons, epistemic warrant, epistemic rationality or epistemic norms are factive. In doing so, I will argue is that proponents of a factive turn in epistemology should observe important distinctions between what are often simply referred to as ‘bad cases.’ Recognizing epistemologically significant differences between deception cases raises serious challenges for those (...) who deny a central role for non-factive aspects of epistemology. (shrink)

A fundamental question asked in modal logic is whether a given theory is consistent. But consistent with what? A typical way to address this question identifies a choice of background knowledge axioms (say, S4, D, etc.) and then shows the assumptions codified by the theory in question to be consistent with those background axioms. But determining the specific choice and division of background axioms is, at least sometimes, little more than tradition. This paper introduces generic theories for propositional modal logic (...) to address consistency results in a more robust way. As building blocks for background knowledge, generic theories provide a standard for categorical determinations of consistency. We argue that the results and methods of this paper help to elucidate problems in epistemology and enjoy sufficient scope and power to have purchase on problems bearing on modalities in judgement, inference, and decision making. (shrink)

Mark Schroeder’s Reasons First is admirable in its scope and execution, deftly demonstrating the theoretical promise of extending the reasons-first approach from ethics to epistemology. In what follows we explore how (not) to account for the evidence-that relation within the reasons-first program, we explain how factive content views of evidence can be resilient in the face of Schroeder’s criticisms, and we explain how knowledge from falsehood threatens Schroeder’s view of knowledge. Along the way we sketch a reliabilist account of the (...) reasons-for relation (inspired by Alston) and show how it limits the applicability of Schroeder’s arguments. (shrink)

There has been a push in recent years to provide better explanations for how AIs make their decisions. Most of this push has come from the ethical concerns that go hand in hand with AIs making decisions that affect humans. Outside of the strictly ethical concerns that have prompted the study of explainable AIs (XAIs), there has been research interest in the mere possibility of creating XAIs in various domains. In general, the more accurate we make our models the harder (...) they are to explain. Go playing AIs like AlphaGo and KataGo provide fantastic examples of this phenomenon. In this paper, I discuss a non-exhaustive list of the leading theories of explanation and what each of these theories would say about the explainability of AI-played moves of Go. Finally, I consider the possibility of ever explaining AI-played Go moves in a way that meets the four principles of XAI. I conclude, somewhat pessimistically, that Go is not as imminently explainable as other domains. As such, the probability of having an XAI for Go that meets the four principles is low. (shrink)

The philosophical interest in the nature, value, and varieties of human understanding has swelled in recent years. This article will provide an overview of new research in the epistemology of understanding, with a particular focus on the following questions: What is understanding and why should we care about it? Is understanding reducible to knowledge? Does it require truth, belief, or justification? Can there be lucky understanding? Does it require ‘grasping’ or some kind of ‘know-how’? This cluster of questions has largely (...) set the research agenda for the study of understanding in epistemology. This article will conclude by discussing some varieties of understanding and highlight directions for future research. (shrink)

According to Anscombe, acting intentionally entails knowledge in ac- tion. This thesis has been near-universally rejected due to a well-known counter- example by Davidson: a man intending to make ten legible carbon copies might not believe with confidence, and hence not know, that he will succeed. If he does, however, his action surely counts as intentional. Damaging as it seems, an even more powerful objection can be levelled against Anscombe: while act- ing, there is as yet no fact of the (...) matter as to whether the agent will succeed. Since his belief that he will is not yet true while his action is in progress, he can- not possibly know that he is indeed bringing about the intended goal. Knowl- edge in action is not only unnecessary for intentional action, it seems, but–at least as regards success-bound types of action–impossible to attain in the first place. -/- In this paper I argue that traditional strategies to counter these objections are unsatisfactory and propose a new account of knowledge in action which has two core features: (i) It invokes an externalist conception of justification which not only meets Davidson’s challenge, but also casts doubts on the tacit internalist premise on which his example relies. (ii) Drawing on recent work about by John MacFarlane, the proposed account conceives of claims to in action as assessment-sensitive so as to overcome the factivity objection. From a retrospective point of evaluation, previous claims about future events and actions can not only be deemed as having been true, but also as having been known. (shrink)

While epistemologists have long debated what it takes for beliefs to be justified, they've devoted much less collective attention to the question of what it takes for beliefs to be excused, and how excuses differ from justifications. This stands in contrast to the state of affairs in legal scholarship, where the contrast between justifications and excuses is a standard topic in introductory criminal law textbooks. My goal in this paper is to extract some lessons from legal theory for epistemologists seeking (...) to distinguish epistemic justifications from epistemic excuses. First, I'll provide a brief overview of appeals to the justification/excuse distinction in epistemology, focusing on the internalism/externalism debate. Then, I'll identify some choice points in legal theory for how to think about the function of the justification/excuse distinction, and will in turn correspond to analogous choice points in epistemology. Ultimately, I'll argue that depending on which way we go at some crucial junctures, we can motivate either a traditional, non‐factive, internalist conception of justification, or a purely factive one where truth is all that's required for a belief to be justified. But there's no clear path, via legal theory, to view that only beliefs that amount to knowledge are justified. (shrink)

Many ethical frameworks require artificial intelligence (AI) systems to be explainable. Explainable AI (XAI) models are frequently tested for their adequacy in user studies. Since different people may have different explanatory needs, it is important that participant samples in user studies are large enough to represent the target population to enable generalizations. However, it is unclear to what extent XAI researchers reflect on and justify their sample sizes or avoid broad generalizations across people. We analyzed XAI user studies (N = (...) 220) published between 2012 and 2022. Most studies did not offer rationales for their sample sizes. Moreover, most papers generalized their conclusions beyond their target population, and there was no evidence that broader conclusions in quantitative studies were correlated with larger samples. These methodological problems can impede evaluations of whether XAI systems implement the explainability called for in ethical frameworks. We outline principles for more inclusive XAI user studies. (shrink)

Can AI developers be held epistemically responsible for the processing of their AI systems when these systems are epistemically opaque? And can explainable AI (XAI) provide public justificatory reasons for opaque AI systems’ outputs? Koskinen (2024) gives negative answers to both questions. Here, I respond to her and argue for affirmative answers. More generally, I suggest that when considering people’s uncertainty about the factors causally determining an opaque AI’s output, it might be worth keeping in mind that a degree of (...) uncertainty about conclusions is inevitable even in entirely human-based empirical science because in induction there’s always a risk of getting it wrong. Keeping this in mind may help appreciate that requiring full transparency from AI systems before epistemically trusting their outputs might be unusually (and potentially overly) demanding. (shrink)

There is a kind of objectivism in epistemology that involves the acceptance of objective epistemic norms. It is generally regarded as harmless. There is another kind of objectivism in epistemology that involves the acceptance of an objectivist account of justification, one that takes the justification of a belief to turn on its accuracy. It is generally regarded as hopeless. It is a strange and unfortunate sociological fact that these attitudes are so prevalent. Objectivism about norms and justification stand or fall (...) together. Justification is simply a matter of conforming to norms. In this essay, I shall make the case for objectivism about justification. (shrink)

This paper advances our understanding of the norms of assertion in two ways. First, I evaluate recent studies claiming to discredit an important earlier finding which supports the hypothesis that assertion has a factive norm. In particular, I evaluate whether it was due to stimuli mentioning that a speaker’s evidence was fallible. Second, I evaluate the hypothesis that assertion has a truth-insensitive standard of justification. In particular, I evaluate the claim that switching an assertion from true to false, while holding (...) all else objectively constant, is irrelevant to attributions of justification. Two pre-registered experiments provide decisive evidence against each claim. In the first experiment, switching from mentioning to not mentioning fallibility made no difference to assertability attributions, thereby disproving the criticism concerning fallibility. By contrast, switching an assertion from true to false decreased the rate of assertability attribution from over 90% to less than 20%, thereby replicating and vindicating the original finding supporting a factive norm. In the second experiment, switching an assertion from true to false decreased the rate of justification attribution from over 80 to 10%, thereby undermining the hypothesis that assertion’s standard of justification is truth-insensitive. The second experiment also demonstrates that perspective-taking influences attributions of justification, and it provides initial evidence that the standard of justification for assertion is stricter than the standard for belief. (shrink)

Necessity and sufficiency are the building blocks of all successful explanations. Yet despite their importance, these notions have been conceptually underdeveloped and inconsistently applied in explainable artificial intelligence (XAI), a fast-growing research area that is so far lacking in firm theoretical foundations. Building on work in logic, probability, and causality, we establish the central role of necessity and sufficiency in XAI, unifying seemingly disparate methods in a single formal framework. We provide a sound and complete algorithm for computing explanatory factors (...) with respect to a given context, and demonstrate its flexibility and competitive performance against state of the art alternatives on various tasks. (shrink)

It is almost universally presumed that knowledge is factive: in order to know that p it must be the case that p is true. This idea is often justified by appealing to knowledge ascriptions and related linguistic phenomena; i.e., an utterance of the form ‘S knows that p, but not-p’ sounds contradictory. In a recent article, Allan Hazlett argues that our ordinary concept of knowledge is not factive. From this it seems to follow that epistemologists cannot appeal to ordinary language (...) to justify the truth condition of knowledge. More significantly, Hazlett claims that epistemologists theorizing about knowledge should not concern themselves with the ordinary concept of knowledge as revealed by knowledge ascriptions and related linguistic phenomena. My paper has two goals: first, to defend the orthodox view that the ordinary concept of knowledge is factive; second, to undermine Hazlett’s claim that epistemologists should not theorize about knowledge on the basis of how ‘knows’ is used in everyday speech. (shrink)

Artificial intelligence (AI) systems are increasingly adopted to make decisions in domains such as business, education, health care, and criminal justice. However, such algorithmic decision systems can have prevalent biases against marginalized social groups and undermine social justice. Explainable artificial intelligence (XAI) is a recent development aiming to make an AI system’s decision processes less opaque and to expose its problematic biases. This paper argues against technical XAI, according to which the detection and interpretation of algorithmic bias can be handled (...) more or less independently by technical experts who specialize in XAI methods. Drawing on resources from feminist epistemology, we show why technical XAI is mistaken. Specifically, we demonstrate that the proper detection of algorithmic bias requires relevant interpretive resources, which can only be made available, in practice, by actively involving a diverse group of stakeholders. Finally, we suggest how feminist theories can help shape integrated XAI: an inclusive social-epistemic process that facilitates the amelioration of algorithmic bias. (shrink)

This study demonstrates the combinations of multiple causal factors that formulate a startup’s strategy to successfully “exit”, namely “recipes for a successful exit,” in the clean- and hard-tech sector. We identify seven key causal factors (i.e., causal conditions) that impact startup success, including commercial readiness, investor interactions, favorable industry, non-financial support, straightforward development path, experienced team, and visibility to investors. We also investigate the combinations of selective causal conditions that can provide further synergetic impact. We conduct the fuzzy-set qualitative comparative (...) analysis (fsQCA) on seven U.S. clean- and hard-tech startups that exited between 2005 and 2016. The successful companies all demonstrate distinctive characteristics based on three general categories (1) robust ecosystem; (2) heavy-lifting team; or (3) external opportunity. Across these three categories, commercial readiness and strong investor interactions are necessary conditions for all exit cases. But there are important differences that drive success in each category, such as the interaction with non-financial support (in the robust ecosystem case), experienced team (in the heavy-lifting team case), and favorable industry (in the external opportunity case). Our findings are meant to support entrepreneurs in reaching an exit by optimizing the given internal and external circumstances, and policymakers to build a robust ecosystem that can increase the success rate of the clean- and hard-tech development. (shrink)