Artificial intelligence (AI) has become nearly ubiquitous in modern society, from components of mobile applications to medical support systems, and everything in between. In societally impactful systems imbued with AI, there has been increasing concern related to opaque AI, that is, artificial intelligence where it is unclear how or why certain decisions are reached. This has led to a recent boom in research on “explainable AI” (XAI), or approaches to making AI more explainable and understandable to human users. (...) In the military domain, numerous bodies have argued that autonomous and AI-enabled weapon systems ought not incorporate unexplainable AI, with the International Committee of the Red Cross and the United States Department of Defense both explicitly including explainability as a relevant factor in the development and use of such systems. In this article, I present a cautiously critical assessment of this view, arguing that explainability will be irrelevant for many current and near-future autonomous systems in the military (which do not incorporate any AI), that it will be trivially incorporated into most military systems which do possess AI (as these generally possess simpler AI systems), and that for those systems with genuinely opaque AI, explainability will prove to be of more limited value than one might imagine. In particular, I argue that explainability, while indeed a virtue in design, is a virtue aimed primarily at designers and troubleshooters of AI-enabled systems, but is far less relevant for users and handlers actually deploying these systems. I further argue that human–machine teaming is a far more important element of responsibly using AI for military purposes, adding that explainability may undermine efforts to improve human–machine teamings by creating a prima facie sense that the AI, due to its explainability, may be utilized with little (or less) potential for mistakes. I conclude by clarifying that the arguments are not against XAI in the military, but are instead intended as a caution against over-inflating the value of XAI in this domain, or ignoring the limitations and potential pitfalls of this approach. (shrink)

[[[ (Here only the chapters 3 – 8, see *** ) First I argue that the prohibition of linguistic self-reference as a solution to the antinomy problem contains a pragmatic contradiction and is thus not only too restrictive, but just inconsistent (chap.1). Furthermore, the possibilities of non-restrictive strategies for antinomy avoidance are discussed, whereby the explicit inclusion of the – pragmatically presuposed – consistency requirement proves to be the optimal strategy (chap.2). ]]] The central question here is that about the (...) actual reason for antinomic structures. It turns out to be a form of negative self-conditioning (chap.3). This makes it necessary to clarify the status of negative concepts (chap.4). The generalization of these considerations (chap.5) leads to the actual analysis of the antinomic basic structure (chap.6): Decisively for the pragmatics of the concept is that it positively owns a meaning, so that positivity is always constituted by the concept qua concept. Thus a negative concept is characterized by a fundamental ambivalence: From a semantic point of view it has negative character, in its pragmatic status as a concept, however, it has positive character. If the meaning is especially that of a negative self-reference, the ambivalence leads to the antinomic constellation of a negative self-condition – that is the crux of the matter here! The concept thus possesses the property defined by it exactly when it does not possess it and vice versa. A closer analysis shows that the function of reflective structures for the occurrence of antinomies has to be judged much more differentiated than previous opinions suggest. Not only are four forms of reflectivity to be distinguished in this context – ontic, semantic, pragmatic reflectivity, and especially the form of negative self-condition; but it is also apparent that these are intertwined with each other in a way that is difficult to be understood. The astonishing variety of relationships associated with this makes the irritation that has always emanated from the antinomy problem appear more comprehensible. In the developed pragmatic perspective furtheron parallels to the structure of self-consciousness become visible (chap.7). I conclude with considerations about the significance of the antinomic structure for the problem of dialectics, especially for the synthesis formation of mutually exclusive terms (chap.8). (shrink)

Carl Craver investigates what we are doing when we sue neuroscience to explain what's going on in the brain.

In this lucid portrayal of human behavior, Fred Dretske provides an original account of the way reasons function in the causal explanation of behavior.

Two studies examined the specificity of effects of explanation on learning by prompting 3- to 6-year-old children to explain a mechanical toy and comparing what they learned about the toy’s causal and non-causal properties to children who only observed the toy, both with and without accompanying verbalization. In Study 1, children were experimentally assigned to either explain or observe the mechanical toy. In Study 2, children were classified according to whether the content of their response to an undirected prompt involved (...) explanation. Dependent measures included whether children understood the toy’s functional-mechanical relationships, remembered perceptual features of the toy, effectively reconstructed the toy, and (for Study 2) generalized the function of the toy when constructing a new one. Results demonstrate that across age groups, explanation promotes causal learning and generalization, but does not improve (and in younger children, can even impair) memory for causally-irrelevant perceptual details. (shrink)

How well can we explain natural occurrences of cognitive behaviours given the theoretical frameworks available to us today? The thesis explores what has to be assumed in cognitive theory in order to provide such an explanation, in contrast to being able to predict behaviour under controlled circumstances. The behaviours considered are all of the type described as involving higher level cognition or being representation hungry. Examples are problem solving and certain types of decision-making. Three different theoretical frameworks are examined: general (...) theories such as Newell and Simon’s Universal Problem Solver, the stances sometimes referred to as “situated cognition,” particularly those that try to exclude cognitive components from their explanatory framework ; and domain-specific theories of cognition. The last category can broadly be separated into two different types of theories: those who claim that specialised cognitive processes are activated through situational features and those who claim that specialised cognitive processes are successful relative such features. The conclusion is that none of the stances are sufficient on their own. The proposed solution is instead to take both the hypothesised domain-specific process and the information it utilises into account. If we allow for “high-quality” information of a more domain-general nature, that indicates which parts of the current situation are important, or makes it possible for the individual to assume certain relations in advance, then this information can account for parts of the flexibility and stability required of the cognitive processes. Examples of such information are causal relations and epistemic information connected to social information. Since these point us to how the features of the situation at hand were produced, they help us identify relevant aspects of, and relations between different situations. By utilising high-quality information, domain-specific processes can be made broader, and less specialised, while keeping a close tie to the situations in which they are active. In this way we can begin to explain real life cognitive behaviours. (shrink)

Although storybooks are often used as pedagogical tools for conveying moral lessons to children, the ability to spontaneously extract "the moral" of a story develops relatively late. Instead, children tend to represent stories at a concrete level - one that highlights surface features and understates more abstract themes. Here we examine the role of explanation in 5- and 6-year-old children's developing ability to learn the moral of a story. Two experiments demonstrate that, relative to a control condition, prompts to explain (...) aspects of a story facilitate children's ability to override salient surface features, abstract the underlying moral, and generalize that moral to novel contexts. In some cases, generating an explanation is more effective than being explicitly told the moral of the story, as in a more traditional pedagogical exchange. These findings have implications for moral comprehension, the role of explanation in learning, and the development of abstract reasoning in early childhood. (shrink)

Explaining Attitudes develops a new account of propositional attitudes - practical realism.

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)

Norms are a pervasive yet mysterious feature of social life. In Explaining Norms, four philosophers and social scientists team up to grapple with some of the many mysteries, offering a comprehensive account of norms: what they are; how and why they emerge, persist and change; and how they work.

Explaining the behaviour of ecosystems is one of the key challenges for the biological sciences. Since 2000, new-mechanicism has been the main model to account for the nature of scientific explanation in biology. The universality of the new-mechanist view in biology has been however put into question due to the existence of explanations that account for some biological phenomena in terms of their mathematical properties (mathematical explanations). Supporters of mathematical explanation have argued that the explanation of the behaviour of ecosystems (...) is usually provided in terms of their mathematical properties, and not in mechanistic terms. They have intensively studied the explanation of the properties of ecosystems that behave following the rules of a non-random network. However, no attention has been devoted to the study of the nature of the explanation in those that form a random network. In this paper, we cover that gap by analysing the explanation of the stability behaviour of the microbiome recently elaborated by Coyte and colleagues, to determine whether it fits with the model of explanation suggested by the new-mechanist or by the defenders of mathematical explanation. Our analysis of this case study supports three theses: (1) that the explanation is not given solely in terms of mechanisms, as the new-mechanists understand the concept; (2) that the mathematical properties that describe the system play an essential explanatory role, but they do not exhaust the explanation; (3) that a non-previously identified appeal to the type of interactions that the entities in the network can exhibit, as well as their abundance, is also necessary for Coyte and colleagues’ account to be fully explanatory. From the combination of these three theses we argue for the necessity of an integrative pluralist view of the nature of behaviour explanation when this is given by appealing to the existence of a random network. (shrink)

​Imagination will remain a mystery—we will not be able to explain imagination—until we can break it into parts we already understand. Explaining Imagination is a guidebook for doing just that, where the parts are other ordinary mental states like beliefs, desires, judgments, and decisions. In different combinations and contexts, these states constitute cases of imagining. This reductive approach to imagination is at direct odds with the current orthodoxy, according to which imagination is a sui generis mental state or process—one with (...) its own inscrutable principles of operation. Explaining Imagination upends that view, showing how, on closer inspection, the imaginings at work in hypothetical reasoning, pretense, the enjoyment of fiction, and creativity are reducible to other familiar mental states—judgments, beliefs, desires, and decisions among them. Crisscrossing contemporary philosophy of mind, cognitive science, and aesthetics, Explaining Imagination argues that a clearer understanding of imagination is already well within reach. (shrink)

It is clear that lack of awareness of the consequences of an action can undermine moral responsibility and blame for these consequences. But when and how it does so is controversial. Sometimes an agent believing that the outcome might occur is excused because it seemed unlikely to her, and sometimes an agent having no idea that it would occur is nevertheless to blame. A low or zero degree of belief might seem to excuse unless the agent “should have known better”, (...) but it is unclear how to spell out this normative condition. -/- This chapter combines (a) an independently motivated account of responsibility, blame, and credit as grounded in a normal explanatory relation between agential qualities and objects of responsibility with (b) the familiar Strawsonian idea that moral blame and credit depend on the agent’s quality of will. The resulting explanatory quality of will condition on moral responsibility is then further motivated by being shown to account for the effects on moral blame and credit of justifications, as well as of excuses and undermined control in cases not involving ignorance. -/- The explanatory quality of will condition is finally applied to cases involving various degrees of lack of awareness. Though this condition itself involves no awareness requirement, it is shown how it accounts for the degrees to which lack of awareness can excuse. It is also explained how lack of awareness fails as an excuse exactly when the agent should have known better and can be blamed for not doing so. (shrink)

In the book, I argue that the mind can be explained computationally because it is itself computational—whether it engages in mental arithmetic, parses natural language, or processes the auditory signals that allow us to experience music. All these capacities arise from complex information-processing operations of the mind. By analyzing the state of the art in cognitive science, I develop an account of computational explanation used to explain the capacities in question.

This book is an expanded and revised edition of the author's critically acclaimed volume Nuts and Bolts for the Social Sciences. In twenty-six succinct chapters, Jon Elster provides an account of the nature of explanation in the social sciences. He offers an overview of key explanatory mechanisms in the social sciences, relying on hundreds of examples and drawing on a large variety of sources - psychology, behavioral economics, biology, political science, historical writings, philosophy and fiction. Written in accessible and jargon-free (...) language, Elster aims at accuracy and clarity while eschewing formal models. In a provocative conclusion, Elster defends the centrality of qualitative social sciences in a two-front war against soft and hard forms of obscurantism. (shrink)

To create fair and accountable AI and robotics, we need precise regulation and better methods to certify, explain, and audit inscrutable systems.

This chapter discusses whether Quasi-Realism gains any advantage over Robust Realism with respect to the problem of explaining supervenience. The chapter starts with a summary of what the supervenience problem is and recounts the history of expressivist thinking about supervenience: the supervenience problem was a challenge raised by expressivist Robust Realists, with the idea that expressivism had an excellent explanation of the phenomenon and realism had none. The chapter then contrasts Quasi-Realism and Robust Realism in order to bring the big (...) problem out in the open, namely that Quasi-Realists have not provided any explanation at all for the phenomenon of supervenience. In this respect they are, it appears, in the same boat as Robust Realists. The chapter maps out the possible paths for Quasi-Realists to travel in pursuit of Quasi-Explananda, and gives some reasons to be optimistic. The chapter ends with a suggestive analogy and a synopsis of the state of the dialectic. (shrink)

In this volume, first published in 1983, Jon Elster approaches the study of technical change from an epistemological perspective.

Carl F. Craver investigates what we are doing when we use neuroscience to explain what's going on in the brain. When does an explanation succeed and when does it fail? Craver offers explicit standards for successful explanation of the workings of the brain, on the basis of a systematic view about what neuroscientific explanations are.

Many artificial intelligence (AI) systems currently used for decision-making are opaque, i.e., the internal factors that determine their decisions are not fully known to people due to the systems’ computational complexity. In response to this problem, several researchers have argued that human decision-making is equally opaque and since simplifying, reason-giving explanations (rather than exhaustive causal accounts) of a decision are typically viewed as sufficient in the human case, the same should hold for algorithmic decision-making. Here, I contend that this argument (...) overlooks that human decision-making is sometimes significantly more transparent and trustworthy than algorithmic decision-making. This is because when people explain their decisions by giving reasons for them, this frequently prompts those giving the reasons to govern or regulate themselves so as to think and act in ways that confirm their reason reports. AI explanation systems lack this self-regulative feature. Overlooking it when comparing algorithmic and human decision-making can result in underestimations of the transparency of human decision-making and in the development of explainable AI that may mislead people by activating generally warranted beliefs about the regulative dimension of reason-giving. (shrink)

Recently, a number of authors have suggested that the epistemic condition on moral responsibility makes blameworthiness much less common than we ordinarily suppose, and much harder to identify. This paper argues that such epistemically based responsibility skepticism is mistaken. Section 2 sketches a general account of moral responsibility, building on the Strawsonian idea that blame and credit relates to the agent’s quality of will. Section 3 explains how this account deals with central cases that motivate epistemic skepticism and how it (...) avoids some objections to quality of will accounts recently raised by Gideon Rosen. But an intuitive worry brought out by these objections remains. Section 4 spells out this remaining worry and argues that, like traditional metaphysical responsibility skepticism, it has its source in a non-standard explanatory perspective on action, suggesting that strategies for explaining away the intuitive pull of traditional skepticism are applicable in this case too. (shrink)

Explaining Value is a selection of the best of Gilbert Harman's shorter writings in moral philosophy. The thirteen essays are divided into four sections, which focus in turn on moral relativism, values and valuing, character traits and virtue ethics, and ways of explaining aspects of morality. Harman's distinctive approach to moral philosophy has provoked much interest; this volume offers a fascinating conspectus of his most important work in the area.

There is general consensus that explainable artificial intelligence is valuable, but there is significant divergence when we try to articulate why, exactly, it is desirable. This question must be distinguished from two other kinds of questions asked in the XAI literature that are sometimes asked and addressed simultaneously. The first and most obvious is the ‘how’ question—some version of: ‘how do we develop technical strategies to achieve XAI?’ Another question is specifying what kind of explanation is worth having in (...) the first place. As difficult and important as the challenges are in answering these questions, they are distinct from a third question: why do we want XAI at all? There is vast literature on this question as well, but I wish to explore a different kind of answer. The most obvious way to answer this question is by describing a desirable outcome that would likely be achieved with the right kind of explanation, which would make the explanation valuable instrumentally. That is, XAI is desirable to attain some other value, such as fairness, trust, accountability, or governance. This family of arguments is obviously important, but I argue that explanations are also intrinsically valuable, because unexplainable systems can be dehumanizing. I argue that there are at least three independently valid versions of this kind of argument: an argument from participation, from knowledge, and from actualization. Each of these arguments that XAI is intrinsically valuable is independently compelling, in addition to the more obvious instrumental benefits of XAI. (shrink)

In some cases, there appears to be an asymmetry in the evidential value of statistical and more individualized evidence. For example, while I may accept that Alex is guilty based on eyewitness testimony that is 80% likely to be accurate, it does not seem permissible to do so based on the fact that 80% of a group that Alex is a member of are guilty. In this paper I suggest that rather than reflecting a deep defect in statistical evidence, this (...) asymmetry might arise from a general constraint on rational inquiry. Plausibly the degree of evidential support needed to justify taking a proposition to be true depends on the stakes of error. While relying on statistical evidence plausibly raises the stakes by introducing new kinds of risk to members of the reference class, paradigmatically `individualized' evidence---evidence tracing back to A's voluntary behavior---can lower the stakes. The net result explains the apparent evidential asymmetry without positing a deep difference in the brute justificatory power of different types of evidence. (shrink)

Little, Brown, 1992 Review by Glenn Branch on Jul 5th 1999 Volume: 3, Number: 27.

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)

Over the last 20 years the comparator model for delusions of control has received considerable support in terms of empirical studies. However, the original version clearly needs to be replaced by a model with a much greater degree of sophistication and specificity. Future developments are likely to involve the specification of the role of dopamine in the model and a generalisation of its explanatory power to the whole range of positive symptoms. However, we will still need to explain why symptoms (...) can be so variable and we still do not understand the origin of the most mysterious symptom of all: thought insertion. (shrink)

Normativity is what gives reasons their force, makes words meaningful, and makes rules and laws binding. It is present whenever we use such terms as ‘correct,' ‘ought,' ‘must,' and the language of obligation, responsibility, and logical compulsion. Yet normativists, the philosophers committed to this idea, admit that the idea of a non-causal normative realm and a body of normative objects is spooky. Explaining the Normative is the first systematic, historically grounded critique of normativism. It identifies the standard normativist pattern of (...) argument, and shows how this pattern depends on circularities, assumptions about the unique correctness of preferred descriptions, problematic transcendental arguments, and regress arguments that end in mysteries. The book considers in detail a paradigm case: legal normativity as constructed by Hans Kelsen. This case exemplifies the problems with normativist arguments. But it also shows how normativism was constructed as an alternative to ordinary social science explanation. The normativist argument is that social science explanations themselves are forced to rely on normative conceptsÑminimally, on normative rationality and on a normative view of ‘concepts' themselves. Empathic understanding of the reasoning and meanings of others, however, can solve the regress problems about meaning and rationality that are central to the appeal of normativism. This account has no need for a parallel normative world, and has a surprising and revealing lineage in the history of philosophy, as well as a basis in neuroscience. (shrink)

Modern explainable AI methods remain far from providing human-like answers to ‘why’ questions, let alone those that satisfactorily agree with human-level understanding. Instead, the results that such methods provide boil down to sets of causal attributions. Currently, the choice of accepted attributions rests largely, if not solely, on the explainee’s understanding of the quality of explanations. The paper argues that such decisions may be transferred from a human to an XAI agent, provided that its machine-learning algorithms perform genuinely abductive (...) inferences. The paper outlines the key predicament in the current inductive paradigm of ML and the associated XAI techniques, and sketches the desiderata for a truly participatory, second-generation XAI, which is endowed with abduction. (shrink)

There has been much debate recently on the question whether essence can explain modality. Here, I examine two routes to an essentialist account of modality. The first is Hale's argument for the necessity of essence, which I will argue is — notwithstanding recent attempted defences of it — invalid by its very structure. The second is the proposal that it is essential to essential truth that it is necessary. After offering three possible versions of the view, I will argue that (...) each fails to provide a metaphysical explanation of necessity in terms of essence. (shrink)

The physical realm is causally closed, according to physicalists like me. But why is it causally closed, what metaphysically explains causal closure? I argue that reductive physicalists are committed to one explanation of causal closure to the exclusion of any independent explanation, and that as a result, they must give up on using a causal argument to attack mind–body dualism. Reductive physicalists should view dualism in much the way that we view the hypothesis that unicorns exist, or that the Kansas (...) City Royals won the 2003 World Series: false, but not objectionable in any distinctively causal way. My argument turns on connections between explanation, counterfactuals, and inductive confirmation. (shrink)

The debate between compatibilists and incompatibilists depends in large part on what ordinary people mean by ‘free will’, a matter on which previous experimental philosophy studies have yielded conflicting results. In Nahmias, Morris, Nadelhoffer, and Turner (2005, 2006), most participants judged that agents in deterministic scenarios could have free will and be morally responsible. Nichols and Knobe (2007), though, suggest that these apparent compatibilist responses are performance errors produced by using concrete scenarios, and that their abstract scenarios reveal the folk (...) theory of free will for what it actually is—incompatibilist. Here, we argue that the results of two new studies suggest just the opposite. Most participants only give apparent incompatibilist judgments when they mistakenly interpret determinism to imply that agents’ mental states are bypassed in the causal chains that lead to their behavior. Determinism does not entail bypassing, so these responses do not reflect genuine incompatibilist intuitions. When participants understand what determinism does mean, the vast majority take it to be compatible with free will. Further results indicate that most people’s concepts of choice and the ability to do otherwise do not commit them to incompatibilism, either, putting pressure on incompatibilist arguments that rely on transfer principles, such as the Consequence Argument. We discuss the implications of these findings for philosophical debates about free will, and suggest that incompatibilism appears to be either false, or else a thesis about something other than what most people mean by ‘free will’. (shrink)

Why does social injustice exist? What role, if any, do implicit biases play in the perpetuation of social inequalities? Individualistic approaches to these questions explain social injustice as the result of individuals’ preferences, beliefs, and choices. For example, they explain racial injustice as the result of individuals acting on racial stereotypes and prejudices. In contrast, structural approaches explain social injustice in terms of beyond-the-individual features, including laws, institutions, city layouts, and social norms. Often these two approaches are seen as competitors. (...) Framing them as competitors suggests that only one approach can win and that the loser offers worse explanations of injustice. In this essay, we explore each approach and compare them. Using implicit bias as an example, we argue that the relationship between individualistic and structural approaches is more complicated than it may first seem. Moreover, we contend that each approach has its place in analyses of injustice and raise the possibility that they can work together—synergistically—to produce deeper explanations of social injustice. If so, the approaches may be complementary, rather than competing. (shrink)

Legal theorists tend to focus on disagreement over the law, and yet a theory of law should also explain why lawyers and judges agree on the law as often as they do. To that end, this article first pins down a precise sense in which there can be pervasive agreement on the law. It then argues that such agreement obtains in the United States and likely in many other jurisdictions as well. Finally, it contends that Hartian Positivism offers a straightforward (...) explanation of this phenomenon and, indeed, a better explanation than rival theories, like Ronald Dworkin’s Law as Integrity or Scott Shapiro’s Planning Theory. Given the ubiquity of legal agreement, Hartian Positivism’s explanatory superiority in this regard is a strong, albeit still not dispositive, consideration in its favour. The broader lesson is that theories of law should give greater priority to explaining legal agreement than many theories now do. (shrink)

From time to time we explain what people do by referring to their habits. We explain somebody’s putting the kettle on in the morning as done through “force of habit”. We explain somebody’s missing a turning by saying that she carried straight on “out of habit”. And we explain somebody’s biting her nails as a manifestation of “a bad habit”. These are all examples of what will be referred to here as habit explanations. Roughly speaking, they explain by referring to (...) a pattern of a particular kind of behaviour which is regularly performed in characteristic circumstances, and has become automatic for that agent due to this repetition. (shrink)

The increased presence of medical AI in clinical use raises the ethical question which standard of explainability is required for an acceptable and responsible implementation of AI-based applications in medical contexts. In this paper, we elaborate on the emerging debate surrounding the standards of explainability for medical AI. For this, we first distinguish several goods explainability is usually considered to contribute to the use of AI in general, and medical AI in specific. Second, we propose to understand the value of (...) explainability relative to other available norms of explainable decision-making. Third, in pointing out that we usually accept heuristics and uses of bounded rationality for medical decision-making by physicians, we argue that the explainability of medical decisions should not be measured against an idealized diagnostic process, but according to practical considerations. We conclude, fourth, to resolve the issue of explainability-standards by relocating the issue to the AI’s certifiability and interpretability. (shrink)

The operations of deep networks are widely acknowledged to be inscrutable. The growing field of Explainable AI has emerged in direct response to this problem. However, owing to the nature of the opacity in question, XAI has been forced to prioritise interpretability at the expense of completeness, and even realism, so that its explanations are frequently interpretable without being underpinned by more comprehensive explanations faithful to the way a network computes its predictions. While this has been taken to be (...) a shortcoming of the field of XAI, I argue that it is broadly the right approach to the problem. (shrink)

Chaotic dynamics has been hailed as the third great scientific revolution in physics this century, comparable to relativity and quantum mechanics. In this book, Peter Smith takes a cool, critical look at such claims. He cuts through the hype and rhetoric by explaining some of the basic mathematical ideas in a clear and accessible way, and by carefully discussing the methodological issues which arise. In particular, he explores the new kinds of explanation of empirical phenomena which modern dynamics can deliver. (...) Explaining Chaos will be compulsory reading for philosophers of science and for anyone who has wondered about the conceptual foundations of chaos theory. (shrink)

There is a current debate about if, and in what sense, machine learning systems used in the medical context need to be explainable. Those arguing in favor contend these systems require post hoc explanations for each individual decision to increase trust and ensure accurate diagnoses. Those arguing against suggest the high accuracy and reliability of the systems is sufficient for providing epistemic justified beliefs without the need for explaining each individual decision. But, as we show, both solutions have limitations—and (...) it is unclear either address the epistemic worries of the medical professionals using these systems. We argue these systems do require an explanation, but an institutional explanation. These types of explanations provide the reasons why the medical professional should rely on the system in practice—that is, they focus on trying to address the epistemic concerns of those using the system in specific contexts and specific occasions. But ensuring that these institutional explanations are fit for purpose means ensuring the institutions designing and deploying these systems are transparent about the assumptions baked into the system. This requires coordination with experts and end-users concerning how it will function in the field, the metrics used to evaluate its accuracy, and the procedures for auditing the system to prevent biases and failures from going unaddressed. We contend this broader explanation is necessary for either post hoc explanations or accuracy scores to be epistemically meaningful to the medical professional, making it possible for them to rely on these systems as effective and useful tools in their practices. (shrink)