The pragmatic character of the Dutch book argument makes it unsuitable as an "epistemic" justification for the fundamental probabilist dogma that rational partial beliefs must conform to the axioms of probability. To secure an appropriately epistemic justification for this conclusion, one must explain what it means for a system of partial beliefs to accurately represent the state of the world, and then show that partial beliefs that violate the laws of probability are invariably less accurate than they could be otherwise. (...) The first task can be accomplished once we realize that the accuracy of systems of partial beliefs can be measured on a gradational scale that satisfies a small set of formal constraints, each of which has a sound epistemic motivation. When accuracy is measured in this way it can be shown that any system of degrees of belief that violates the axioms of probability can be replaced by an alternative system that obeys the axioms and yet is more accurate in every possible world. Since epistemically rational agents must strive to hold accurate beliefs, this establishes conformity with the axioms of probability as a norm of epistemic rationality whatever its prudential merits or defects might be. (shrink)

This paper advances a reductive semantics for ‘ought’ and a naturalistic theory of normativity. It gives a unified analysis of predictive, instrumental, and categorical uses of ‘ought’: the predictive ‘ought’ is basic, and is interpreted in terms of probability. Instrumental ‘oughts’ are analyzed as predictive ‘oughts’ occurring under an ‘in order that’ modifer (the end-relational theory). The theory is then extended to categorical uses of ‘ought’: it is argued that they are special rhetorical uses of the instrumental ‘ought’. Plausible conversational (...) principles explain how this end-relational ‘ought’ can perform the expressive functions of the moral ‘ought’. The notion of an ‘ought-simpliciter’ is also discussed. (shrink)

"This book is the first of two volumes on belief and counterfactuals. It consists of six of a total of eleven chapters. The first volume is concerned primarily with questions in epistemology and is expository in parts. Among others, it provides an accessible introduction to belief revision and ranking theory. Ranking theory specifies how conditional beliefs should behave. It does not tell us why they should do so nor what they are. This book fills these two gaps. The consistency argument (...) tells us why conditional beliefs should obey the laws of ranking theory by showing them to be the means to attaining the end of holding true and informative beliefs. The conditional theory of conditional belief tells us what conditional beliefs are by specifying their nature in terms of non-conditional belief and counterfactuals. In addition, the book contains several novel arguments, accounts, and applications. These include an argument for the thesis that there are only hypothetical imperatives and no categorical imperatives; an account of the instrumentalist understanding of normativity, or rationality, according to which one ought to take the means to one's ends; as well as solutions to the problems of conceptual belief change, logical learning, and learning conditionals. A distinctive feature of the book is its unifying methodological approach: means-end philosophy. Means-end philosophy takes serious that philosophy is a normative discipline, and that philosophical problems are entangled with each other. It also explains the importance of logic to philosophy, without being a technical theory itself"--. (shrink)

Scientists imagine for epistemic reasons, and these imaginings can be better or worse. But what does it mean for an imagining to be epistemically better or worse? There are at least three metaepistemological frameworks that present different answers to this question: epistemological consequentialism, deontic epistemology, and virtue epistemology. This paper presents empirical evidence that scientists adopt each of these different epistemic frameworks with respect to imagination, but argues that the way they do this is best explained if scientists are fundamentally (...) epistemic consequentialists about imagination. (shrink)

This paper argues that a notion of statistical explanation, based on Salmon’s statistical relevance model, can help us better understand deep neural networks. It is proved that homogeneous partitions, the core notion of Salmon’s model, are equivalent to minimal sufficient statistics, an important notion from statistical inference. This establishes a link to deep neural networks via the so-called Information Bottleneck method, an information-theoretic framework, according to which deep neural networks implicitly solve an optimization problem that generalizes minimal sufficient statistics. The (...) resulting notion of statistical explanation is general, mathematical, and subcausal. (shrink)

Should decision-making algorithms be held to higher standards of transparency than human beings? The way we answer this question directly impacts what we demand from explainable algorithms, how we govern them via regulatory proposals, and how explainable algorithms may help resolve the social problems associated with decision making supported by artificial intelligence. Some argue that algorithms and humans should be held to the same standards of transparency and that a double standard of transparency is hardly justified. We give two arguments (...) to the contrary and specify two kinds of situations for which higher standards of transparency are required from algorithmic decisions as compared to humans. Our arguments have direct implications on the demands from explainable algorithms in decision-making contexts such as automated transportation. (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)

We argue that artificial networks are explainable and offer a novel theory of interpretability. Two sets of conceptual questions are prominent in theoretical engagements with artificial neural networks, especially in the context of medical artificial intelligence: Are networks explainable, and if so, what does it mean to explain the output of a network? And what does it mean for a network to be interpretable? We argue that accounts of “explanation” tailored specifically to neural networks have ineffectively reinvented the wheel. In (...) response to, we show how four familiar accounts of explanation apply to neural networks as they would to any scientific phenomenon. We diagnose the confusion about explaining neural networks within the machine learning literature as an equivocation on “explainability,” “understandability” and “interpretability.” To remedy this, we distinguish between these notions, and answer by offering a theory and typology of interpretation in machine learning. Interpretation is something one does to an explanation with the aim of producing another, more understandable, explanation. As with explanation, there are various concepts and methods involved in interpretation: Total or Partial, Global or Local, and Approximative or Isomorphic. Our account of “interpretability” is consistent with uses in the machine learning literature, in keeping with the philosophy of explanation and understanding, and pays special attention to medical artificial intelligence systems. (shrink)

Scientists depend on complex computational systems that are often ineliminably opaque, to the detriment of our ability to give scientific explanations and detect artifacts. Some philosophers have s...

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)

We are sceptical of concerns over the opacity of algorithmic decision tools. While transparency and explainability are certainly important desiderata in algorithmic governance, we worry that automated decision-making is being held to an unrealistically high standard, possibly owing to an unrealistically high estimate of the degree of transparency attainable from human decision-makers. In this paper, we review evidence demonstrating that much human decision-making is fraught with transparency problems, show in what respects AI fares little worse or better and argue that (...) at least some regulatory proposals for explainable AI could end up setting the bar higher than is necessary or indeed helpful. The demands of practical reason require the justification of action to be pitched at the level of practical reason. Decision tools that support or supplant practical reasoning should not be expected to aim higher than this. We cast this desideratum in terms of Daniel Dennett’s theory of the “intentional stance” and argue that since the justification of action for human purposes takes the form of intentional stance explanation, the justification of algorithmic decisions should take the same form. In practice, this means that the sorts of explanations for algorithmic decisions that are analogous to intentional stance explanations should be preferred over ones that aim at the architectural innards of a decision tool. (shrink)

Philosophers have established that certain ethically important val- ues are modally robust in the sense that they systematically deliver correlative benefits across a range of counterfactual scenarios. In this paper, we contend that recourse – the systematic process of reversing unfavorable decisions by algorithms and bureaucracies across a range of counterfactual scenarios – is such a modally ro- bust good. In particular, we argue that two essential components of a good life – temporally extended agency and trust – are under- (...) written by recourse. We critique existing approaches to the conceptualization, op- erationalization and implementation of recourse. Based on these criticisms, we suggest a revised approach to recourse and give ex- amples of how it might be implemented – especially for those who are least well off. (shrink)

The usefulness of machine learning algorithms has led to their widespread adoption prior to the development of a conceptual framework for making sense of them. One common response to this situation is to say that machine learning suffers from a “black box problem.” That is, machine learning algorithms are “opaque” to human users, failing to be “interpretable” or “explicable” in terms that would render categorization procedures “understandable.” The purpose of this paper is to challenge the widespread agreement about the existence (...) and importance of a black box problem. The first section argues that “interpretability” and cognates lack precise meanings when applied to algorithms. This makes the concepts difficult to use when trying to solve the problems that have motivated the call for interpretability. Furthermore, since there is no adequate account of the concepts themselves, it is not possible to assess whether particular technical features supply formal definitions of those concepts. The second section argues that there are ways of being a responsible user of these algorithms that do not require interpretability. In many cases in which a black box problem is cited, interpretability is a means to a further end such as justification or non-discrimination. Since addressing these problems need not involve something that looks like an “interpretation” of an algorithm, the focus on interpretability artificially constrains the solution space by characterizing one possible solution as the problem itself. Where possible, discussion should be reformulated in terms of the ends of interpretability. (shrink)

Although machine learning has been successful in recent years and is increasingly being deployed in the sciences, enterprises or administrations, it has rarely been discussed in philosophy beyond the philosophy of mathematics and machine learning. The present contribution addresses the resulting lack of conceptual tools for an epistemological discussion of machine learning by conceiving of deep learning networks as ‘judging machines’ and using the Kantian analysis of judgments for specifying the type of judgment they are capable of. At the center (...) of the argument is the fact that the functionality of deep learning networks is established by training and cannot be explained and justified by reference to a predefined rule-based procedure. Instead, the computational process of a deep learning network is barely explainable and needs further justification, as is shown in reference to the current research literature. Thus, it requires a new form of justification, that is to be specified with the help of Kant’s epistemology. (shrink)

Epistemic consequentialists maintain that the epistemically right (e.g., the justified) is to be understood in terms of conduciveness to the epistemic good (e.g., true belief). Given the wide variety of epistemological approaches that assume some form of epistemic consequentialism, and the controversies surrounding consequentialism in ethics, it is surprising that epistemic consequentialism remains largely uncontested. However, in a recent paper, Selim Berker has provided arguments that allegedly lead to a ‘rejection’ of epistemic consequentialism. In the present paper, it is shown (...) that reliabilism—the most prominent form of epistemic consequentialism, and one of Berker’s main targets—survives Berker’s arguments unscathed. (shrink)

It is argued that it is not sufficient to consider only the sentences included in the explanans and explanandum when determining whether they constitute an explanation, but these sentences must always be evaluated relative to a knowledge situation. The central criterion on an explanation is that the explanans in a non-trivial way increases the belief value of the explanandum, where the belief value of a sentence is determined from the given knowledge situation. The outlined theory of explanations is applied to (...) some well-known examples and is also compared to other theories of explanation. (shrink)

The philosophical theory of scientific explanation proposed here involves a radically new treatment of causality that accords with the pervasively statistical character of contemporary science. Wesley C. Salmon describes three fundamental conceptions of scientific explanation--the epistemic, modal, and ontic. He argues that the prevailing view (a version of the epistemic conception) is untenable and that the modal conception is scientifically out-dated. Significantly revising aspects of his earlier work, he defends a causal/mechanical theory that is a version of the ontic conception. (...) Professor Salmon's theory furnishes a robust argument for scientific realism akin to the argument that convinced twentieth-century physical scientists of the existence of atoms and molecules. To do justice to such notions as irreducibly statistical laws and statistical explanation, he offers a novel account of physical randomness. The transition from the "reviewed view" of scientific explanation (that explanations are arguments) to the causal/mechanical model requires fundamental rethinking of basic explanatory concepts. (shrink)

Scientific explanations must bear the proper relationship to the world: they must depict what, out in the world, is responsible for the explanandum. But explanations must also bear the proper relationship to their audience: they must be able to create human understanding. With few exceptions, philosophical accounts of explanation either ignore entirely the relationship between explanations and their audience or else demote this consideration to an ancillary role. In contrast, I argue that considering an explanation’s communicative role is crucial to (...) any satisfactory account of explanation. (shrink)

This paper describes the corner-stones of a means-ends approach to the philosophy of inductive inference. I begin with a fallibilist ideal of convergence to the truth in the long run, or in the 'limit of inquiry'. I determine which methods are optimal for attaining additional epistemic aims (notably fast and steady convergence to the truth). Means-ends vindications of (a version of) Occam's Razor and the natural generalizations in a Goodmanian Riddle of Induction illustrate the power of this approach. The paper (...) establishes a hierarchy of means-ends notions of empirical success, and discusses a number of issues, results and applications of means-ends epistemology. (shrink)

Explainable artificial intelligence (XAI) is an emerging, multidisciplinary field of research that seeks to develop methods and tools for making AI systems more explainable or interpretable. XAI researchers increasingly recognise explainability as a context-, audience- and purpose-sensitive phenomenon, rather than a single well-defined property that can be directly measured and optimised. However, since there is currently no overarching definition of explainability, this poses a risk of miscommunication between the many different researchers within this multidisciplinary space. This is the problem we (...) seek to address in this paper. We outline a framework, called Explanatory Pragmatism, which we argue has two attractive features. First, it allows us to conceptualise explainability in explicitly context-, audience- and purpose-relative terms, while retaining a unified underlying definition of explainability. Second, it makes visible any normative disagreements that may underpin conflicting claims about explainability regarding the purposes for which explanations are sought. Third, it allows us to distinguish several dimensions of AI explainability. We illustrate this framework by applying it to a case study involving a machine learning model for predicting whether patients suffering disorders of consciousness were likely to recover consciousness. (shrink)

Issues concerning scientific explanation have been a focus of philosophical attention from Pre- Socratic times through the modern period. However, recent discussion really begins with the development of the Deductive-Nomological (DN) model. This model has had many advocates (including Popper 1935, 1959, Braithwaite 1953, Gardiner, 1959, Nagel 1961) but unquestionably the most detailed and influential statement is due to Carl Hempel (Hempel 1942, 1965, and Hempel & Oppenheim 1948). These papers and the reaction to them have structured subsequent discussion concerning (...) scientific explanation to an extraordinary degree. After some general remarks by way of background and orientation (Section 1), this entry describes the DN model and its extensions, and then turns to some well-known objections (Section 2). It next describes a variety of subsequent attempts to develop alternative models of explanation, including Wesley Salmon's Statistical Relevance (Section 3) and Causal Mechanical (Section 4) models and the Unificationist models due to Michael Friedman and Philip Kitcher (Section 5). Section 6 provides a summary and discusses directions for future work. (shrink)

I defend the claim that understanding is the goal of explanation against various persistent criticisms, especially the criticism that understanding is not truth-connected in the appropriate way, and hence is a merely psychological state. Part of the reason why understanding has been dismissed as the goal of explanation, I suggest, is because the psychological dimension of the goal of explanation has itself been almost entirely neglected. In turn, the psychological dimension of understanding—the Aha! experience, the sense that a certain explanation (...) “feels right”, and so on—has been conspicuously overemphasized. I try to correct for both of these exaggerations. Just as the goal of explanation includes a richer psychological—including phenomenological—dimension than is generally acknowledged, so too understanding has a stronger truth connection than is generally acknowledged.Keywords: Understanding; Explanation; Philosophy of science; Epistemology. (shrink)

Scientific methods are heuristic in nature. Heuristics are simplifying, incomplete, underdetermined and fallible problem-solving rules that can nevertheless serve certain goals in certain contexts better than truth-preserving algorithms. Because of their goal- and context-dependence, a framework is needed for systematic choosing between them. This is the domain of scientific methodology. Such a methodology, I argue, relies on a form of instrumental rationality. Three challenges to such an instrumentalist account are addressed. First, some authors have argued that the rational choice of (...) at least some methods, namely those supporting belief formation, is not goal-dependent. Second, some authors have observed that some method choices seem intuitively rational, even though relevant goals are lacking. Thirdly, some authors have argued that instrumental rationality itself depends on a goal-independent form of rationality. It is the heuristic nature of scientific methods that affords me the arguments against these challenges. This heuristic-instrumentalist account provides the means-ends analysis needed to evaluate heuristic method choice. The paper thus offers the conceptual basis for a scientific methodology that is both compatible with the heuristic nature of actual scientific practice and also normatively relevant for assessing method choice. (shrink)

Epistemic consequentialists maintain that the epistemically right (e.g., the justified) is to be understood in terms of conduciveness to the epistemic good (e.g., true belief). Given the wide variety of epistemological approaches that assume some form of epistemic consequentialism, and the controversies surrounding consequentialism in ethics, it is surprising that epistemic consequentialism remains largely uncontested. However, in a recent paper, Selim Berker has provided arguments that allegedly lead to a ‘rejection’ of epistemic consequentialism. In the present paper, it is shown (...) that reliabilism—the most prominent form of epistemic consequentialism, and one of Berker’s main targets—survives Berker’s arguments unscathed. (shrink)

This paper explores the connective ‘because’, based on the idea that ‘_C_ because _A_’ implies the acceptance/truth of the antecedent _A_ as well as of the consequent _C_, and additionally that the antecedent makes a difference for the consequent. To capture this idea of difference-making a ‘relevantized’ version of the Ramsey Test for conditionals is employed that takes the antecedent to be relevant to the consequent in the following sense: a conditional is true/accepted in a state \(\sigma \) just in (...) case (i) the consequent is true/accepted when \(\sigma \) is revised by the antecedent _and_ (ii) the consequent fails to be true/accepted when \(\sigma \) is revised by the antecedent’s negation. To extend this to a semantics for ‘because’, we add that (iii) the antecedent and (iv) the consequent are accepted/true in the state \(\sigma \). We get metaphysical or doxastic interpretations of these clauses, depending on what we mean by a model and a state. We introduce several semantics known from suppositional conditionals, which we reinterpret for difference-making conditionals and ‘because’. We present a minimal logic for ‘because’ sentences and show how it can be extended in ways that parallel the hierarchy of extensions of the logic of suppositional conditionals. We establish correspondence results between axioms for ‘because’ and properties of states, and prove that the specified logics are sound with respect to the semantics. (shrink)

This paper addresses the logical foundations of goal-regression planning in autonomous rational agents. It focuses mainly on three problems. The first is that goals and subgoals will often be conjunctions, and to apply goal-regression planning to a conjunction we usually have to plan separately for the conjuncts and then combine the resulting subplans. A logical problem arises from the fact that the subplans may destructively interfere with each other. This problem has been partially solved in the AI literature (e.g., in (...) SNLP and UCPOP), but the solutions proposed there work only when a restrictive assumption is satisfied. This assumption pertains to the computability of threats. It is argued that this assumption may fail for an autonomous rational agent operating in a complex environment. Relaxing this assumption leads to a theory of defeasible planning. The theory is formulated precisely and an implementation in the OSCAR architecture is discussed. (shrink)