This article considers the issue of opacity as a problem for socially consequential mechanisms of classification and ranking, such as spam filters, credit card fraud detection, search engines, news trends, market segmentation and advertising, insurance or loan qualification, and credit scoring. These mechanisms of classification all frequently rely on computational algorithms, and in many cases on machine learning algorithms to do this work. In this article, I draw a distinction between three forms of opacity: opacity as intentional corporate or state (...) secrecy, opacity as technical illiteracy, and an opacity that arises from the characteristics of machine learning algorithms and the scale required to apply them usefully. The analysis in this article gets inside the algorithms themselves. I cite existing literatures in computer science, known industry practices, and do some testing and manipulation of code as a form of lightweight code audit. I argue that recognizing the distinct forms of opacity that may be coming into play in a given application is a key to determining which of a variety of technical and non-technical solutions could help to prevent harm. (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)

The ever-increasing application of algorithms to decision-making in a range of social contexts has prompted demands for algorithmic accountability. Accountable decision-makers must provide their decision-subjects with justifications for their automated system’s outputs, but what kinds of broader principles should we expect such justifications to appeal to? Drawing from political philosophy, I present an account of algorithmic accountability in terms of the democratic ideal of ‘public reason’. I argue that situating demands for algorithmic accountability within this justificatory framework enables us to (...) better articulate their purpose and assess the adequacy of efforts toward them. (shrink)

It is now part and parcel of the official philosophical wisdom that models are essential to the acquisition and organisation of scientific knowledge. It is also generally accepted that most models represent their target systems in one way or another. But what does it mean for a model to represent its target system? I begin by introducing three conundrums that a theory of scientific representation has to come to terms with and then address the question of whether the semantic view (...) of theories, which is the currently most widely accepted account of theories and models, provides us with adequate answers to these questions. After having argued in some detail that it does not, I conclude by pointing out in what direction a tenable account of scientific representation might be sought. (shrink)

According to a causal model of explanation, we explain phenomena by giving their causes or, where the phenomena are themselves causal regularities, we explain them by giving a mechanism linking cause and effect. If we explain why smoking causes cancer, we do not give the cause of this causal connection, but we do give the causal mechanism that makes it. The claim that to explain is to give a cause is not only natural and plausible, but it also avoids many (...) of the objections to other accounts of explanation, such as the views that to explain is to give a reason to believe the phenomenon occurred, to somehow make the phenomenon familiar, or to give a Deductive-Nomological argument. Unlike the reason for belief account, a causal model makes a clear distinction between understanding why a phenomenon occurs and merely knowing that it does, and the model does so in a way that makes understanding unmysterious and objective. Understanding is not some sort of super-knowledge, but simply more knowledge: knowledge of the phenomenon and knowledge of its causal history. A causal model makes it clear how something can explain without itself being explained, and so avoids the regress of whys, since we can know a phenomenon's cause without knowing the cause of the cause. It also accounts for legitimate self-evidencing explanations, explanations where the phenomenon is an essential part of the evidence that the explanation is correct, so the explanation can not supply a non-circular reason for believing the phenomenon occurred. There is no barrier to knowing a cause through its effects and also knowing that it is their cause. The speed of recession of a star explains its observed red-shift, even though the shift is an essential part of the evidence for its speed of recession. The model also avoids the most serious objection to the familiarity view, which is that some phenomena are familiar yet not understood, since a phenomenon can be perfectly familiar, such as the blueness of the sky or the fact that the same side of the moon always faces the earth, even if we do not know its cause. Finally, a causal model avoids many of the objections to the Deductive-Nomological model. Ordinary explanations do not have to meet the requirements of the Deductive-Nomological model, because one does not need to give a law to give a cause, and one does not need to know a law to have good reason to believe that a cause is a cause. As for the notorious over-permissiveness of the Deductive-Nomological model, the reason recession explains red-shift but not conversely, is simply that causes explain effects but not conversely, and the reason a conjunction of laws does not explain its conjuncts is that conjunctions do not cause their conjuncts. (shrink)

This book introduces readers to the topic of explanation. The insights of Plato, Aristotle, J.S. Mill and Carl Hempel are examined, and are used to argue against the view that explanation is merely a problem for the philosophy of science. Having established its importance for understanding knowledge in general, the book concludes with a bold and original explanation of explanation.

A major contribution to contemporary social theory. Not only does it provide a compelling critique of some of the main perspectives in 20th century philosophy and social science, but it also presents a systematic synthesis of the many themse which have preoccupied Habermas for thirty years. --Times Literary Supplement.

This article compares causal and constitutive explanation. While scientific inquiry usually addresses both causal and constitutive questions, making the distinction is crucial for a detailed understanding of scientific questions and their interrelations. These explanations have different kinds of explananda and they track different sorts of dependencies. Constitutive explanations do not address events or behaviors, but causal capacities. While there are some interesting relations between building and causal manipulation, causation and constitution are not to be confused. Constitution is a synchronous and (...) asymmetric relation between relata that cannot be conceived as independent existences. However, despite their metaphysical differences, the same key ideas about explanation largely apply to both. Causal and constitutive explanations face similar challenges (such as the problems of relevance and explanatory regress) and both are in the business of mapping networks of counterfactual dependence—i.e. mechanisms—although the relevant counterfactuals are of a different sort. In the final section the issue of developmental explanation is discussed. It is argued that developmental explanations deserve their own place in taxonomy of explanations, although ultimately developmental dependencies can be analyzed as combinations of causal and constitutive dependencies. Hence, causal and constitutive explanation are distinct, but not always completely separate forms of explanation. (shrink)

On the received view, counterfactuals are analysed using the concept of closeness between possible worlds: the counterfactual 'If it had been the case that p, then it would have been the case that q' is true at a world w just in case q is true at all the possible p-worlds closest to w. The degree of closeness between two worlds is usually thought to be determined by weighting different respects of similarity between them. The question I consider in the (...) paper is which weights attach to different respects of similarity. I start by considering Lewis's answer to the question and argue against it by presenting several counterexamples. I use the same examples to motivate a general principle about closeness: if a fact obtains in both of two worlds, then this similarity is relevant to the closeness between them if and only if the fact has the same explanation in the two worlds. I use this principle and some ideas of Lewis's to formulate a general account of counterfactuals, and I argue that this account can explain the asymmetry of counterfactual dependence. The paper concludes with a discussion of some examples that cannot be accommodated by the present version of the account and therefore necessitate further work on the details. (shrink)

This essay warns of eroding accountability in computerized societies. It argues that assumptions about computing and features of situations in which computers are produced create barriers to accountability. Drawing on philosophical analyses of moral blame and responsibility, four barriers are identified: 1) the problem of many hands, 2) the problem of bugs, 3) blaming the computer, and 4) software ownership without liability. The paper concludes with ideas on how to reverse this trend.

Explanation-Aware Computing: Papers from the 2007 AAAI Workshop, Association for the Advancement of Artificial Intelligence, Technical Report WS-07-06, Menlo Park California, AAAI Press, 2007, 1-9.

Recent theoretical and empirical work suggests that explanation and categorization are intimately related. This paper explores the hypothesis that explanations can help structure conceptual representations, and thereby influence the relative importance of features in categorization decisions. In particular, features may be differentially important depending on the role they play in explaining other features or aspects of category membership. Two experiments manipulate whether a feature is explained mechanistically, by appeal to proximate causes, or functionally, by appeal to a function or goal. (...) Explanation type has a significant impact on the relative importance of features in subsequent categorization judgments, with functional explanations reversing previously documented effects of [”]causal status’. The findings suggest that a feature’s explanatory importance can impact categorization, and that explanatory relationships, in addition to causal relationships, are critical to understanding conceptual representation. (shrink)

Good explanations are not only true or probably true, but are also relevant to a causal question. Current models of causal explanation either only address the question of the truth of an explanation, or do not distinguish the probability of an explanation from its relevance. The tasks of scenario construction and conversational explanation are distinguished, which in turn shows how scenarios can interact with conversational principles to determine the truth and relevance of explanations. The proposed model distinguishes causal discounting from (...) causal backgrounding , and makes predictions concerning the differential effects of contextual information about alternative explanations on: (a) the kind of mental models constructed; (b) belief revision about probable cause; and (c) the perceived quality of a focal explanation. Four experiments are reported that test these predictions. The significance of the notion of explanatory relevance for research on causal explanation is then discussed. (shrink)

This paper offers a dialogue theory of explanation. A successful explanation is defined as a transfer of understanding in a dialogue system in which a questioner and a respondent take part. The questioner asks a special sort of why-question that asks for understanding of something and the respondent provides a reply that transfers understanding to the questioner. The theory is drawn from recent work on explanation in artificial intelligence (AI), especially in expert systems, but applies to scientific, legal and everyday (...) conversational explanations. (shrink)

Reviewed Work: Explaining Explanation by David-Hillel Ruben .

Contents: PART 1. MODELS IN SCIENTIFIC PROCESSES. Joseph AGASSI: Why there is no theory of models. Ma??l??gorzata CZARNOCKA: Models and symbolic nature of knowledge. Adam GROBLER: The representational and the non-representational in models of scientific theories. Stephan HARTMANN: Models as a tool for the theory construction; some strategies of preliminary physics. William HERFEL: Nonlinear dynamical models as concrete construction. Elzbieta KA??L??USZY??N??SKA: Styles of thinking. Stathis PSILLOS: The cognitive interplay between theories and models: the case of 19th century optics. PART 2. (...) TOOLS OF SCIENCE. Nancy D. CARTWRIGHT, Towfic SHOMAR, Maricio SUAREZ: The tool-box of science. Javier ECHEVERRIA: The four contexts of scientific activity. Katline HAVAS: Continuity and change; kinds of negation in scientific progress. Matthias KAISER: The independence of scientific phenomena. W??l??adys??l??aw KRAJEWSKI: Scientific meta-philosophy. Ilkka NIINILUOTO: The emergence if scientific specialities: six models. Leszek NOWAK: Antirealism, realism and idealization. Rinat M. NUGAYEV: Classic, modern and postmodern scientific unification. Veikko RANTALA: Translation and scientific change. Gerhard SCHURZ: Theories and their applications - a case of nonmonotonic reasoning. Witold STRAWI??N??SKI: The unity of science today. Vardan TOROSIAN: Are the ethic and logic of science compatible. PART 3. UNSHARP APPROACHES IN SCIENCE. Ernest W. ADAMS: Problems and prospects in a theory of inexact first-order theories. Wolfgang BALZER and Gerhard ZOUBEK: On the comparison of approximative empirical claims. Gianpierro CATTANEO, Maria Luisa DALLA CHIARA, Roberto GIUNTINI: The unsharp approaches to quantum theory. Theo A.F. KUIPERS: Falsification versus efficient truth approximation. Bernhard LAUTH: Limiting decidability and probability. Jaros??l??aw PYKACZ: Many-valued logics in foundations of quantum mechanics. Roman R. ZAPATRIN: Logico-algebraic approach to spacetime quantization. (shrink)