How do we go about weighing evidence, testing hypotheses, and making inferences? The model of " inference to the best explanation " -- that we infer the hypothesis that would, if correct, provide the best explanation of the available evidence--offers a compelling account of inferences both in science and in ordinary life. Widely cited by epistemologists and philosophers of science, IBE has nonetheless remained little more than a slogan. Now this influential work has been thoroughly revised and updated, (...) and features a new introduction and two new chapters. Inference to the Best Explanation is an unrivaled exposition of a theory of particular interest in the fields both of epistemology and the philosophy of science. (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 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 to the causal/mechanical model requires fundamental rethinking of basic explanatory concepts. (shrink)
As Aristotle stated, scientific explanation is based on deductive argument--yet, Wesley C. Salmon points out, not all deductive arguments are qualified explanations. The validity of the explanation must itself be examined. _Four Decades of Scientific Explanation_ provides a comprehensive account of the developments in scientific explanation that transpired in the last four decades of the twentieth century. It continues to stand as the most comprehensive treatment of the writings on the subject during these years. Building on the (...) historic 1948 essay by Carl G. Hempel and Paul Oppenheim, "Studies in the Logic of Explanation,” which introduced the deductive-nomological model on which most work on scientific explanation was based for the following four decades, Salmon goes beyond this model's inherent basis of describing empirical knowledge to tells us “not only _what,_ but also _why_.” Salmon examines the predominant models in chronological order and describes their development, refinement, and criticism or rejection. _Four Decades of Scientific Explanation_ underscores the need for a consensus of approach and ongoing evaluations of methodology in scientific explanation, with the goal of providing a better understanding of natural phenomena. (shrink)
_This paper deals with the interrelationship between causal explanation and methodology in a relatively young discipline in biology: epigenetics. Based on cases from molecular and ecological epigenetics, I show that James Woodward’s interventionist account of causation captures essential features about how epigeneticists using highly diverse methods, i.e. laboratory experiments and purely observational studies, think about causal explanation. I argue that interventionism thus qualifies as a useful unifying explanatory approach when it comes to cross-methodological research efforts: It can act (...) as a guiding rationale to link causal models in molecular biology with statistical models derived from observational data analysis and to identify test-criteria for reciprocal transparent studies in different fields of research, which is a shared issue across the sciences.___. (shrink)
Approaches to explanation -- Causal and explanatory relevance -- The kairetic account of /D making -- The kairetic account of explanation -- Extending the kairetic account -- Event explanation and causal claims -- Regularity explanation -- Abstraction in regularity explanation -- Approaches to probabilistic explanation -- Kairetic explanation of frequencies -- Kairetic explanation of single outcomes -- Looking outward -- Looking inward.
A compelling idea holds that reality has a layered structure. We often disagree about what inhabits the bottom layer, but we agree that higher up we find chemical, biological, geological, psychological, sociological, economic, /etc./, entities: molecules, human beings, diamonds, mental states, cities, interest rates, and so on. How is this intuitive talk of a layered structure of entities to be understood? Traditionally, philosophers have proposed to understand layered structure in terms of either reduction or supervenience. But these traditional views face (...) well-known problems. A plausible alternative is that layered structure is to be explicated by appeal to explanations of a certain sort, termed / grounding explanations/. Grounding explanations tell us what obtains in virtue of what. Unfortunately, the use of grounding explanations to articulate the layered conception faces a problem, which I call /the collapse/. The collapse turns on the question of how to ground the facts stated by the explanations themselves. In this paper I make a suggestion about how to ground explanations that avoids the collapse. Briefly, the suggestion is that the fact stated by a grounding explanation is grounded in its /explanans/. (shrink)
Explanations are very important to us in many contexts: in science, mathematics, philosophy, and also in everyday and juridical contexts. But what is an explanation? In the philosophical study of explanation, there is long-standing, influential tradition that links explanation intimately to causation: we often explain by providing accurate information about the causes of the phenomenon to be explained. Such causal accounts have been the received view of the nature of explanation, particularly in philosophy of science, since (...) the 1980s. However, philosophers have recently begun to break with this causal tradition by shifting their focus to kinds of explanation that do not turn on causal information. The increasing recognition of the importance of such non-causal explanations in the sciences and elsewhere raises pressing questions for philosophers of explanation. What is the nature of non-causal explanations – and which theory best captures it? How do non-causal explanations relate to causal ones? How are non-causal explanations in the sciences related to those in mathematics and metaphysics? This volume of new essays explores answers to these and other questions at the heart of contemporary philosophy of explanation. The essays address these questions from a variety of perspectives, including general accounts of non-causal and causal explanations, as well as a wide range of detailed case studies of non-causal explanations from the sciences, mathematics and metaphysics. (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)
Normative explanations of why things are wrong, good, or unfair are ubiquitous in ordinary practice and normative theory. This paper argues that normative explanation is subject to a justification condition: a correct complete explanation of why a normative fact holds must identify features that would go at least some way towards justifying certain actions or attitudes. I first explain and motivate the condition I propose. I then support it by arguing that it fits well with various theories of (...) normative reasons, makes good sense of certain legitimate moves in ordinary normative explanatory discourse, and helps to make sense of our judgments about explanatory priority in certain cases of normative explanation. This last argument also helps to highlight respects in which normative explanation won’t be worryingly discontinuous with explanations in other domains even though these other explanations aren’t subject to the justification condition. Thus the paper aims not only to do some constructive theorizing about the relatively neglected topic of normative explanation but also to cast light on the broader question of how normative explanation may be similar to and different from explanations in other domains. (shrink)
Explanations in the life sciences frequently involve presenting a model of the mechanism taken to be responsible for a given phenomenon. Such explanations depart in numerous ways from nomological explanations commonly presented in philosophy of science. This paper focuses on three sorts of differences. First, scientists who develop mechanistic explanations are not limited to linguistic representations and logical inference; they frequently employ dia- grams to characterize mechanisms and simulations to reason about them. Thus, the epistemic resources for presenting mechanistic explanations (...) are considerably richer than those suggested by a nomological framework. Second, the fact that mechanisms involve organized systems of component parts and operations provides direction to both the discovery and testing of mech- anistic explanations. Finally, models of mechanisms are developed for specific exemplars and are not represented in terms of universally quantified statements. Generalization involves investigating both the similarity of new exemplars to those already studied and the variations between them. Ó 2005 Elsevier Ltd. All rights reserved. (shrink)
Some properties are causally relevant for a certain effect, others are not. In this paper we describe a problem for our understanding of this notion and then offer a solution in terms of the notion of a program explanation.
This article considers the prospects of inference to the best explanation as a method of confirming causal claims vis-à-vis the medical evidence of mechanisms. I show that IBE is actually descriptive of how scientists reason when choosing among hypotheses, that it is amenable to the balance/weight distinction, a pivotal pair of concepts in the philosophy of evidence, and that it can do justice to interesting features of the interplay between mechanistic and population level assessments.
Recent work on interpretability in machine learning and AI has focused on the building of simplified models that approximate the true criteria used to make decisions. These models are a useful pedagogical device for teaching trained professionals how to predict what decisions will be made by the complex system, and most importantly how the system might break. However, when considering any such model it’s important to remember Box’s maxim that "All models are wrong but some are useful." We focus on (...) the distinction between these models and explanations in philosophy and sociology. These models can be understood as a "do it yourself kit" for explanations, allowing a practitioner to directly answer "what if questions" or generate contrastive explanations without external assistance. Although a valuable ability, giving these models as explanations appears more difficult than necessary, and other forms of explanation may not have the same trade-offs. We contrast the different schools of thought on what makes an explanation, and suggest that machine learning might benefit from viewing the problem more broadly. (shrink)
Call an explanation in which a non-mathematical fact is explained—in part or in whole—by mathematical facts: an extra-mathematical explanation. Such explanations have attracted a great deal of interest recently in arguments over mathematical realism. In this article, a theory of extra-mathematical explanation is developed. The theory is modelled on a deductive-nomological theory of scientific explanation. A basic DN account of extra-mathematical explanation is proposed and then redeveloped in the light of two difficulties that the basic (...) theory faces. The final view appeals to relevance logic and uses resources in information theory to understand the explanatory relationship between mathematical and physical facts. 1Introduction2Anchoring3The Basic Deductive-Mathematical Account4The Genuineness Problem5Irrelevance6Relevance and Information7Objections and Replies 7.1Against relevance logic7.2Too epistemic7.3Informational containment8Conclusion. (shrink)
Explanations in the life sciences frequently involve presenting a model of the mechanism taken to be responsible for a given phenomenon. Such explanations depart in numerous ways from nomological explanations commonly presented in philosophy of science. This paper focuses on three sorts of differences. First, scientists who develop mechanistic explanations are not limited to linguistic representations and logical inference; they frequently employ diagrams to characterize mechanisms and simulations to reason about them. Thus, the epistemic resources for presenting mechanistic explanations are (...) considerably richer than those suggested by a nomological framework. Second, the fact that mechanisms involve organized systems of component parts and operations provides direction to both the discovery and testing of mechanistic explanations. Finally, models of mechanisms are developed for specific exemplars and are not represented in terms of universally quantified statements. Generalization involves investigating both the similarity of new exemplars to those already studied and the variations between them. (shrink)
The ultimate source of explanation in biology is the principle of natural selection. Natural selection means differential reproduction of genes and gene combinations. It is a mechanistic process which accounts for the existence in living organisms of end-directed structures and processes. It is argued that teleological explanations in biology are not only acceptable but indeed indispensable. There are at least three categories of biological phenomena where teleological explanations are appropriate.
Humeanism about laws of nature — the view that the laws reduce to the Humean mosaic — is a popular view, but currently existing versions face powerful objections. The non-supervenience objection, the non-fundamentality objection and the explanatory circularity objection have all been thought to cause problems for the Humean. However, these objections share a guiding thought — they are all based on the idea that there is a certain kind of divergence between the practice of science and the metaphysical picture (...) suggested by Humeanism. -/- I suggest that the Humean can respond to these objections not by rejecting this divergence, but by arguing that is appropriate. In particular the Humean can, in the spirit of Loewer (2012), distinguish between scientific and metaphysical explanation — this is motivated by differing aims of explanation in science and metaphysics. And they can further leverage this into distinctions between scientific and metaphysical fundamentality and scientific and metaphysical possibility. We can use these distinctions to respond to the objections that the Humean faces. (shrink)
This chapter explores an objection to explanatory universalism, the doctrine that the principle of sufficient reason is true or everything has an explanation. This objection is a direct argument to the conclusion that the PSR yields the existence of an omni-explainer, i.e. something that explains everything. The objection crucially relies on the assumption that explanation is dissective in its explanandum place, and its conclusion conflicts with the irreflexivity of explanation. So the chapter considers two responses to the (...) mentioned objection. The first response consisting in restricting the irreflexivity of explanation is criticised in connection with topics in the metaphysics of grounding. The second response consisting in denying that explanation is dissective is vindicated. Finally, the chapter argues that a plausible revised version of the principle that explanation is dissective, the PSR, and the irreflexivity of explanation together yield a striking picture of our universe. (shrink)
This paper offers a new account of metaphysical explanation. The account is modelled on Kitcher’s unificationist approach to scientific explanation. We begin, in Sect. 2, by briefly introducing the notion of metaphysical explanation and outlining the target of analysis. After that, we introduce a unificationist account of metaphysical explanation before arguing that such an account is capable of capturing four core features of metaphysical explanations: irreflexivity, non-monotonicity, asymmetry and relevance. Since the unificationist theory of metaphysical (...) class='Hi'>explanation inherits irreflexivity and non-monotonicity directly from the unificationist theory of scientific explanation that underwrites it, we focus on demonstrating how the account can secure asymmetry and relevance. (shrink)
This paper describes an alternative to the common view that explanation in the special sciences involves subsumption under laws. According to this alternative, whether or not a generalization can be used to explain has to do with whether it is invariant rather than with whether it is lawful. A generalization is invariant if it is stable or robust in the sense that it would continue to hold under a relevant if it is stable or robust in the sense that (...) it would continue to hold under a relevant class of changes. Unlike lawfulness, invariance comes in degrees and has other features that are well suited to capture the characteristics of explanatory generalizations in the special sciences. For example, a generalization can be invariant even if it has exceptions or holds only over a limited spatio-temporal interval. The notion of invariance can be used to resolve a number of dilemmas that arise in standard treatments of explanatory generalizations in the special sciences. (shrink)
The central aim of this paper is to shed light on the nature of explanation in computational neuroscience. I argue that computational models in this domain possess explanatory force to the extent that they describe the mechanisms responsible for producing a given phenomenon—paralleling how other mechanistic models explain. Conceiving computational explanation as a species of mechanistic explanation affords an important distinction between computational models that play genuine explanatory roles and those that merely provide accurate descriptions or predictions (...) of phenomena. It also serves to clarify the pattern of model refinement and elaboration undertaken by computational neuroscientists. (shrink)
Scientists and laypeople alike use the sense of understanding that an explanation conveys as a cue to good or correct explanation. Although the occurrence of this sense or feeling of understanding is neither necessary nor sufficient for good explanation, it does drive judgments of the plausibility and, ultimately, the acceptability, of an explanation. This paper presents evidence that the sense of understanding is in part the routine consequence of two well-documented biases in cognitive psychology: overconfidence and (...) hindsight. In light of the prevalence of counterfeit understanding in the history of science, I argue that many forms of cognitive achievement do not involve a sense of understanding, and that only the truth or accuracy of an explanation make the sense of understanding a valid cue to genuine understanding. (shrink)
Grounding and explanation are said to be intimately connected. Some even maintain that grounding just is a form of explanation. But grounding and explanation also seem importantly different—on the face of it, the former is ‘worldy’ or ‘objective’ while the latter isn’t. In this paper, we develop and respond to an argument to the effect that there is no way to fruitfully address this tension that retains orthodox views about grounding and explanation but doesn’t undermine a (...) central piece of methodology, namely that explanation is a guide to ground. (shrink)
This paper examines explanations that turn on non-local geometrical facts about the space of possible configurations a system can occupy. I argue that it makes sense to contrast such explanations from "geometry of motion" with causal explanations. I also explore how my analysis of these explanations cuts across the distinction between kinematics and dynamics.
Back cover: This book develops a philosophical account that reveals the major characteristics that make an explanation in the life sciences reductive and distinguish them from non-reductive explanations. Understanding what reductive explanations are enables one to assess the conditions under which reductive explanations are adequate and thus enhances debates about explanatory reductionism. The account of reductive explanation presented in this book has three major characteristics. First, it emerges from a critical reconstruction of the explanatory practice of the life (...) sciences itself. Second, the account is monistic since it specifies one set of criteria that apply to explanations in the life sciences in general. Finally, the account is ontic in that it traces the reductivity of an explanation back to certain relations that exist between objects in the world (such as part-whole relations and level relations), rather than to the logical relations between sentences. Beginning with a disclosure of the meta-philosophical assumptions that underlie the author’s analysis of reductive explanation, the book leads into the debate about reduction(ism) in the philosophy of biology and continues with a discussion on the two perspectives on explanatory reduction that have been proposed in the philosophy of biology so far. The author scrutinizes how the issue of reduction becomes entangled with explanation and analyzes two concepts, the concept of a biological part and the concept of a level of organization. The results of these five chapters constitute the ground on which the author bases her final chapter, developing her ontic account of reductive explanation. (shrink)
This volume distinguishes between two main traditions in the philosophy of science - the aristotelian, with its stress on explanation in terms of purpose and intentionality, and the galilean, which takes causal explanation as primary. It then traces the complex history of these competing traditions as they are manifested in such movements as positivism, idealism, Marxism and contemporary linguistic analysis. Hempels's theory of scientific explanation, the claims of cybernetics the rise of an analytic philosophy of action and (...) the revival of hermenuetics are all discussed. The volume also deals with causal explanation, intentionality and teleological explanation, and explanation in history and the social sciences. The author concludes that explanation of human actions cannot be reduced to simple causality, and discusses the implications of this conclusion for the disciplines of history and sociology. (shrink)
This volume addresses relations between macroscopic and microscopic description; essential roles of visualization and representation in chemical understanding; historical questions involving chemical concepts; the impacts of chemical ideas on wider cultural concerns; and relationships between contemporary chemistry and other sciences. The authors demonstrate, assert, or tacitly assume that chemical explanation is functionally autonomous. This volume should he of interest not only to professional chemists and philosophers, but also to workers in medicine, psychology, and other fields in which relationships between (...) explanations based on diverse levels of description and investigation are important. (Listed on Google books). (shrink)
This paper argues that besides mechanistic explanations, there is a kind of explanation that relies upon “topological” properties of systems in order to derive the explanandum as a consequence, and which does not consider mechanisms or causal processes. I first investigate topological explanations in the case of ecological research on the stability of ecosystems. Then I contrast them with mechanistic explanations, thereby distinguishing the kind of realization they involve from the realization relations entailed by mechanistic explanations, and explain how (...) both kinds of explanations may be articulated in practice. The second section, expanding on the case of ecological stability, considers the phenomenon of robustness at all levels of the biological hierarchy in order to show that topological explanations are indeed pervasive there. Reasons are suggested for this, in which “neutral network” explanations are singled out as a form of topological explanation that spans across many levels. Finally, I appeal to the distinction of explanatory regimes to cast light on a controversy in philosophy of biology, the issue of contingence in evolution, which is shown to essentially involve issues about realization. (shrink)
In this paper I show that a novel ontic reading of explanation, intending to capture the de re essential features of individuals, can support the qualitative view of individual essences. It is argued further that the putative harmful consequences of the Leibniz Principle and its converse for the qualitative view can be avoided, provided that individual essences are not construed in the style of the naïve bundle theory with set-theoretical identity- conditions. Adopting either the more sophisticated two-tier BT or, (...) alternatively, the neo-Aristotelian position of taking essences as natures in the Aristotelian sense, can help to evade these main charges against the qualitative view. The functional parallels with the alternative haecceitistic view of individuation and individual essence will also be considered. (shrink)
How far should our realism extend? For many years philosophers of mathematics and philosophers of ethics have worked independently to address the question of how best to understand the entities apparently referred to by mathematical and ethical talk. But the similarities between their endeavours are not often emphasised. This book provides that emphasis. In particular, it focuses on two types of argumentative strategies that have been deployed in both areas. The first—debunking arguments—aims to put pressure on realism by emphasising the (...) seeming redundancy of mathematical or moral entities when it comes to explaining our judgements. In the moral realm this challenge has been made by Gilbert Harman and Sharon Street; in the mathematical realm it is known as the 'Benacerraf-Field' problem. The second strategy—indispensability arguments—aims to provide support for realism by emphasising the seeming intellectual indispensability of mathematical or moral entities, for example when constructing good explanatory theories. This strategy is associated with Quine and Putnam in mathematics and with Nicholas Sturgeon and David Enoch in ethics. Explanation in Ethics and Mathematics addresses these issues through an explicitly comparative methodology which we call the 'companions in illumination' approach. By considering how argumentative strategies in the philosophy of mathematics might apply to the philosophy of ethics, and vice versa, the papers collected here break new ground in both areas. For good measure, two further companions for illumination are also broached: the philosophy of chance and the philosophy of religion. Collectively, these comparisons light up new questions, arguments, and problems of interest to scholars interested in realism in any area. (shrink)
In this article we analyze the methodological commitments of a radical embodied cognition approach to social interaction and social cognition, specifically with respect to the explanatory framework it adopts. According to many representatives of REC, such as enactivists and the proponents of dynamical and ecological psychology, sociality is to be explained by focusing on the social unit rather than the individuals that comprise it and establishing the regularities that hold on this level rather than modeling the sub-personal mechanisms that could (...) be said to underlie social phenomena. We point out that, despite explicit commitment, such a view implies an implicit rejection of the mechanistic explanation framework widely adopted in traditional cognitive science, which, in our view, hinders comparability between REC and these approaches. We further argue that such a position is unnecessary and that enactive mechanistic explanation of sociality is both possible and desirable. We examine three distinct objections from REC against mechanistic explanation, which we dub the decomposability, causality and extended cognition worries. In each case we show that these complaints can be alleviated by either appreciation of the full scope of the mechanistic account or adjustments on both mechanistic and REC sides of the debate. (shrink)
In this paper, I aim to provide access to the current debate on non-causal explanations in philosophy of science. I will first present examples of non-causal explanations in the sciences. Then, I will outline three alternative approaches to non-causal explanations – that is, causal reductionism, pluralism, and monism – and, corresponding to these three approaches, different strategies for distinguishing between causal and non-causal explanation. Finally, I will raise questions for future research on non-causal explanations.
I discuss the methodological passage in the begin- ning of Ethica Eudemia I.6 (1216b26-35), which has received attention in connection with Aristotle’s notion of dialectic and his methodology in Ethics. My central focus is not to discuss whether Aristotle is prescribing and using what has been called the method of endoxa. I will focus on how this passage coheres with the remaining parts of the same chapter, which also are advancing methodological remarks. My claim is that the meth- od of (...) Ethica Eudemia I.6 is in agreement with many features of Aristotle’s theory of explanation as presented in the Posterior Analytics: Aristotle’s main concern is a warning against misuses of explanatory arguments. (shrink)
A story does more than recount events; it recounts events in a way that renders them intelligible, thus conveying not just information but also understanding. We might therefore be tempted to describe narrative as a genre of explanation. When the police invite a suspect to “tell his story,” they are asking him to explain the blood on his shirt or his absence from home on the night of the murder; and whether he is judged to have a “good story” (...) will depend on its adequacy as an explanation. Can we account for the explanatory force of narrative with the models of explanation available in the philosophy of science? Or does narrative convey a different kind of understanding, which requires a different model and perhaps even a term other than ‘explanation’? (shrink)
Does mathematics ever play an explanatory role in science? If so then this opens the way for scientific realists to argue for the existence of mathematical entities using inference to the best explanation. Elsewhere I have argued, using a case study involving the prime-numbered life cycles of periodical cicadas, that there are examples of indispensable mathematical explanations of purely physical phenomena. In this paper I respond to objections to this claim that have been made by various philosophers, and I (...) discuss potential future directions of research for each side in the debate over the existence of abstract mathematical objects. (shrink)
The ontic conception of scientific explanation has been constructed and motivated on the basis of a putative lexical ambiguity in the term explanation. I raise a puzzle for this ambiguity claim, and then give a deflationary solution under which all ontically-rendered talk of explanation is merely elliptical; what it is elliptical for is a view of scientific explanation that altogether avoids the ontic conception. This result has revisionary consequences for New Mechanists and other philosophers of science, (...) many of whom have assimilated their conception of explanation to the ontic conception. (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 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.
We propose a formal framework for interpretable machine learning. Combining elements from statistical learning, causal interventionism, and decision theory, we design an idealised explanation game in which players collaborate to find the best explanation for a given algorithmic prediction. Through an iterative procedure of questions and answers, the players establish a three-dimensional Pareto frontier that describes the optimal trade-offs between explanatory accuracy, simplicity, and relevance. Multiple rounds are played at different levels of abstraction, allowing the players to explore (...) overlapping causal patterns of variable granularity and scope. We characterise the conditions under which such a game is almost surely guaranteed to converge on a optimal explanation surface in polynomial time, and highlight obstacles that will tend to prevent the players from advancing beyond certain explanatory thresholds. The game serves a descriptive and a normative function, establishing a conceptual space in which to analyse and compare existing proposals, as well as design new and improved solutions. (shrink)
Teleological explanations (TEs) account for the existence or properties of an entity in terms of a function: we have hearts because they pump blood, and telephones for communication. While many teleological explanations seem appropriate, others are clearly not warranted-for example, that rain exists for plants to grow. Five experiments explore the theoretical commitments that underlie teleological explanations. With the analysis of [Wright, L. (1976). Teleological Explanations. Berkeley, CA: University of California Press] from philosophy as a point of departure, we examine (...) in Experiment 1 whether teleological explanations are interpreted causally, and confirm that TEs are only accepted when the function invoked in the explanation played a causal role in bringing about what is being explained. However, we also find that playing a causal role is not sufficient for all participants to accept TEs. Experiment 2 shows that this is not because participants fail to appreciate the causal structure of the scenarios used as stimuli. In Experiments 3-5 we show that the additional requirement for TE acceptance is that the process by which the function played a causal role must be general in the sense of conforming to a predictable pattern. These findings motivate a proposal, Explanation for Export, which suggests that a psychological function of explanation is to highlight information likely to subserve future prediction and intervention. We relate our proposal to normative accounts of explanation from philosophy of science, as well as to claims from psychology and artificial intelligence. (shrink)
