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)

What is human understanding and why should we care about it? I propose a method of philosophical investigation called ‘function-first epistemology’ and use this method to investigate the nature and value of understanding-why. I argue that the concept of understanding-why serves the practical function of identifying good explainers, which is an important role in the general economy of our concepts. This hypothesis sheds light on a variety of issues in the epistemology of understanding including the role of explanation, the relationship (...) between understanding-why and knowledge, and the value of understanding-why. I conclude that understanding-why is valuable and yet knowledge plays more important roles in our epistemic life. (shrink)

We might think that thought experiments are at their most powerful or most interesting when they produce new knowledge. This would be a mistake; thought experiments that seek understanding are just as powerful and interesting, and perhaps even more so. A growing number of epistemologists are emphasizing the importance of understanding for epistemology, arguing that it should supplant knowledge as the central notion. In this chapter, I bring the literature on understanding in epistemology to bear on explicating the different ways (...) that thought experiments increase three important kinds of understanding: explanatory, objectual and practical. (shrink)

This chapter examines the status of inference to the best explanation in naturalistic metaphysics. The methodology of inference to the best explanation in metaphysics is studied from the perspective of contemporary views on scientific explanation and explanatory inferences in the history and philosophy of science. This reveals serious shortcomings in prevalent attempts to vindicate metaphysical "explanationism" by reference to similarities between science and naturalistic metaphysics. This critique is brought out by considering a common gambit of methodological unity: (1) Both metaphysics (...) and science employ inference to the best explanation. (2) One has no reason to think that if explanationism is truth-conducive in science, it is not so in metaphysics. (3) One has a positive reason to think that if explanationism is truth-conducive in science, it is also so in metaphysics. (shrink)

An influential suggestion about the relationship between Bayesianism and inference to the best explanation holds that IBE functions as a heuristic to approximate Bayesian reasoning. While this view promises to unify Bayesianism and IBE in a very attractive manner, important elements of the view have not yet been spelled out in detail. I present and argue for a heuristic conception of IBE on which IBE serves primarily to locate the most probable available explanatory hypothesis to serve as a working hypothesis (...) in an agent’s further investigations. Along the way, I criticize what I consider to be an overly ambitious conception of the heuristic role of IBE, according to which IBE serves as a guide to absolute probability values. My own conception, by contrast, requires only that IBE can function as a guide to the comparative probability values of available hypotheses. This is shown to be a much more realistic role for IBE given the nature and limitations of the explanatory considerations with which IBE operates. (shrink)

Advancements in machine learning have fuelled the popularity of using AI decision algorithms in procedures such as bail hearings, medical diagnoses and recruitment. Academic articles, policy texts, and popularizing books alike warn that such algorithms tend to be opaque: they do not provide explanations for their outcomes. Building on a causal account of transparency and opacity as well as recent work on the value of causal explanation, I formulate a moral concern for opaque algorithms that is yet to receive a (...) systematic treatment in the literature: when such algorithms are used in life-changing decisions, they can obstruct us from effectively shaping our lives according to our goals and preferences, thus undermining our autonomy. I argue that this concern deserves closer attention as it furnishes the call for transparency in algorithmic decision-making with both new tools and new challenges. (shrink)

Accounts of scientific explanation disagree about what’s required for a cause, law, or other fact to be a reason why an event occurs. In short, they disagree about the conditions for explanatory relevance. Nonetheless, most accounts presuppose that claims about explanatory relevance play a descriptive role in tracking reality. By rejecting the need for this descriptivist assumption, I develop an expressivist account of explanatory relevance and explanation: to judge that an answer is explanatory is to express an attitude of _being (...) for being satisfied by that answer_. I show how expressivism vindicates ordinary scientific discourse about explanation, including claims about the objectivity and mind-independence of explanations. By avoiding commitment to ontic relevance relations, I rehabilitate an irrealist conception of explanation. (shrink)

In the last few years, biologists and computer scientists have claimed that the introduction of data science techniques in molecular biology has changed the characteristics and the aims of typical outputs (i.e. models) of such a discipline. In this paper we will critically examine this claim. First, we identify the received view on models and their aims in molecular biology. Models in molecular biology are mechanistic and explanatory. Next, we identify the scope and aims of data science (machine learning in (...) particular). These lie mainly in the creation of predictive models which performances increase as data set increases. Next, we will identify a tradeoff between predictive and explanatory performances by comparing the features of mechanistic and predictive models. Finally, we show how this a priori analysis of machine learning and mechanistic research applies to actual biological practice. This will be done by analyzing the publications of a consortium—The Cancer Genome Atlas—which stands at the forefront in integrating data science and molecular biology. The result will be that biologists have to deal with the tradeoff between explaining and predicting that we have identified, and hence the explanatory force of the ‘new’ biology is substantially diminished if compared to the ‘old’ biology. However, this aspect also emphasizes the existence of other research goals which make predictive force independent from explanation. (shrink)

Much recent work on explanation in the interventionist tradition emphasizes the explanatory value of stable causal generalizations—i.e., causal generalizations that remain true in a wide range of background circumstances. We argue that two separate explanatory virtues are lumped together under the heading of `stability’. We call these two virtues breadth and guidance respectively. In our view, these two virtues are importantly distinct, but this fact is neglected or at least under-appreciated in the literature on stability. We argue that an adequate (...) theory of explanatory goodness should recognize breadth and guidance as distinct virtues, as breadth and guidance track different ideals of explanation, satisfy different cognitive and pragmatic ends, and play different theoretical roles in helping us understand the explanatory value of mechanisms. Thus keeping track of the distinction between these two forms of stability yields a more accurate and perspicuous picture of the role that stability considerations play in explanation. (shrink)

Certain characteristics make machine learning a powerful tool for processing large amounts of data, and also particularly unsuitable for explanatory purposes. There are worries that its increasing use in science may sideline the explanatory goals of research. We analyze the key characteristics of ML that might have implications for the future directions in scientific research: epistemic opacity and the ‘theory-agnostic’ modeling. These characteristics are further analyzed in a comparison of ML with the traditional statistical methods, in order to demonstrate what (...) it is specifically that makes ML methodology substantially unsuitable for reaching explanations. The analysis is given broader philosophical context by connecting it with the views on the role of prediction and explanation in science, their relationship, and the value of explanation. We proceed to show, first, that ML disrupts the relationship between prediction and explanation commonly understood as a functional relationship. Then we show that the value of explanation is not exhausted in purely predictive functions, but rather has a ubiquitously recognized value for both science and everyday life. We then invoke two hypothetical scenarios with different degrees of automatization of science, which help test our intuitions on the role of explanation in science. The main question we address is whether ML will reorient or otherwise impact our standard explanatory practice. We conclude with a prognosis that ML would diversify science into purely predictively oriented research based on ML-like techniques and, on the other hand, remaining faithful to anthropocentric research focused on the search for explanation. (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)

People often answer why-questions with what we call experiential explanations: narratives or stories with temporal structure and concrete details. In contrast, on most theories of the epistemic function of explanation, explanations should be abstractive: structured by general relationships and lacking extraneous details. We suggest that abstractive and experiential explanations differ not only in level of abstraction, but also in structure, and that each form of explanation contributes to the epistemic goals of individual learners and of science. In particular, experiential explanations (...) support mental simulation and survive transitions across background theories; as a result, they support learning and help us translate between competing frameworks. Experiential explanations play an irreducible role in human cognition – and perhaps in science. (shrink)

In the last few years, biologists and computer scientists have claimed that the introduction of data science techniques in molecular biology has changed the characteristics and the aims of typical outputs (i.e. models) of such a discipline. In this paper we will critically examine this claim. First, we identify the received view on models and their aims in molecular biology. Models in molecular biology are mechanistic and explanatory. Next, we identify the scope and aims of data science (machine learning in (...) particular). These lie mainly in the creation of predictive models which performances increase as data set increases. Next, we will identify a tradeoff between predictive and explanatory performances by comparing the features of mechanistic and predictive models. Finally, we show how this a priori analysis of machine learning and mechanistic research applies to actual biological practice. This will be done by analyzing the publications of a consortium—The Cancer Genome Atlas—which stands at the forefront in integrating data science and molecular biology. The result will be that biologists have to deal with the tradeoff between explaining and predicting that we have identified, and hence the explanatory force of the ‘new’ biology is substantially diminished if compared to the ‘old’ biology. However, this aspect also emphasizes the existence of other research goals which make predictive force independent from explanation. (shrink)

Throughout science and mathematics, expert inquirers often reformulate existing problem-solving procedures and theories. But what value is there to reformulating, particularly when one already knows how to solve a given problem? Is reformulating merely instrumentally valuable for other practical or epistemic aims, or does it constitute a distinctive kind of epistemic achievement? I argue that by changing what we need to know to solve a problem, significant reformulations constitute a kind of intellectual value. Whereas some reformulations are trivial notational variants, (...) a significant reformulation provides an inferentially distinct problem-solving plan. I situate my preferred position as a middle ground between deflationary and metaphysically-substantial alternatives. (shrink)

The tremendous philosophical focus on how to characterize explanatory metaphysical dependence has eclipsed a number of other unresolved issued about scientific explanation. The purpose of this paper is taxonomical. I will outline a number of other questions about the nature of explanation and its role in science—eight, to be precise—and argue that each is independent. All of these topics have received some philosophical attention, but none nearly so much as it deserves. Furthermore, existing views on these topics have been obscured (...) by not distinguishing among these independent questions and, especially, by not separating them from the question of what metaphysical dependence relation is explanatory. Philosophical analysis of scientific explanation would be much improved by attending more carefully to these, and probably still other, elements of an account of explanation. (shrink)

Occam's razor—the idea that all else being equal, we should pick the simpler hypothesis—plays a prominent role in ordinary and scientific inference. But why are simpler hypotheses better? One attractive hypothesis known as Bayesian Occam's razor is that more complex hypotheses tend to be more flexible—they can accommodate a wider range of possible data—and that flexibility is automatically penalized by Bayesian inference. In two experiments, we provide evidence that people's intuitive probabilistic and explanatory judgments follow the prescriptions of BOR. In (...) particular, people's judgments are consistent with the two most distinctive characteristics of BOR: They penalize hypotheses as a function not only of their numbers of free parameters but also as a function of the size of the parameter space, and they penalize those hypotheses even when their parameters can be “tuned” to fit the data better than comparatively simpler hypotheses. (shrink)

This paper brings together results from the philosophy and the psychology of explanation to argue that there are multiple concepts of explanation in human psychology. Specifically, it is shown that pluralism about explanation coheres with the multiplicity of models of explanation available in the philosophy of science, and it is supported by evidence from the psychology of explanatory judgment. Focusing on the case of a norm of explanatory power, the paper concludes by responding to the worry that if there is (...) a plurality of concepts of explanation, one will not be able to normatively evaluate what counts as good explanation. (shrink)

As a strategy for exploring the relationship between understanding and knowledge, we consider whether epistemic luck – which is typically thought to undermine knowledge – undermines understanding. Questions about the etiology of understanding have also been at the heart of recent theoretical debates within epistemology. Kvanvig (2003) put forward the argument that there could be lucky understanding and produced an example that he deemed persuasive. Grimm (2006) responded with a case that, he argued, demonstrated that there could not be lucky (...) understanding. In this paper, we empirically examine how participants' patterns of understanding attributions line up with the predictions of Kvanvig and Grimm. We argue that the data challenge Kvanvig's position. People do not differentiate between knowing-why and understanding-why on the basis of proper etiology: attributions of knowledge and understanding involve comparable (and minimal) roles for epistemic luck. We thus posit that folk knowledge and understanding are etiologically symmetrical. (shrink)

In recent years, an active research program has emerged that aims to develop a Humean best-system account (BSA) of laws of nature that improves on Lewis’s canonical articulation of the view. Its guiding idea is that the laws are cognitive tools tailored to the specific needs and limitations of creatures like us. While current versions of this “pragmatic Humean” research program fare much better than Lewis’s account along many dimensions, I will argue that they have trouble making sense of certain (...) key features of the practice of fundamental physics. Indeed, these features seem to go against the very idea that laws are useful for agents like us. In my view, Humeans can address these issues by paying more attention to the explanatory role of laws. Following this idea, I will propose an account on which what makes a systematization the best is a kind of explanatory power, understood along the lines of the unificationist theory of explanation. The resulting view, I will argue, can make sense of those features of laws that other pragmatic accounts of laws have trouble explaining. (shrink)

Our concept of actual causation plays a deep, ever-present role in our experiences. I first argue that traditional philosophical methods for understanding this concept are unlikely to be successful. I contend that we should instead use functional analyses and an understanding of the cognitive bases of causal cognition to gain insight into the concept of actual causation. I additionally provide initial, programmatic steps towards carrying out such analyses. The characterization of the concept of actual causation that results is quite different (...) from many standard views: it is graded, context-sensitive, and extrinsic. (shrink)

This paper examines two prominent formal trade-offs in artificial intelligence (AI)---between predictive accuracy and fairness, and between predictive accuracy and interpretability. These trade-offs have become a central focus in normative and regulatory discussions as policymakers seek to understand the value tensions that can arise in the social adoption of AI tools. The prevailing interpretation views these formal trade-offs as directly corresponding to tensions between underlying social values, implying unavoidable conflicts between those social objectives. In this paper, I challenge that prevalent (...) interpretation by introducing a sociotechnical approach to examining the value implications of trade-offs. After providing a set-theoretic understanding of these model trade-offs, I argue that their implications cannot be understood in isolation from the technical, psychological, organizational, and social factors that shape the integration of AI models into decision-making pipelines. Specifically, I identify three key considerations---validity and instrumental relevance, compositionality, and dynamics---for contextualizing and characterizing these implications. These considerations reveal that the relationship between model trade-offs and corresponding values depends on critical choices and assumptions. Crucially, judicious sacrifices in one model property for another can, in fact, promote both sets of corresponding values. The proposed sociotechnical perspective thus shows that we can and should aspire to higher epistemic and ethical possibilities than the prevalent interpretation suggests, while offering practical guidance for achieving those outcomes. Finally, I draw out the broader implications of this perspective for AI design and governance, highlighting the need to broaden normative engagement across the AI lifecycle, develop legal and auditing tools sensitive to sociotechnical considerations, and rethink the vital role and appropriate structure of interdisciplinary collaboration in fostering a responsible AI workforce. (shrink)

This paper provides evidence for a contrastive account of explanation that is motivated by pragmatic theories that recognize the contribution that context makes to the interpretation of a prompt for explanation. This study replicates the primary findings of previous work in explanation-based category learning, extending that work by illustrating the critical role of the context in this type of learning. Participants interacted with items from two categories either by describing the items or explaining their category membership. We manipulated the feature (...) overlap between the categories and examined both the explanations generated and acquired knowledge of the categories. Explanations for membership in a given category were influenced by the unprompted contrast category, indicating an important role for contrastive processing in the generation of explanations. The influence of the contrast category was similarly seen in the transfer performance of the participants. (shrink)

Scientific explanations are widely recognized to have instrumental value by helping scientists make predictions and control their environment. In this paper I raise, and provide a first analysis of, the question whether explanatory proofs in mathematics have analogous instrumental value. I first identify an important goal in mathematical practice: reusing resources from existing proofs to solve new problems. I then consider the more specific question: do explanatory proofs have instrumental value by promoting reuse of the resources they contain? In general, (...) I argue that the answer to this question is “no” and demonstrate this in detail for the theory of mathematical explanation developed by Marc Lange. (shrink)

This paper examines two prominent formal trade-offs in artificial intelligence (AI)---between predictive accuracy and fairness, and between predictive accuracy and interpretability. These trade-offs have become a central focus in normative and regulatory discussions as policymakers seek to understand the value tensions that can arise in the social adoption of AI tools. The prevailing interpretation views these formal trade-offs as directly corresponding to tensions between underlying social values, implying unavoidable conflicts between those social objectives. In this paper, I challenge that prevalent (...) interpretation by introducing a sociotechnical approach to examining the value implications of trade-offs. After providing a set-theoretic understanding of these model trade-offs, I argue that their implications cannot be understood in isolation from the technical, psychological, organizational, and social factors that shape the integration of AI models into decision-making pipelines. Specifically, I identify three key considerations---validity and instrumental relevance, compositionality, and dynamics---for contextualizing and characterizing these implications. These considerations reveal that the relationship between model trade-offs and corresponding values depends on critical choices and assumptions. Crucially, judicious sacrifices in one model property for another can, in fact, promote both sets of corresponding values. The proposed sociotechnical perspective thus shows that we can and should aspire to higher epistemic and ethical possibilities than the prevalent interpretation suggests, while offering practical guidance for achieving those outcomes. Finally, I draw out the broader implications of this perspective for AI design and governance, highlighting the need to broaden normative engagement across the AI lifecycle, develop legal and auditing tools sensitive to sociotechnical considerations, and rethink the vital role and appropriate structure of interdisciplinary collaboration in fostering a responsible AI workforce. (shrink)

People use commonsense science knowledge to flexibly explain, predict, and manipulate the world around them, yet we lack computational models of how this commonsense science knowledge is represented, acquired, utilized, and revised. This is an important challenge for cognitive science: Building higher order computational models in this area will help characterize one of the hallmarks of human reasoning, and it will allow us to build more robust reasoning systems. This paper presents a novel assembled coherence theory of human conceptual change, (...) whereby people revise beliefs and mental models by constructing and evaluating explanations using fragmentary, globally inconsistent knowledge. We implement AC theory with Timber, a computational model of conceptual change that revises its beliefs and generates human-like explanations in commonsense science. Timber represents domain knowledge using predicate calculus and qualitative model fragments, and uses an abductive model formulation algorithm to construct competing explanations for phenomena. Timber then scores competing explanations with respect to previously accepted beliefs, using a cost function based on simplicity and credibility, identifies a low-cost, preferred explanation and accepts its constituent beliefs, and then greedily alters previous explanation preferences to reduce global cost and thereby revise beliefs. Consistency is a soft constraint in Timber; it is biased to select explanations that share consistent beliefs, assumptions, and causal structure with its other, preferred explanations. In this paper, we use Timber to simulate the belief changes of students during clinical interviews about how the seasons change. We show that Timber produces and revises a sequence of explanations similar to those of the students, which supports the psychological plausibility of AC theory. (shrink)

It has often been argued that we face a trade-off between accuracy and opacity in deep learning models. The idea is that we can only harness the accuracy of deep learning models by simultaneously accepting that the grounds for the models’ decision-making are epistemically opaque to us. In this paper, we ask the following question: what are the prospects of making deep learning models transparent without compromising on their accuracy? We argue that the answer to this question depends on which (...) kind of opacity we have in mind. If we focus on the standard notion of opacity, which tracks the internal complexities of deep learning models, we argue that existing explainable AI (XAI) techniques show us that the prospects look relatively good. But, as it has recently been argued in the literature, there is another notion of opacity that concerns factors external to the model. We argue that there are at least two types of external opacity—link opacity and structure opacity—and that existing XAI techniques can to some extent help us reduce the former but not the latter. (shrink)

Explanation is a central concept in human psychology. Drawing upon philosophical theories of explanation, psychologists have recently begun to examine the relationship between explanation, probability and causality. Our study advances this growing literature in the intersection of psychology and philosophy of science by systematically investigating how judgments of explanatory power are affected by the prior credibility of a potential explanation, the causal framing used to describe the explanation, the generalizability of the explanation, and its statistical relevance for the evidence. Collectively, (...) the results of our five experiments support the hypothesis that the prior credibility of a causal explanation plays a central role in explanatory reasoning: first, because of the presence of strong main effects on judgments of explanatory power, and second, because of the gate-keeping role it has for other factors. Highly credible explanations were not susceptible to causal framing effects. Instead, highly credible hypotheses were sensitive to the effects of factors which are usually considered relevant from a normative point of view: the generalizability of an explanation, and its statistical relevance for the evidence. These results advance current literature in the philosophy and psychology of explanation in three ways. First, they yield a more nuanced understanding of the determinants of judgments of explanatory power, and the interaction between these factors. Second, they illuminate the close relationship between prior beliefs and explanatory power. Third, they clarify the relationship between abductive and probabilistic reasoning. (shrink)

Science and mathematics continually change in their tools, methods, and concepts. Many of these changes are not just modifications but progress---steps to be admired. But what constitutes progress? This dissertation addresses one central source of intellectual advancement in both disciplines: reformulating a problem-solving plan into a new, logically compatible one. For short, I call these cases of compatible problem-solving plans "reformulations." Two aspects of reformulations are puzzling. First, reformulating is often unnecessary. Given that we could already solve a problem using (...) an older formulation, what do we gain by reformulating? Second, some reformulations are genuinely trivial or insignificant. Merely replacing one symbol with another does not lead to intellectual progress. What distinguishes significant reformulations from trivial ones? According to what I call "conceptualism" (or "conceptual empiricism"), reformulations are intellectually significant when they provide a different plan for solving problems. Significant reformulations provide inferentially different routes to the same solution. In contrast, trivial reformulations provide exactly the same problem-solving plans, and hence they do not change our understanding. This answers the second question about what distinguishes trivial from significant reformulations. However, the first question remains: what makes a new way of solving an old problem valuable? Here, a bevy of practical considerations come to mind: one formulation might be faster, less complicated, or use more familiar concepts. According to "instrumentalism," these practical benefits are all there is to reformulating. Some reformulations are simply more instrumentally valuable for meeting the aims of science than others. At another extreme, "fundamentalism" contends that a reformulation is valuable when it provides a more fundamental description of reality. According to this view, some reformulations directly contribute to the metaphysical aim of carving reality at its joints. Conceptualism develops a middle ground between instrumentalism and fundamentalism, preserving their benefits without their costs. I argue that the epistemic value of significant reformulations does not reduce to either practical or metaphysical value. Reformulations are valuable because they are a constitutive part of problem-solving. Both science and mathematics aim at solving all possible problems within their respective domains. Meeting this aim requires being able to plan for any possible problem-solving context, and this requires reformulating. By reformulating, we clarify what we need to know to solve problems. Still, one might wonder whether the value of reformulations requires underlying differences in explanatory power. According to "explanationism," a reformulation is valuable only when it provides a better explanation. Explanationism stands as a rival middle ground position to my own. However, it faces numerous counterexamples. In many cases, two reformulations provide the same explanation while nonetheless providing different ways of understanding a phenomenon. Hence, reformulating can be valuable even when neither formulation is more explanatory. Methodologically, I draw on a variety of case studies to support my account of reformulation. These range from classical mechanics to quantum chemistry, along with examples from mathematics. Symmetry arguments provide a paradigmatic example: the mathematics of symmetry groups radically recasts quantum mechanics and quantum chemistry. Nevertheless, elementary approaches exist that eschew this additional mathematical apparatus, solving problems in a more tedious but less mathematically-demanding manner. Further examples include reformulations of quantum field theory, Arabic vs. Roman numerals, and Fermat's little theorem in number theory. In each case, my account identifies how reformulations change and improve our understanding of science and mathematics. (shrink)

Explanations are conceived to ensure the trustworthiness of AI systems. Yet, relying solemnly on algorithmic solutions, as provided by explainable artificial intelligence (XAI), might fall short to account for sociotechnical risks jeopardizing their factuality and informativeness. To mitigate these risks, we delve into the complex landscape of ethical risks surrounding XAI systems and their generated explanations. By employing a literature review combined with rigorous thematic analysis, we uncover a diverse array of technical risks tied to the robustness, fairness, and evaluation (...) of XAI systems. Furthermore, we address a broader range of contextual risks jeopardizing their security, accountability, reception alongside other cognitive, social, and ethical concerns of explanations. We advance a multi-layered risk assessment framework, where each layer advances strategies for practical intervention, management, and documentation of XAI systems within organizations. Recognizing the theoretical nature of the framework advanced, we discuss it in a conceptual case study. For the XAI community, our multifaceted investigation represents a path to practically address XAI risks while enriching our understanding of the ethical ramifications of incorporating XAI in decision-making processes. (shrink)

Among the many causes of an event, how do we distinguish the important ones? Are there ways to distinguish among causes on principled grounds that integrate both practical aims and objective knowledge? Psychologist Tania Lombrozo has suggested that causal explanations “identify factors that are ‘exportable’ in the sense that they are likely to subserve future prediction and intervention” (Lombrozo 2010, 327). Hence portable causes are more important precisely because they provide objective information to prediction and intervention as practical aims. However, (...) I argue that this is only part of the epistemology of causal selection. Recent work on portable causes has implicitly assumed them to be portable within the same causal system at a later time. As a result, it has appeared that the objective content of causal selection includes only facts about the causal structure of that single system. In contrast, I present a case study from systems biology in which scientists are searching for causal factors that are portable across rather than within causal systems. By paying careful attention to how these biologists find portable causes, I show that the objective content of causal selection can extend beyond the immediate systems of interest. In particular, knowledge of the evolutionary history of gene networks is necessary for correctly identifying causal patterns in these networks that explain cellular behavior in a portable way. (shrink)

Abductive reasoning assigns special status to the explanatory power of a hypothesis. But how do people make explanatory judgments? Our study clarifies this issue by asking: How does the explanatory power of a hypothesis cohere with other cognitive factors? How does probabilistic information affect explanatory judgments? In order to answer these questions, we conducted an experiment with 671 participants. Their task was to make judgments about a potentially explanatory hypothesis and its cognitive virtues. In the responses, we isolated three constructs: (...) Explanatory Value, Rational Acceptability, and Entailment. Explanatory judgments strongly cohered with judgments of causal relevance and with a sense of understanding. Furthermore, we found that Explanatory Value was sensitive to manipulations of statistical relevance relations between hypothesis and evidence, but not to explicit information about the prior probability of the hypothesis. These results indicate that probabilistic information about statistical relevance is a strong determinant of Explanatory Value. More generally, our study suggests that abductive and probabilistic reasoning are two distinct modes of inference. (shrink)

The relation between probabilistic and explanatory reasoning is a classical topic in philosophy of science. Most philosophical analyses are concerned with the compatibility of Inference to the Best Explanation with probabilistic, Bayesian inference, and the impact of explanatory considerations on the assignment of subjective probabilities. This paper reverses the question and asks how causal and explanatory considerations are affected by probabilistic information. We investigate how probabilistic information determines the explanatory value of a hypothesis, and in which sense folk explanatory practice (...) can be said to be rational. Our study identifies three main factors in reasoning about a explanatory hypothesis: cognitive salience, rational acceptability and logical entailment. This corresponds well to the variety of philosophical accounts of explanation. Moreover, we show that these factors are highly sensitive to manipulations of probabilistic information. This finding suggests that probabilistic reasoning is a crucial part of explanatory inferences, and it motivates new avenues of research in the debate about Inference to the Best Explanation and probabilistic measures of explanatory power. (shrink)

The relatively new movement of Experimental Philosophy applies different systematic experimental methods to further illuminate classical philosophical issues. This book brings together experts from the field to give the reader a compact yet extensive overview, offering a ready at hand introduction to the state of the art.

Following an analysis of the state of investigations and clinical outcomes in the Alzheimer's research field, I argue that the widely-accepted 'amyloid cascade' mechanistic explanation of Alzheimer's disease appears to be fundamentally incomplete. In this context, I propose that a framework termed 'principled mechanism' (PM) can help with remedying this problem. First, using a series of five 'tests', PM systematically compares different components of a given mechanistic explanation against a paradigmatic set of criteria, and hints at various ways of making (...) the mechanistic explanation more 'complete'. These steps will be demonstrated using the amyloid explanation, and its missing or problematic mechanistic elements will be highlighted. Second, PM makes an appeal for the discovery and application of 'biological principles' (BPs), which approximate ceteris paribus laws and are operative at the level of a biological cell. As such, although thermodynamic, evolutionary, ecological and other laws or principles from chemistry and the broader life sciences could inform them, BPs should be considered ontologically unique. BPs could augment different facets of the mechanistic explanation but also allow further independent nomological explanation of the phenomenon. Whilst this overall strategy can be complementary to certain 'New Mechanist' approaches, an important distinction of the PM framework is its equal attention to the explanatory utility of biological principles. Lastly, I detail two hypothetical BPs, and show how they could each inform and improve the potentially incomplete mechanistic aspects of the amyloid explanation and also how they could provide independent explanations of the cellular features associated with Alzheimer's disease. (shrink)