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

In this paper, I critically evaluate several related, provocative claims made by proponents of data-intensive science and “Big Data” which bear on scientific methodology, especially the claim that scientists will soon no longer have any use for familiar concepts like causation and explanation. After introducing the issue, in Section 2, I elaborate on the alleged changes to scientific method that feature prominently in discussions of Big Data. In Section 3, I argue that these methodological claims are in tension with a (...) prominent account of scientific method, often called “Inference to the Best Explanation”. Later on, in Section 3, I consider an argument against IBE that will be congenial to proponents of Big Data, namely, the argument due to Roche and Sober Analysis, 73:659–668, that “explanatoriness is evidentially irrelevant.” This argument is based on Bayesianism, one of the most prominent general accounts of theory-confirmation. In Section 4, I consider some extant responses to this argument, especially that of Climenhaga Philosophy of Science, 84:359–368,. In Section 5, I argue that Roche and Sober’s argument does not show that explanatory reasoning is dispensable. In Section 6, I argue that there is good reason to think explanatory reasoning will continue to prove indispensable in scientific practice. Drawing on Cicero’s oft-neglected De Divinatione, I formulate what I call the “Ciceronian Causal-nomological Requirement”, which states, roughly, that causal-nomological knowledge is essential for relying on correlations in predictive inference. I defend a version of the CCR by appealing to the challenge of “spurious correlations,” chance correlations which we should not rely upon for predictive inference. In Section 7, I offer some concluding remarks. (shrink)

To succeed, political science usually requires either prediction or contextual historical work. Both of these methods favor explanations that are narrow-scope, applying to only one or a few cases. Because of the difficulty of prediction, the main focus of political science should often be contextual historical work. These epistemological conclusions follow from the ubiquity of causal fragility, under-determination, and noise. They tell against several practices that are widespread in the discipline: wide-scope retrospective testing, such as much large-n statistical work; lack (...) of emphasis on prediction; and resources devoted to ‘pure theory’ divorced from frequent empirical application. I illustrate, via Donatella della Porta’s work on political violence, the important role that is still left for theory. I conclude by assessing the scope for political science to offer policy advice. (shrink)

This book addresses key conceptual issues relating to the modern scientific and engineering use of computer simulations. It analyses a broad set of questions, from the nature of computer simulations to their epistemological power, including the many scientific, social and ethics implications of using computer simulations. The book is written in an easily accessible narrative, one that weaves together philosophical questions and scientific technicalities. It will thus appeal equally to all academic scientists, engineers, and researchers in industry interested in questions (...) related to the general practice of computer simulations. (shrink)

We introduce a novel point of view on the “models as mediators” framework in order to emphasize certain important epistemological questions about models in science which have so far been little investigated. To illustrate how this perspective can help answer these kinds of questions, we explore the use of simplified models in high energy physics research beyond the Standard Model. We show in detail how the construction of simplified models is grounded in the need to mitigate pressing epistemic problems concerning (...) the uncertainty inherent in the present theoretical and experimental contexts. (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)

Recently, it has been argued that the use of Big Data transforms the sciences, making data-driven research possible and studying causality redundant. In this paper, I focus on the claim on causal knowledge by examining the Big Data project EXPOsOMICS, whose research is funded by the European Commission and considered capable of improving our understanding of the relation between exposure and disease. While EXPOsOMICS may seem the perfect exemplification of the data-driven view, I show how causal knowledge is necessary for (...) the project, both as a source for handling complexity and as an output for meeting the project’s goals. Consequently, I argue that data-driven claims about causality are fundamentally flawed and causal knowledge should be considered a necessary aspect of Big Data science. In addition, I present the consequences of this result on other data-driven claims, concerning the role of theoretical considerations. I argue that the importance of causal knowledge and other kinds of theoretical engagement in EXPOsOMICS undermine theory-free accounts and suggest alternative ways of framing science based on Big Data. (shrink)

In recent years, the biomedical field has witnessed the emergence of novel tools and modelling techniques driven by the rise of the so-called Big Data. In this paper, we address the issue of predictability in biomedical Big Data models of cancer patients, with the aim of determining the extent to which computationally driven predictions can be implemented by medical doctors in their clinical practice. We show that for a specific class of approaches, called k-Nearest Neighbour algorithms, the ability to draw (...) predictive inferences relies on a geometrical, or topological, notion of similarity encoded in a well-defined metric, which determines how close the characteristics of distinct patients are on average. We then discuss the conditions under which the relevant models can yield reliable and trustworthy predictive outcomes. (shrink)

The prospects of a causal interpretation of probability are examined. Various accounts both from the history of scientific method and from recent developments in the tradition of the method of arbitrary functions, in particular by Strevens, Rosenthal, and Abrams, are briefly introduced and assessed. I then present a specific account of causal probability with the following features: First, the link between causal probability and a particular account of induction and causation is established, namely eliminative induction and the related difference-making account (...) of causation in the tradition of Bacon, Herschel, and Mill. Second, it is shown how a causal approach is useful beyond applications of the method of arbitrary functions and is able to deal with various shades of both ontic and epistemic probabilities. Furthermore, I clarify the notion of causal symmetry as a central element of an objective version of the principle of indifference and relate probabilistic independence to causal irrelevance. (shrink)

In climate science, climate models are one of the main tools for understanding phenomena. Here, we develop a framework to assess the fitness of a climate model for providing understanding. The framework is based on three dimensions: representational accuracy, representational depth, and graspability. We show that this framework does justice to the intuition that classical process-based climate models give understanding of phenomena. While simple climate models are characterized by a larger graspability, state-of-the-art models have a higher representational accuracy and representational (...) depth. We then compare the fitness-for-providing understanding of process-based to data-driven models that are built with machine learning. We show that at first glance, data-driven models seem either unnecessary or inadequate for understanding. However, a case study from atmospheric research demonstrates that this is a false dilemma. Data-driven models can be useful tools for understanding , specifically for phenomena for which scientists can argue from the coherence of the models with background knowledge to their representational accuracy and for which the model complexity can be reduced such that they are graspable to a satisfactory extent. When citing this paper, please use the full journal title Studies in History and Philosophy of Science. (shrink)

Has the rise of data-intensive science, or ‘big data’, revolutionized our ability to predict? Does it imply a new priority for prediction over causal understanding, and a diminished role for theory and human experts? I examine four important cases where prediction is desirable: political elections, the weather, GDP, and the results of interventions suggested by economic experiments. These cases suggest caution. Although big data methods are indeed very useful sometimes, in this paper’s cases they improve predictions either limitedly or not (...) at all, and their prospects of doing so in the future are limited too. (shrink)

Supervised machine learning has found its way into ever more areas of scientific inquiry, where the outcomes of supervised machine learning applications are almost universally classified as predictions. I argue that what researchers often present as a mere terminological particularity of the field involves the consequential transformation of tasks as diverse as classification, measurement, or image segmentation into prediction problems. Focusing on the case of machine-learning enabled poverty prediction, I explore how reframing a measurement problem as a prediction task alters (...) the primary epistemic aim of the application. Instead of measuring a property, machine learning developers conceive of their models as predicting a given measurement of this property. I argue that this _predictive reframing_ common to supervised machine learning applications is epistemically and ethically problematic, as it allows developers to externalize concerns critical to the epistemic validity and ethical implications of their model’s inferences. I further hold that the predictive reframing is not a necessary feature of supervised machine learning by offering an alternative conception of machine learning models as measurement models. An interpretation of supervised machine learning applications to measurement tasks as _automatically-calibrated model-based measurements_ internalizes questions of construct validity and ethical desirability critical to the measurement problem these applications are intended to and presented as solving. Thereby, this paper introduces an initial framework for exploring technical, historical, and philosophical research at the intersection of measurement and machine learning. (shrink)

Modern science’s ability to produce, store, and analyze big datasets is changing the way that scientific research is practiced. Philosophers have only begun to comprehend the changed nature of scientific reasoning in this age of “big data.” We analyze data-focused practices in biology and climate modeling, identifying distinct species of data-centric science: phenomena-laden in biology and phenomena-agnostic in climate modeling, each better suited for its own domain of application, though each entail trade-offs. We argue that data-centric practices in science are (...) not monolithic because the opportunities and challenges presented by big data vary across scientific domains. (shrink)

Reflexivity is, roughly, when studying or theorising about a target itself influences that target. Fragility is, roughly, when causal or other relations are hard to predict, holding only intermittently or fleetingly. Which is more important, methodologically? By going systematically through cases that do and do not feature each of them, I conclude that it is fragility that matters, not reflexivity. In this light, I interpret and extend the claims made about reflexivity in a recent paper by Jessica Laimann. I finish (...) by assessing the benefits and costs of a focus on reflexivity. (shrink)

A difference-making account of causality is proposed that is based on a counterfactual definition, but differs from traditional counterfactual approaches to causation in a number of crucial respects: it introduces a notion of causal irrelevance; it evaluates the truth-value of counterfactual statements in terms of difference-making; it renders causal statements background-dependent. On the basis of the fundamental notions 'causal relevance' and 'causal irrelevance', further causal concepts are defined including causal factors, alternative causes, and importantly inus-conditions. Problems and advantages of the (...) proposed account are discussed. Finally, it is shown how the account can shed new light on three classic problems in epistemology, the problem of induction, the logic of analogy, and the framing of eliminative induction. (shrink)

Large part of contemporary science is in fact technoscience, in the sense that it crucially depends on several technologies for the generation, collection, and analysis of data. This prompts a re-examination of the relations between science and technologies. In this essay, I advance the view that we’d better move beyond the ‘subordination view’ and the ‘instrumental’ view. The first aims to establish the primacy of science over technology, and the second uses technology instrumentally to support a realist position about theoretical (...) entities. I suggest that we should instead concentrate on how science and technology interact. This will reveal that technology has a poietic character, namely it actively partakes in the production of knowledge. But this poietic character can only be understood within the cognitive activity of scientific communities. Current research in molecular epidemiology, notably the projects funded within the ‘European exposome initiative’, serves as a motivation for such discussion and as an illustration of the claims made. (shrink)