Deep learning is currently the most prominent and widely successful method in artificial intelligence. Despite having played an active role in earlier artificial intelligence and neural network research, philosophers have been largely silent on this technology so far. This is remarkable, given that deep learning neural networks have blown past predicted upper limits on artificial intelligence performance—recognizing complex objects in natural photographs and defeating world champions in strategy games as complex as Go and chess—yet there remains no universally accepted explanation (...) as to why they work so well. This article provides an introduction to these networks as well as an opinionated guidebook on the philosophical significance of their structure and achievements. It argues that deep learning neural networks differ importantly in their structure and mathematical properties from the shallower neural networks that were the subject of so much philosophical reflection in the 1980s and 1990s. The article then explores several different explanations for their success and ends by proposing three areas of inquiry that would benefit from future engagement by philosophers of mind and science. (shrink)

Recent scholarship in philosophy of science and technology has shown that scientific and technological decision making are laden with values, including values of a social, political, and/or ethical character. This paper examines the role of value judgments in the design of machine-learning systems generally and in recidivism-prediction algorithms specifically. Drawing on work on inductive and epistemic risk, the paper argues that ML systems are value laden in ways similar to human decision making, because the development and design of ML systems (...) requires human decisions that involve tradeoffs that reflect values. In many cases, these decisions have significant—and, in some cases, disparate—downstream impacts on human lives. After examining an influential court decision regarding the use of proprietary recidivism-prediction algorithms in criminal sentencing, Wisconsin v. Loomis, the paper provides three recommendations for the use of ML in penal systems. (shrink)

Science is one of the most important forces in contemporary society. The most reliable source of knowledge about the world, science shapes the technological possibilities before us, informs public policy, and is crucial to measuring the efficacy of public policy. Yet it is not a simple repository of facts on which we can draw. It is an ongoing process of evidence gathering, discovery, contestation, and criticism. I will argue that an understanding of the nature of science and the scientific process (...) should be the central goal for scientific literacy, rather than a grasp of specific facts. With this understanding of science as a backdrop, the paper then turns to modes for citizen engagement with science. This paper articulates different ways citizens can engage with science, including four avenues for legitimate contestation of scientific claims. I then look more closely at contestation of science on the basis of values. That science can be legitimately contested by non-experts on a range of grounds means that science communication should not just aim at getting citizens to accept scientific claims, but rather to engage in a more robust two-way conversation about science. (shrink)

Many of the computing systems programmed using Machine Learning are opaque: it is difficult to know why they do what they do or how they work. Explainable Artificial Intelligence aims to develop analytic techniques that render opaque computing systems transparent, but lacks a normative framework with which to evaluate these techniques’ explanatory successes. The aim of the present discussion is to develop such a framework, paying particular attention to different stakeholders’ distinct explanatory requirements. Building on an analysis of “opacity” from (...) philosophy of science, this framework is modeled after accounts of explanation in cognitive science. The framework distinguishes between the explanation-seeking questions that are likely to be asked by different stakeholders, and specifies the general ways in which these questions should be answered so as to allow these stakeholders to perform their roles in the Machine Learning ecosystem. By applying the normative framework to recently developed techniques such as input heatmapping, feature-detector visualization, and diagnostic classification, it is possible to determine whether and to what extent techniques from Explainable Artificial Intelligence can be used to render opaque computing systems transparent and, thus, whether they can be used to solve the Black Box Problem. (shrink)

Over the last several years, there has been an explosion of interest and attention devoted to the problem of Uncertainty Quantification (UQ) in climate science—that is, to giving quantitative estimates of the degree of uncertainty associated with the predictions of global and regional climate models. The technical challenges associated with this project are formidable, and so the statistical community has understandably devoted itself primarily to overcoming them. But even as these technical challenges are being met, a number of persistent conceptual (...) difficulties remain. So why is UQ so important in climate science? UQ, I would like to argue, is first and foremost a tool for communicating knowledge from experts to policy makers in a way that is meant to be free from the influence of social and ethical values. But the standard ways of using probabilities to separate ethical and social values from scientific practice cannot be applied in a great deal of climate modeling, because the roles of values in creating the models cannot be discerned after the fact—the models are too complex and the result of too much distributed epistemic labor. I argue, therefore, that typical approaches for handling ethical/social values in science do not work well here. (shrink)

We address some of the epistemological challenges highlighted by the Critical Data Studies literature by reference to some of the key debates in the philosophy of science concerning computational modeling and simulation. We provide a brief overview of these debates focusing particularly on what Paul Humphreys calls epistemic opacity. We argue that debates in Critical Data Studies and philosophy of science have neglected the problem of error management and error detection. This is an especially important feature of the epistemology of (...) Big Data. In “Error” section we explain the main characteristics of error detection and correction along with the relationship between error and path complexity in software. In this section we provide an overview of conventional statistical methods for error detection and review their limitations when faced with the high degree of conditionality inherent to modern software systems. (shrink)

Simple idealized models seem to provide more understanding than opaque, complex, and hyper-realistic models. However, an increasing number of scientists are going in the opposite direction by utilizing opaque machine learning models to make predictions and draw inferences, suggesting that scientists are opting for models that have less potential for understanding. Are scientists trading understanding for some other epistemic or pragmatic good when they choose a machine learning model? Or are the assumptions behind why minimal models provide understanding misguided? In (...) this paper, using the case of deep neural networks, I argue that it is not the complexity or black box nature of a model that limits how much understanding the model provides. Instead, it is a lack of scientific and empirical evidence supporting the link that connects a model to the target phenomenon that primarily prohibits understanding. (shrink)

Critical Data Studies explore the unique cultural, ethical, and critical challenges posed by Big Data. Rather than treat Big Data as only scientifically empirical and therefore largely neutral phenomena, CDS advocates the view that Big Data should be seen as always-already constituted within wider data assemblages. Assemblages is a concept that helps capture the multitude of ways that already-composed data structures inflect and interact with society, its organization and functioning, and the resulting impact on individuals’ daily lives. CDS questions the (...) many assumptions about Big Data that permeate contemporary literature on information and society by locating instances where Big Data may be naively taken to denote objective and transparent informational entities. In this introduction to the Big Data & Society CDS special theme, we briefly describe CDS work, its orientations, and principles. (shrink)

The question of the relationship of the making of value judgments in a typically ethical sense to the methods and procedures of science has been discussed in the literature at least to that point which e. e. cummings somewhere refers to as “The Mystical Moment of Dullness.” Nevertheless, albeit with some trepidation, I feel that something more may fruitfully be said on the subject.

Machine learning (ML) has been praised as a tool that can advance science and knowledge in radical ways. However, it is not clear exactly how radical are the novelties that ML generates. In this article, I argue that this question can only be answered contextually, because outputs generated by ML have to be evaluated on the basis of the theory of the science to which ML is applied. In particular, I analyze the problem of novelty of ML outputs in the (...) context of molecular biology. In order to do this, I first clarify the nature of the models generated by ML. Next, I distinguish three ways in which a model can be novel (from the weakest to the strongest). Third, I dissect the way ML algorithms work and generate models in molecular biology and genomics. On these bases, I argue that ML is either a tool to identify instances of knowledge already present and codified, or to generate models that are novel in a weak sense. The notable contribution of ML to scientific discovery in the context of biology is that it can aid humans in overcoming potential bias by exploring more systematically the space of possible hypotheses implied by a theory. (shrink)

Philosophers continue to debate both the actual and the ideal roles of values in science. Recently, Eric Winsberg has offered a novel, model-based challenge to those who argue that the internal workings of science can and should be kept free from the influence of social values. He contends that model-based assignments of probability to hypotheses about future climate change are unavoidably influenced by social values. I raise two objections to Winsberg’s argument, neither of which can wholly undermine its conclusion but (...) each of which suggests that his argument exaggerates the influence of social values on estimates of uncertainty in climate prediction. I then show how a more traditional challenge to the value-free ideal seems tailor-made for the climate context. (shrink)

I examine the argument that scientific theories are typically 'underdetermined' by the data, an argument which has often been used to combat scientific realism. I deal with two objections to the underdetermination argument: (i) that the argument conflicts with the holistic nature of confirmation, and (ii) that the argument rests on an untenable theory/data dualism. I discuss possible responses to both objections, and argue that in both cases the proponent of underdetermination can respond in ways which are individually plausible, but (...) that the best response to the first objection conflicts with the best response to the second. Consequently underdetermination poses less of a problem for scientific realism than has often been thought. (shrink)

In this paper I argue for a distinction between subjective and value laden aspects of judgements showing why equating the former with the latter has the potential to confuse matters when the goal is uncovering the influence of political influences on scientific practice. I will focus on three separate but interrelated issues. The first concerns the issue of ‘verification’ in computational modelling. This is a practice that involves a number of formal techniques but as I show, even these allegedly objective (...) methods ultimately rely on subjective estimation and evaluation of different types of parameters. This has implications for my second point which relates to uncertainty quantification—an assessment of the degree of uncertainty present in a particular modelling scenario. I argue that while this practice also involves subjective elements, in no way does that detract from its status as an epistemic exercise. Finally I discuss the relation between accuracy and uncertainty and how each relates to judgements that embody social/ethical/political concerns, in particular those associated with high consequence systems. (shrink)

Reasons are given to justify the claim that computer simulations and computational science constitute a distinctively new set of scientific methods and that these methods introduce new issues in the philosophy of science. These issues are both epistemological and methodological in kind.

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

Although epistemic values have become widely accepted as part of scientific reasoning, non-epistemic values have been largely relegated to the "external" parts of science (the selection of hypotheses, restrictions on methodologies, and the use of scientific technologies). I argue that because of inductive risk, or the risk of error, non-epistemic values are required in science wherever non-epistemic consequences of error should be considered. I use examples from dioxin studies to illustrate how non-epistemic consequences of error can and should be considered (...) in the internal stages of science: choice of methodology, characterization of data, and interpretation of results. (shrink)

This is the first textbook on pattern recognition to present the Bayesian viewpoint. The book presents approximate inference algorithms that permit fast approximate answers in situations where exact answers are not feasible. It uses graphical models to describe probability distributions when no other books apply graphical models to machine learning. No previous knowledge of pattern recognition or machine learning concepts is assumed. Familiarity with multivariate calculus and basic linear algebra is required, and some experience in the use of probabilities would (...) be helpful though not essential as the book includes a self-contained introduction to basic probability theory. (shrink)

Science is one of the most important forces in contemporary society. The most reliable source of knowledge about the world, science shapes the technological possibilities before us, informs public policy, and is crucial to measuring the efficacy of public policy. Yet it is not a simple repository of facts on which we can draw. It is an ongoing process of evidence gathering, discovery, contestation, and criticism. I will argue that an understanding of the nature of science and the scientific process (...) should be the central goal for scientific literacy, rather than a grasp of specific (often disciplinary) facts. With this understanding of science as a backdrop, the paper then turns to modes for citizen engagement with science. This paper articulates different ways citizens can engage with science, including four avenues for legitimate contestation of scientific claims. I then look more closely at contestation of science on the basis of values. That science can be legitimately contested by non-experts on a range of grounds means that science communication should not just aim at getting citizens to accept scientific claims, but rather to engage in a more robust two-way conversation about science. (shrink)