In a recent paper, Erasmus et al. (2021) defend the idea that the ambiguity of the term “explanation” in explainable AI (XAI) can be solved by adopting any of four different extant accounts of explanation in the philosophy of science: the Deductive Nomological, Inductive Statistical, Causal Mechanical, and New Mechanist models. In this chapter, I show that the authors’ claim that these accounts can be applied to deep neural networks as they would to any natural phenomenon is mistaken. I also (...) provide a more general argument as to why the notion of explainability as it is currently used in the XAI literature bears little resemblance to the traditional concept of scientific explanation. It would be more fruitful to use the label “understandable AI” to avoid the confusion that surrounds the goal and purposes of XAI. In the second half of the chapter, I argue for a pragmatic conception of understanding that is better suited to play the central role attributed to explanation in XAI. Following De Regt (2017) and Kuorikoski and Ylikoski (2015), the conditions of satisfaction for understanding an ML system are fleshed out in terms of an agent’s success in using the system. (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)

Starting with the slogan that understanding is a ‘knowledge of causes’, Stephen Grimm and John Greco have argued that understanding comes from a knowledge of dependence relations. Grounding is the trendiest dependence relation on the market, and if Grimm and Greco are correct, then instances of grounding should also give rise to understanding. In this paper, I will show that this prediction is correct – grounding does indeed generate understanding in just the way that Grimm and Greco anticipate. However, grounding (...) examples of understanding also show that Grimm and Greco are not telling the full story when it comes to understanding. Understanding can only be generated by a particular subset of dependence relations – those dependence relations that are also explanatory. Grimm and Greco should thus appeal to a privileged class of dependence relations, relations like grounding that can also give rise to explanation. (shrink)

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 present an account of group competence that is explicitly framed for cases of epistemic performances. According to it, we must consider group epistemic competence as the group agents’ capacity to produce knowledge, and not the result of the summation of its individual members’ competences to produce knowledge. Additionally, I contend that group competence must be understood in terms of group normative status. To introduce my view, I present Jesper Kallestrup’s denial that group competence involves anything over (...) and beyond the aggregation of individual competences. I have divided my response into two parts. First, I compare two conceptions of competence from Ernest Sosa’s reliabilist virtue epistemology, Thinking about oneself: The place and value of reflection in philosophy and psychology, Springer, 2019) and David Löwenstein’s account of know-how. Second, I take the results from this comparison and apply them to the issue of group know-how, by the hand of Orestis Palermos and Deborah Tollefsen’s twofold approach to the topic Socially extended epistemology, Oxford University Press, 2018). Finally, I return to Kallestrup’s denial to make my point in favour of the conception of genuine group competence as the group normative status to achieve success. (shrink)

ABSTRACT: According to a traditional account, understanding why X occurred is equivalent to knowing that X was caused by Y. This paper defends the account against a major objection, viz., knowing-that is not sufficient for understanding-why, for understanding-why requires a kind of grasp while knowledge-that does not. I discuss two accounts of grasp in recent literature and argue that if either is true, then knowing that X was caused by Y entails at least a rudimentary understanding of why X occurred. (...) If my defense is successful, it would cast doubt on an influential account of the epistemic value of understanding. (shrink)

Hot-cold empathy gaps are a pervasive phenomena wherein one’s predictions about others tend to skew ‘in the direction’ of one’s own current visceral states. For instance, when one predicts how hungry someone else is, one’s prediction will tend to reflect one’s own current hunger state. These gaps also obtain intrapersonally, when one attempts to predict what one oneself would do at a different time. In this paper, we do three things: We draw on empirical evidence to argue that so-called hot-cold (...) empathy gaps arise when one projects one’s own current state into a simulation about another. Second, we argue that this process does not typically confer knowledge, even when the predictions it produces happen to be accurate. Third, we suggest that these results can be used to develop a challenge for L.A. Paul's view that authentic action sometimes requires subjective knowledge of one’s own values and how these values relate to relevant outcomes. We then sketch an alternative view of the epistemic grounds of authenticity, one on which authenticity requires a kind of understanding. The relevant form of understanding can be achieved by subjective knowledge but can also be achieved elsewise, such as through testimony from a close friend about what one values. (shrink)

The philosophical interest in the nature, value, and varieties of human understanding has swelled in recent years. This article will provide an overview of new research in the epistemology of understanding, with a particular focus on the following questions: What is understanding and why should we care about it? Is understanding reducible to knowledge? Does it require truth, belief, or justification? Can there be lucky understanding? Does it require ‘grasping’ or some kind of ‘know-how’? This cluster of questions has largely (...) set the research agenda for the study of understanding in epistemology. This article will conclude by discussing some varieties of understanding and highlight directions for future research. (shrink)

Many AI systems that make important decisions are black boxes: how they function is opaque even to their developers. This is due to their high complexity and to the fact that they are trained rather than programmed. Efforts to alleviate the opacity of black box systems are typically discussed in terms of transparency, interpretability, and explainability. However, there is little agreement about what these key concepts mean, which makes it difficult to adjudicate the success or promise of opacity alleviation methods. (...) I argue for a unified account of these key concepts that treats the concept of understanding as fundamental. This allows resources from the philosophy of science and the epistemology of understanding to help guide opacity alleviation efforts. A first significant benefit of this understanding account is that it defuses one of the primary, in-principle objections to post hoc explainable AI (XAI) methods. This “rationalization objection” argues that XAI methods provide mere rationalizations rather than genuine explanations. This is because XAI methods involve using a separate “explanation” system to approximate the original black box system. These explanation systems function in a completely different way than the original system, yet XAI methods make inferences about the original system based on the behavior of the explanation system. I argue that, if we conceive of XAI methods as idealized scientific models, this rationalization worry is dissolved. Idealized scientific models misrepresent their target phenomena, yet are capable of providing significant and genuine understanding of their targets. (shrink)

The counterfactual approach to explainable AI (XAI) seeks to provide understanding of AI systems through the provision of counterfactual explanations. In a recent systematic review, Chou et al. (Inform Fus 81:59–83, 2022) argue that the counterfactual approach does not clearly provide causal understanding. They diagnose the problem in terms of the underlying framework within which the counterfactual approach has been developed. To date, the counterfactual approach has not been developed in concert with the approach for specifying causes developed by Pearl (...) (Causality: Models, reasoning, and inference. Cambridge University Press, 2000) and Woodward (Making things happen: A theory of causal explanation. Oxford University Press, 2003). In this paper, I build on Chou et al.’s work by applying the Pearl-Woodward approach. I argue that the standard counterfactual approach to XAI is capable of delivering causal understanding, but that there are limitations on its capacity to do so. I suggest a way to overcome these limitations. (shrink)

Philosophical debates on how to account for the progress of science have traditionally divided along the realism-anti-realism axis. Relatively recent developments in epistemology, however, have opened up a new knowledge-understanding axis to the debate. This chapter presents a novel understanding-based account of scientific progress that takes its motivation from problem-solving practices in science. Problem-solving is characterized as a means of measuring degree of understanding, which is argued to be the principal epistemic (or cognitive) aim of science, over and against knowledge. (...) This account is then defended against knowledge reductionism by showing how each plays an important and distinctive role in science. (shrink)

In this talk I present the main results from Anta (2021), namely, that the theoretical division between Boltzmannian and Gibbsian statistical mechanics should be understood as a separation in the epistemic capabilities of this physical discipline. In particular, while from the Boltzmannian framework one can generate powerful explanations of thermal processes by appealing to their microdynamics, from the Gibbsian framework one can predict observable values in a computationally effective way. Finally, I argue that this statistical mechanical schism contradicts the Hempelian (...) (1958) thesis that the predictive power of a scientific theory is directly proportional to its explanatory potential, and vice versa. (shrink)