In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory strategies will (...) lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

The feeling of clarity can be dangerously seductive. It is the feeling associated with understanding things. And we use that feeling, in the rough-and-tumble of daily life, as a signal that we have investigated a matter sufficiently. The sense of clarity functions as a thought-terminating heuristic. In that case, our use of clarity creates significant cognitive vulnerability, which hostile forces can try to exploit. If an epistemic manipulator can imbue a belief system with an exaggerated sense of clarity, then they (...) can induce us to terminate our inquiries too early — before we spot the flaws in the system. How might the sense of clarity be faked? Let’s first consider the object of imitation: genuine understanding. Genuine understanding grants cognitive facility. When we understand something, we categorize its aspects more easily; we see more connections between its disparate elements; we can generate new explanations; and we can communicate our understanding. In order to encourage us to accept a system of thought, then, an epistemic manipulator will want the system to provide its users with an exaggerated sensation of cognitive facility. The system should provide its users with the feeling that they can easily and powerfully create categorizations, generate explanations, and communicate their understanding. And manipulators have a significant advantage in imbuing their systems with a pleasurable sense of clarity, since they are freed from the burdens of accuracy and reliability. I offer two case studies of seductively clear systems: conspiracy theories; and the standardized, quantified value systems of bureaucracies. (shrink)

In this article I develop a model of theoretical understanding in science. This is a philosophical theory that specifies the conditions that are both necessary and sufficient for a scientist to satisfy the construction ‘S understands theory T ’. I first consider how this construction is preferable to others, then build a model of the requisite conditions on the basis of examples from elementary physics. I then show how this model of theoretical understanding can be made philosophically robust and provide (...) a more sophisticated account than we see from models of a similar kind developed by those working in the psychology of physics and artificial intelligence. 1 Introduction2 The Explicandum/Analysandum3 Analysis of ‘S understands T’ 4 The Inferential Model4.1 Which problems are we talking about?4.2 Does the solution have to be true?4.3 Does each cycle have to be correct for S to understand T?4.4 What is problem-solving reliability?4.5 Does every specific inference of a cycle have to be correct?4.6 Does each inference have to use a principle that is part of the theory?5 Theoretical Understanding and Conceptual Expertise: Empirical Considerations6 Conclusion: Inferential Model of Scientific Understanding and Our Traditional Problems. (shrink)

In this paper, I sketch a new model for the format of the content of understanding states, Compressible Graph Maximalism (CGM). In this model, the format of the content of understanding is graphical, and compressible. It thus combines ideas from approaches that stress the link between understanding and holistic structure (like as reported by Grimm (in: Ammon SGCBS (ed) Explaining Understanding: New Essays in Epistemollogy and the Philosophy of Science, Routledge, New York, 2016)), and approaches that emphasize the connection between (...) understanding and compression (like Wilkenfeld (Synthese 190(6):997–1016, 2018)). I argue that the combination of these ideas has several attractive features, and I defend the idea against some challenges. (shrink)

I argue for an account of grasping, or understanding that, on which we grasp via a higher-order mental act of Husserlian fulfillment. Fulfillment is the act of matching up the objects of our phenomenally presentational experiences with those of our phenomenally representational thought. Grasping-by-fulfilling is importantly different from standard epistemic aims, in part because it is phenomenal rather than inferential. (I endorse Bourget’s 2017 arguments to that effect.) I show that grasping-by-fulfilling cannot be a species of propositional knowledge or belief, (...) and that it is not essentially connected to justification. I motivate a revisionary epistemology on which achieving propositional knowledge and coming to grasp are dual epistemic aims. My account makes sense of a common occurrence—that we are often unmoved to act on our beliefs until we come to phenomenally experience them in some way. It also explains puzzling features of human inquiry. (shrink)

Explaining phenomena is a primary goal of science. Consequently, it is unsurprising that gaining a proper understanding of the nature of explanation is an important goal of science education. In order to properly understand explanation, however, it is not enough to simply consider theories of the nature of explanation. Properly understanding explanation requires grasping the relation between explanation and understanding, as well as how explanations can lead to scientific knowledge. This article examines the nature of explanation, its relation to understanding, (...) and how explanations are used to generate scientific knowledge via inferences to the best explanation. Studying these features and applications of explanation not only provides insight into a concept that is important for science education in its own right, but also sheds light on an aspect of recent debates concerning the so-called consensus view of nature of science (NOS). Once the relation between explanation, understanding, and knowledge is clear, it becomes apparent that science is unified in important ways. Seeing this unification provides some support for thinking that there are general features of NOS of the sort proposed by the consensus view and that teaching about these general features of NOS should be a goal of science education. (shrink)

This paper aims to propose an inferential and dynamic approach to understanding with models in biology. Understanding plays a central role in the practice of modeling. From its links with the other two central elements of scientific research, experimentation, and explanation, we show its epistemic relevance to the case of explanation in biology. Furthermore, by including the notion of understanding, we propose a non-referentialist perspective on scientific models, which is determined by their use.

In pedagogical contexts and in everyday life, we often come to believe something because it would best explain the data. What is it about the explanatory endeavor that makes it essential to everyday learning and to scientific progress? There are at least two plausible answers. On one view, there is something special about having true explanations. This view is highly intuitive: it’s clear why true explanations might improve one’s epistemic position. However, there is another possibility—it could be that the process (...) of seeking, generating, or evaluating explanations itself puts one in a better epistemic position, even when the outcome of the process is not a true explanation. In other words, it could be that accurate explanations are beneficial, or it could be that high-quality explaining is beneficial, where there is something about the activity of looking for an explanation that improves our epistemic standing. The main goal of this paper is to tease apart these two possibilities, both theoretically and empirically, which we align with “Inference to the Best Explanation” and “Explaining for the Best Inference”, respectively. We also provide some initial support for EBI and identify promising directions for future research. (shrink)

Several argue that truth cannot be science’s sole epistemic goal, for it would fail to do justice to several scientific practices that advance understanding. I challenge these arguments, but only after making a small concession: science’s sole epistemic goal is not truth as such; rather, its goal is finding true answers to relevant questions. Using examples from the natural and social sciences, I then show that scientific understanding’s epistemically valuable features are either true answers to relevant questions or a means (...) thereof. (shrink)

Many have argued that the human sciences feature a unique form of understanding that is absent from the natural sciences. However, in the last decade or so, epistemologists and philosop...

The literature on the nature of understanding can be divided into two broad camps. Explanationists believe that it is knowledge of explanations that is key to understanding. In contrast, their manipulationist rivals maintain that understanding essentially involves an ability to manipulate certain representations. The aim of this paper is to provide a novel knowledge based account of understanding. More specifically, it proposes an account of maximal understanding of a given phenomenon in terms of fully comprehensive and maximally well-connected knowledge of (...) it and of degrees of understanding in terms of approximations to such knowledge. It is completed by a contextualist semantics for outright attributions of understanding according to which an attribution of understanding is true of one just in case one knows enough about it to perform some contextually determined task. It is argued that this account has an edge over both its explanationist and manipulationist competitors. (shrink)

This paper connects two important debates in epistemology—to wit, on the goal of inquiry and on the nature of understanding—and offers a unified knowledge-based account of both.

What does it take to understand a phenomenon ideally, or to the highest conceivable extent? In this paper, we answer this question by arguing for five necessary conditions for ideal understanding: (i) representational accuracy, (ii) intelligibility, (iii) truth, (iv) reasonable endorsement, and (v) fitting. Even if one disagrees that there is some form of ideal understanding, these five conditions can be regarded as sufficient conditions for a particularly deep level of understanding. We then argue that grasping, novel predictions, and transparency (...) are not reasonable conditions for ideal understanding. (shrink)

I examine situations in which we say that different subjects have ‘different’, ‘competing’, or ‘conflicting understandings’ of a phenomenon. In order to make sense of such situations, we should turn our attention to an often neglected ambiguity in the word ‘understanding’. Whereas the notion of understanding that is typically discussed in philosophy is, to use Elgin’s terms, tethered to the facts, there is another notion of understanding that is not tethered in the same way. This latter notion is relevant because, (...) typically, talk of two subjects having ‘different’, ‘competing’, or ‘conflicting understandings’ of a phenomenon does not entail any commitment to the proposition that these subjects understand the phenomenon in the tethered sense of the word. This paper aims, first, to analyze the non-tethered notion of understanding, second, to clarify its relationship to the tethered notion, third, to explore what exactly goes on when ‘different’, ‘competing’, or ‘conflicting understandings’ clash and, fourth, to discuss the significance of such situations in our epistemic practices. In particular, I argue for a version of scientific pluralism according to which such situations are important because they help scientific communities achieve their fundamental epistemic goals—most importantly, the goal of understanding the world in the tethered sense. (shrink)

Name der Zeitschrift: Deutsche Zeitschrift für Philosophie Jahrgang: 67 Heft: 3 Seiten: 373-391.

This paper offers a normative account of the speech act of explanation with understanding as its norm. The previous accounts of the speech act of explanation rely on the factive notion of understanding and maintain that proper explanations require knowledge. I argue, however, that such accounts are too demanding and do not reflect the everyday practice of explanation and the attribution of understanding. Instead, I argue that the non-factive, objectual attitude of understanding is sufficient for a proper explanation. On the (...) normative level, explanations are governed by an audience-centred norm, i.e., they are sensitive to the epistemic position of the audience. According to the proposed account, an explanation is a communicative act in which one puts the audience in a position to understand the explained phenomenon. This proposal fits into the recent wave of applications of the normative account and makes space for the pluralism of illocutionary acts. (shrink)

Artificial general intelligence is not a new notion, but it has certainly been gaining traction in recent years, and academic as well as industry resources are redirected to research in AGI. The main reason for this is that current AI techniques are limited as they are designed to operate in specific problem-domains, following meticulous preparation. These systems cannot operate in an unknown environment or under conditions of uncertainty, reuse knowledge gained in another problem domain, or autonomously learn and understand the (...) problem-domain. We shall call AI systems capable of such feats artificial general intelligent systems. The three tasks of this paper are to provide a working definition of the term AGI, examine the “missing G”, i.e., the set of abilities that current AI systems lack and whose implementation will result in a basic AGI system, and consider different approaches, including a hybrid one, to a comprehensive solution for an AGI. (shrink)

In recent years, state-of-the-art artificial intelligence systems have started to show signs of what might be seen as human level intelligence. More specifically, large language models such as OpenAI’s GPT-3, and more recently Google’s PaLM and DeepMind’s GATO, are performing amazing feats involving the generation of texts. However, it is acknowledged by many researchers that contemporary language models, and more generally, learning systems, still lack important capabilities, such as understanding, reasoning and the ability to employ knowledge of the world and (...) common sense in order to reach or at least advance toward general intelligence. Some believe that scaling will eventually bring about these capabilities; others think that a different architecture is needed. In this paper, we focus on the latter, with the purpose of integrating a theoretical–philosophical conception of understanding as knowledge of dependence relations, with the high-level requirements and engineering design of a robust AI system, which integrates machine learning and symbolic components. (shrink)

The way in which one understands information and concepts, and the way a student works to develop this, is an individual aspect of learning that cannot be universally defined as (at least manifested) the same for everyone. ‘Understanding’ is a broad term, and the way one achieves understanding is dependent on the way that material is presented. In this article, we argue that the philosophy of science can be important to nursing education—in particular, by showing that the way we imbue (...) understanding might depend on the meaning of ‘understanding’. Diagrams and concept maps are meant to guide newly formed knowledge and connections to develop proper thinking (e.g., the order in which nursing students must prioritize data) that a student requires in the field. We argue that whether or not an image/diagram/concept map confers understanding will depend on both what the object is and what we mean by ‘understanding’. (shrink)

This paper aims to clarify the nature of understanding in medicine. The first part describes in more detail what it means to understand something and links a type of understanding (i.e., objectual understanding) to explanations. The second part proceeds to investigate what objectual understanding of a disease (i.e., biomedical understanding) requires by considering the case of scurvy from the history of medi- cine. The main hypothesis is that grasping a mechanistic explanation of a condi- tion is necessary for a biomedical (...) understanding of that condition. The third part of the paper argues that biomedical understanding is necessary, but not sufficient for understanding in a clinical context (i.e., clinical understanding). The hypothe- sis is that clinical understanding combines biomedical understanding of a disease or pathological condition with understanding illness, which involves some degree of personal understanding of the patient. It is argued that, in many cases, clinical understanding necessitates adopting a particular second-personal stance and using cognitive resources in addition to those involved in biomedical understanding. (shrink)

We provide two programmatic frameworks for integrating philosophical research on understanding with complementary work in computer science, psychology, and neuroscience. First, philosophical theories of understanding have consequences about how agents should reason if they are to understand that can then be evaluated empirically by their concordance with findings in scientific studies of reasoning. Second, these studies use a multitude of explanations, and a philosophical theory of understanding is well suited to integrating these explanations in illuminating ways.

The paper provides a systematic overview of recent debates in epistemology and philosophy of science on the nature of understanding. We explain why philosophers have turned their attention to understanding and discuss conditions for “explanatory” understanding of why something is the case and for “objectual” understanding of a whole subject matter. The most debated conditions for these types of understanding roughly resemble the three traditional conditions for knowledge: truth, justification and belief. We discuss prominent views about how to construe these (...) conditions for understanding, whether understanding indeed requires conditions of all three types and whether additional conditions are needed. (shrink)