Most recent philosophical thought about the scientific representation of the world has focused on dyadic relationships between language-like entities and the world, particularly the semantic relationships of reference and truth. Drawing inspiration from diverse sources, I argue that we should focus on the pragmatic activity of representing, so that the basic representational relationship has the form: Scientists use models to represent aspects of the world for specific purposes. Leaving aside the terms "law" and "theory," I distinguish principles, specific conditions, models, (...) hypotheses, and generalizations. I argue that scientists use designated similarities between models and aspects of the world to form both hypotheses and generalizations. (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)

Some machine learning models, in particular deep neural networks (DNNs), are not very well understood; nevertheless, they are frequently used in science. Does this lack of understanding pose a problem for using DNNs to understand empirical phenomena? Emily Sullivan has recently argued that understanding with DNNs is not limited by our lack of understanding of DNNs themselves. In the present paper, we will argue, _contra_ Sullivan, that our current lack of understanding of DNNs does limit our ability to understand with (...) DNNs. Sullivan’s claim hinges on which notion of understanding is at play. If we employ a weak notion of understanding, then her claim is tenable, but rather weak. If, however, we employ a strong notion of understanding, particularly explanatory understanding, then her claim is not tenable. (shrink)

Under what conditions does machine learning (ML) model opacity inhibit the possibility of explaining and understanding phenomena? In this article, I argue that nonepistemic values give shape to the ML opacity problem even if we keep researcher interests fixed. Treating ML models as an instance of doing model-based science to explain and understand phenomena reveals that there is (i) an external opacity problem, where the presence of inductive risk imposes higher standards on externally validating models, and (ii) an internal opacity (...) problem, where greater inductive risk demands a higher level of transparency regarding the inferences the model makes. (shrink)

Scientists depend on complex computational systems that are often ineliminably opaque, to the detriment of our ability to give scientific explanations and detect artifacts. Some philosophers have s...

Science is replete with falsehoods that epistemically facilitate understanding by virtue of being the very falsehoods they are. In view of this puzzling fact, some have relaxed the truth requirement on understanding. I offer a factive view of understanding that fully accommodates the puzzling fact in four steps: (i) I argue that the question how these falsehoods are related to the phenomenon to be understood and the question how they figure into the content of understanding it are independent. (ii) I (...) argue that the falsehoods do not figure into the understanding’s content by being elements of its periphery or core. (iii) Drawing lessons from case studies, I argue that the falsehoods merely enable understanding. When working with such falsehoods, only the truths we extract from them are elements of the content of our understanding. (iv) I argue that the extraction view is compatible with the thesis that falsehoods can have an epistemic value by virtue of being the very falsehoods they are. (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)

One puzzle concerning highly idealized models is whether they explain. Some suggest they provide so-called ‘how-possibly explanations’. However, this raises an important question about the nature of how-possibly explanations, namely what distinguishes them from ‘normal’, or how-actually, explanations? I provide an account of how-possibly explanations that clarifies their nature in the context of solving the puzzle of model-based explanation. I argue that the modal notions of actuality and possibility provide the relevant dividing lines between how-possibly and how-actually explanations. Whereas how-possibly (...) explanations establish claims of possible explanations, how-actually explanations establish claims of actual ones. Models, in turn, simply provide evidence for these claims. (shrink)

Science relies on models and idealizations that are known not to be true. Even so, science is epistemically reputable. To accommodate science, epistemology should focus on understanding rather than knowledge and should recognize that the understanding of a topic need not be factive. This requires reconfiguring the norms of epistemic acceptability. If epistemology has the resources to accommodate science, it will also have the resources to show that art too advances understanding.

Idealization and abstraction are central concepts in the philosophy of science and in science itself. My goal in this paper is suggest an account of these concepts, building on and refining an existing view due to Jones Idealization XII: correcting the model. Idealization and abstraction in the sciences, vol 86. Rodopi, Amsterdam, pp 173–217, 2005) and Godfrey-Smith Mapping the future of biology: evolving concepts and theories. Springer, Berlin, 2009). On this line of thought, abstraction—which I call, for reasons to be (...) explained, abstractness—involves the omission of detail, whereas idealization consists in a deliberate mismatch between a description and the world. I will suggest that while the core idea underlying these authors’ view is correct, they make several assumptions and stipulations that are best avoided. For one thing, they tie abstractness too close to truth. For another, they do not allow sufficient room to the difference between idealization and error. Taking these points into account leads to a refined account of the distinction, in which abstractness is seen in terms of relative richness of detail, and idealization is seen as closely connected with the knowledge and intentions of idealizers. I lay out these accounts in turn, and then discuss the relationship between the two concepts, and several other upshots of the present way of construing the distinction. (shrink)

Science is the study of our world, as it is in its messy reality. Nonetheless, science requires idealization to function—if we are to attempt to understand the world, we have to find ways to reduce its complexity. Idealization and the Aims of Science shows just how crucial idealization is to science and why it matters. Beginning with the acknowledgment of our status as limited human agents trying to make sense of an exceedingly complex world, Angela Potochnik moves on to explain (...) how science aims to depict and make use of causal patterns—a project that makes essential use of idealization. She offers case studies from a number of branches of science to demonstrate the ubiquity of idealization, shows how causal patterns are used to develop scientific explanations, and describes how the necessarily imperfect connection between science and truth leads to researchers’ values influencing their findings. The resulting book is a tour de force, a synthesis of the study of idealization that also offers countless new insights and avenues for future exploration. (shrink)

Philosophers have recently focused on critical, epistemological challenges that arise from the opacity of deep neural networks. One might conclude from this literature that doing good science with opaque models is exceptionally challenging, if not impossible. Yet, this is hard to square with the recent boom in optimism for AI in science alongside a flood of recent scientific breakthroughs driven by AI methods. In this paper, I argue that the disconnect between philosophical pessimism and scientific optimism is driven by a (...) failure to examine how AI is actually used in science. I show that, in order to understand the epistemic justification for AI-powered breakthroughs, philosophers must examine the role played by deep learning as part of a wider process of discovery. The philosophical distinction between the 'context of discovery' and the 'context of justification' is helpful in this regard. I demonstrate the importance of attending to this distinction with two cases drawn from the scientific literature, and show that epistemic opacity need not diminish AI's capacity to lead scientists to significant and justifiable breakthroughs. (shrink)

Deep neural networks have become increasingly successful in applications from biology to cosmology to social science. Trained DNNs, moreover, correspond to models that ideally allow the prediction of new phenomena. Building in part on the literature on ‘eXplainable AI’, I here argue that these models are instrumental in a sense that makes them non-explanatory, and that their automated generation is opaque in a unique way. This combination implies the possibility of an unprecedented gap between discovery and explanation: When unsupervised models (...) are successfully used in exploratory contexts, scientists face a whole new challenge in forming the concepts required for understanding underlying mechanisms. (shrink)

How can a model that stops short of representing the whole truth about the causal production of a phenomenon help us to understand the phenomenon? I answer this question from the perspective of what I call the simple view of understanding, on which to understand a phenomenon is to grasp a correct explanation of the phenomenon. Idealizations, I have argued in previous work, flag factors that are casually relevant but explanatorily irrelevant to the phenomena to be explained. Though useful to (...) the would-be understander, such flagging is only a first step. Are there any further and more advanced ways that idealized models aid understanding? Yes, I propose: the manipulation of idealized models can provide considerable insight into the reasons that some causal factors are difference-makers and others are not, which helps the understander to grasp the nature of explanatory connections and so to better grasp the explanation itself. (shrink)

In this paper, I propose that the debate in epistemology concerning the nature and value of understanding can shed light on the role of scientific idealizations in producing scientific understanding. In philosophy of science, the received view seems to be that understanding is a species of knowledge. On this view, understanding is factive just as knowledge is, i.e., if S knows that p, then p is true. Epistemologists, however, distinguish between different kinds of understanding. Among epistemologists, there are those who (...) think that a certain kind of understanding—objectual understanding—is not factive, and those who think that objectual understanding is quasi-factive. Those who think that understanding is not factive argue that scientific idealizations constitute cognitive success, which we then consider as instances of understanding, and yet they are not true. This paper is an attempt to draw lessons from this debate as they pertain to the role of idealizations in producing scientific understanding. I argue that scientific understanding is quasi-factive. (shrink)

This article discusses minimal model explanations, which we argue are distinct from various causal, mechanical, difference-making, and so on, strategies prominent in the philosophical literature. We contend that what accounts for the explanatory power of these models is not that they have certain features in common with real systems. Rather, the models are explanatory because of a story about why a class of systems will all display the same large-scale behavior because the details that distinguish them are irrelevant. This story (...) explains patterns across extremely diverse systems and shows how minimal models can be used to understand real systems. (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)