Questions about idealizations in science are often framed along the lines of, How can science be so effective when it gets so much wrong? Rice’s book, Leveraging Distortions: Explanation, Idealization, and Universality in Science offers a refinement on this framing, where we need not commit to the premise that idealizations are, in fact, wrong, that they need to be contained to the irrelevant parts of a model, or should be explained away as mere appearance. Rice takes a holist approach in which idealization is more like a process by which models as a whole are leveraged into better fit with their targets. Idealizations should not be carved out one by one on this approach; they make sense in the context of the models in which they figure, and they distort in ways that illuminate features like universal behavior in the systems being modeled. This is a refreshing approach to how idealizations work, one that does not require the common presupposition that idealizations are simply false.By universality, Rice means “the stability of certain patterns or behaviors across systems that are heterogeneous in their features. Universality classes are, then, just the group of systems that will display those universal patterns or behaviors” (155). Universality enables a more abstract description of systems than what scientists may have started with, and this process of making the description of the behavior more universal serves to identify common causal structures implemented in very different physical mediums. Different descriptions of causal relata facilitate identification of more unifying patterns of behavior. Given how often philosophers think of abstraction as somehow eliminating causation, by identifying causation too strongly with microphysical details, universality is a helpful way to bring the process of abstracting description back into contact with the way in which models inevitably involve causal structure, and how that causal structure itself can be better understood by connecting classes of systems with heterogeneous physical media and similar behavior, by showing how the more abstract descriptions of causal structure are deployed in each.There are two specific features of his view that set Rice’s book apart from most other contemporary views on idealizations. The first is the explicit emphasis on holism. Often, idealizations are isolated from models and then assessed on their own after extraction from the modeling context in which they were made. In evaluating idealizations as individual propositions removed from surrounding context, it is somewhat unsurprising that many look inaccurate. Rice aptly shows how idealization plays a key role in identifying universality behavior by distorting a whole, undecomposed model. This focus on holism and the role idealizations play in a larger modeling context helps Rice’s treatment of idealizations stand apart from many others, including those he explicitly engages with, such as Angela Potochnik (2017), Michael Strevens (2011), and Kareem Khalifa (2017). This approach fits better with the usage of idealizations in science by not needing to explain away the widespread reliance on idealization in so many sciences. Even if one thinks the other accounts are successful in trying to explain why idealizations can be used in science despite falsity and misrepresentation, there is something uncomfortable about explaining such widespread use of them by framing it as apparently irrational. Rice’s account does not require starting from a framing where scientists rampantly engage in apparently irrational practices and then explain why it is not as bad as it looks. Instead of using idealizations despite falsity, idealizations are part of a coherent package that can be used for explanatory leverage.The second feature that sets his view apart follows from this: idealizations are a tool to be actively used, not peculiarities to be explained away or dubious commitments to be minimized. Too often, idealizations are treated as some kind of representational failure, a compensation for epistemic limitations. In a more epistemically perfect world, on such thinking, idealizations could be done away with. Rice turns this around: idealizations are not something we put up with or have to be resigned to; they are a key tool to be used in positive ways to generate explanations and for building bodies of understanding. This is where the “leveraging” part of the title comes in: idealizations are actively relied on to achieve modeling techniques that would be impossible otherwise. They are a lever by which to torque a model into better alignment. This positive feature of idealizations accounts for the advantageous character of idealizations as a feature, not a bug.While Rice is, in my view, exactly right to reject these background presuppositions about the falsity of idealizations, I would also add that he could go further in this regard; the book would benefit from more explicit discussion of what he means by truth or falsity. There are pragmatist versions of truth, for example, that are quite consonant with his final view, so that it need not be framed as a puzzle that false statements somehow work to return genuine knowledge. Idealizations are usually presupposed to be false; authors like Potochnik (2017), in fact, define them as false, such that if it is an idealization, then by definition, it could not be true. Rice does not seem to endorse this, yet accurate representation is left hanging somewhat. A discussion of epistemic standards of veridicality that should be used for the holistic evaluation the of models, and the ways in which various identifiable components of those models accomplish this without decomposition, would strengthen his overall push toward a more explicit and foregrounded holism about models and his claims in chapter 8 about realism.That is quite mild, as critical remarks go, and most of the book is full of detailed examples and other discussions that don’t require a further discussion of truth. There is a lot covered in this book, much of which Rice has written about elsewhere and some of which he extends, refines, or adds to in new ways in the book. In the introduction, Rice stakes the main claim that pervasive distortion doesn’t just happen in science; it is central to science working as well as it does that such distortion take place. This sets up the later chapters on universality as a behavior that can be instantiated in physically heterogeneous systems and identified with more abstract (and distorting) descriptions of those systems. This introduction does a good job of situating why this alternative stance toward idealizations as pervasive distortions that are used for purposes that cannot be served with other tools differs from approaches where idealizations are considered after isolating them from modeling contexts and then evaluating them as false yet useful.Chapter 2 discusses what Rice calls the causal or causal-mechanical paradigm in literature on explanation. The causal approach, as he characterizes it, explains an event by giving the relevant factors in the event’s causal history. Wesley Salmon, James Woodward, Michael Strevens, Angela Potochnik, and the wide range of authors working in the ‘new mechanisms’ discussion are highlighted as examples of this. Rice is right to highlight how widespread discussions of causation are in discussions of explanation, and it is great to see Salmon given more credit. At the same time, this chapter lumps together some heterogeneous approaches, like Woodward’s (2005) account of causal explanation, for example. Woodward gives an account of those explanations that are causal without claiming that this is exhaustive of all explanation; there could be noncausal explanations, but he just isn’t discussing this possibility. Strevens (2011), in contrast, takes himself to be providing a complete account of explanation based on causation; Potochnik (2017), as well, offers an account of explanation in which causation, in the form of causal patterns, plays a necessary role.Chapter 3 follows this up by demonstrating with a series of examples a number of explanations that do not involve causation. This chapter may be overkill if the goal was to demonstrate that not all explanations need be causal explanations, since some of the apparent targets, like Woodward, already agree with this, and there is a lot of interesting work on distinctively mathematical explanations that highlights how they contrast with and complement causal explanations that he does not engage with. But as a collection of examples of noncausal explanation, this chapter has new material to add to existing examples, especially to the examples of distinctively statistical explanations given by Marc Lange (2016).In chapter 4, Rice lays out his own counterfactual account of explanation and contrasts it with other such accounts. He offers three criteria that any such account should meet that will be useful in these discussions (93), even if one does not want to adopt Rice’s own particular account. The details of Rice’s own account here seem compressed, and if one just reads this chapter, it is hard to see how this is supposed to work and be a genuine move forward. The later chapters, especially chapters 6, 7, and 9, show how the account works when applied, which is illuminating. It would thus be useful, for instance if teaching from the book in a seminar, to pair chapter 4 with one of these further chapters, especially chapter 6.Chapter 5 is brief, focused on how decomposition of models into subcomponents that are then treated separately simply doesn’t work for most models. Rice makes some very clear points about why models must be treated holistically, solidifying his point about idealizations as distortions in those models that don’t make sense when taken out of that context through attempts at decomposition.Universality, a term of art here that follows on Rice’s other work (see, e.g., Rice 2018, 2019; Batterman and Rice 2014), is given detailed treatment in chapter 6. This chapter lays out some detailed case studies and illustrates how the holistic distortion involved in idealization is what conveys or captures the specifically modal information in a model. Chapter 7 continues with themes Rice has written about elsewhere: multiscale models and how universality fits into considerations of scale and renormalization.Chapter 8 moves on to consider how models can provide understanding even when they do not do so by providing explanations. Rice’s examples involve cases where scientists have incomplete explanations, so some might consider these to be explanations already, since one need not require that an explanation be fully complete in order to count as an explanation. This chapter also connects understanding to realism and scientific progress. Idealizations have often been treated as failures for realism, where an otherwise successful model is purportedly decomposed into elements, some of which are clearly not literally representationally accurate in the way one might suppose necessary to be a realist about that component (another way in which naive correspondence treatments of truth sneak into philosophy of science by way of assuming that bits of models should map one-to-one to bits of the world and that realism about a model fails if there are idealizations that don’t map in this simplified way). He draws on his own account of factive understanding, in the first part of the chapter, to lay out an alternative approach to realism where the focus is not on isolated model components but on the body of understanding that models produce for scientists. This body of understanding, which again requires holism, can serve as an epistemic basis for realism about the behavior thus understood.Finally, in chapter 9, Rice brings together all the themes in the book and makes the clearest case yet for how idealizations are used as “holistic distortions” that are not merely part of science but central and positively contributory to the success of modeling techniques in providing both explanation and understanding. This chapter is a great conclusion to bring together the different topics in the book. Many of the other topics are ones Rice has written about elsewhere, and this concluding chapter helps make sense of the synoptic project into which all this work fits. If one were teaching with this this text, this might be a good chapter to start with rather than to end with.Overall, this book does a nice job of bringing together Rice’s previous work while also extending that work with new examples and ones worked out in more detail, and connecting the different topics in a cohesive way around the orientation toward holism and idealizations as holistic model distortions. This makes it a great addition to a range of contemporary discussions around explanation, models, understanding, and realism, and a good starting point for graduate students to get into these topics.

中文翻译：

---

利用扭曲：科学中的解释、理想化和普遍性

关于科学理想化的问题通常是这样的：当科学犯了这么多错误时，它怎么能如此有效呢？赖斯的书《利用扭曲：科学中的解释、理想化和普遍性》对这一框架进行了改进，我们不需要承诺这样的前提：理想化实际上是错误的，它们需要包含在科学的不相关部分中。模型，或者应该被解释为纯粹的外观。赖斯采用了一种整体方法，其中理想化更像是一个过程，通过该过程，模型作为一个整体被利用以更好地适应其目标。不应该对这种方法一一进行理想化；它们在它们所描绘的模型的背景下是有意义的，并且它们以阐明被建模系统中的普遍行为等特征的方式扭曲。这是一种关于理想化如何运作的令人耳目一新的方法，它不需要理想化完全错误的共同前提。 赖斯所说的普遍性是指“在特征不同的系统中某些模式或行为的稳定性。那么，普遍性类就是将显示这些普遍模式或行为的一组系统”（155）。普遍性使得对系统的描述比科学家可能开始时更加抽象，并且使行为描述更加普遍的过程有助于识别在非常不同的物理介质中实现的常见因果结构。对因果关系的不同描述有助于识别更统一的行为模式。鉴于哲学家经常认为抽象是在某种程度上消除因果关系，通过将因果关系与微观物理细节过于强烈地识别在一起，普遍性是一种有用的方式，可以使抽象描述的过程重新与模型不可避免地涉及因果结构的方式联系起来，以及如何通过将系统类别与异构物理介质和相似行为联系起来，通过展示因果结构的更抽象描述如何在每个系统中部署，可以更好地理解因果结构本身。他的观点有两个具体特征，使赖斯的书与大多数人的书不同。其他当代关于理想化的观点。首先是明确强调整体论。通常，理想化与模型分离，然后从生成模型的建模环境中提取后自行评估。在将理想化评估为脱离周围环境的个体命题时，许多看起来不准确也就不足为奇了。赖斯恰当地展示了理想化如何通过扭曲整个未分解的模型来在识别普遍性行为中发挥关键作用。这种对整体论和理想化在更大的建模背景中发挥的作用的关注，有助于莱斯对理想化的处理与许多其他人不同，包括他明确参与的那些，例如 Angela Potochnik (2017)，迈克尔·斯特文斯 (2011) 和卡里姆·哈利法 (2017)。这种方法更适合科学中理想化的使用，因为不需要解释许多科学中对理想化的广泛依赖。即使人们认为其他说法成功地试图解释为什么理想化可以在科学中使用，尽管存在虚假和歪曲，但通过将其定义为明显的非理性来解释理想化的如此广泛使用，还是会让人感到不舒服。赖斯的叙述不需要从这样一个框架开始：科学家们猖獗地从事明显不合理的做法，然后解释为什么它并不像看起来那么糟糕。尽管存在虚假，但理想化不是使用理想化，而是可以用于解释杠杆的连贯包的一部分。使他的观点与众不同的第二个特征如下：理想化是一种可以积极使用的工具，而不是需要解释或消除的特殊性。尽量减少可疑的承诺。很多时候，理想化被视为某种表征失败，是对认知局限性的补偿。在一个认知上更加完美的世界中，基于这种思考，理想化可以被消除。赖斯扭转了这一局面：理想化不是我们必须忍受或必须屈服的东西；我们必须将理想化视为现实。它们是一个重要的工具，可以积极地使用它们来产生解释和建立理解体系。这就是标题中“利用”部分的用武之地：积极依赖理想化来实现否则不可能实现的建模技术。它们是一个杠杆，可以通过扭矩使模型更好地对齐。理想化的这一积极特征说明了理想化作为一种​​特征而不是缺陷的有利特征。虽然在我看来，赖斯拒绝这些关于理想化虚假性的背景预设是完全正确的，但我还想补充一点，他可以走得更远在这方面; 对他所说的真理或谬误的含义进行更明确的讨论将使这本书受益匪浅。例如，真理有一些实用主义版本，与他的最终观点非常一致，因此不必将错误陈述以某种方式返回真实知识这一难题视为一个谜题。理想化通常被认为是错误的。事实上，像 Potochnik (2017) 这样的作者将它们定义为错误，这样如果它是一种理想化，那么根据定义，它不可能是真的。赖斯似乎并不赞同这一点，但准确的表述却有些悬而未决。对应用于模型整体评估的真实性认知标准的讨论，以及这些模型的各种可识别组件在不分解的情况下实现这一目标的方式，将加强他对模型和模型的更明确和更突出的整体论的整体推动。他在第八章中关于现实主义的主张。正如批评言论所说，这是相当温和的，本书的大部分内容都充满了详细的例子和其他讨论，不需要进一步讨论真相。本书涵盖了很多内容，其中大部分内容赖斯在其他地方写过，并且他在书中以新的方式扩展、完善或添加了其中一些内容。在引言中，赖斯提出了一个主要观点：普遍存在的扭曲不仅发生在科学领域；也存在于科学领域。它是科学工作的核心，而且这种扭曲的发生也确实如此。这将后面关于普遍性的章节设置为一种行为，可以在物理异构系统中实例化，并用这些系统的更抽象（和扭曲）的描述来识别。这个介绍很好地解释了为什么这种对理想化的另一种立场是普遍的扭曲，用于无法用其他工具实现的目的，这与将理想化与建模上下文隔离然后将其评估为错误但有用之后考虑理想化的方法不同第二章讨论赖斯在解释文献中所说的因果或因果机械范式。正如他所描述的那样，因果方法通过给出事件因果历史中的相关因素来解释事件。Wesley Salmon、James Woodward、Michael Strevens、Angela Potochnik 以及从事“新机制”讨论的众多作者都是这方面的例子。赖斯正确地强调了在解释讨论中对因果关系的讨论是多么广泛，而且很高兴看到鲑鱼得到更多的信任。与此同时，本章将一些异质的方法集中在一起，例如伍德沃德（Woodward，2005）对因果解释的解释。伍德沃德对那些因果关系的解释进行了说明，但并未声称这已是所有解释的详尽解释；可能存在非因果解释，但他只是没有讨论这种可能性。相比之下，Strevens（2011）认为自己提供了基于因果关系的完整解释；Potochnik（2017）也提供了一种解释说明，其中因果关系以因果模式的形式发挥着必要的作用。第三章通过一系列例子展示了一些不涉及因果关系的解释。如果本章的目标是证明并非所有解释都需要因果解释，那么本章可能有点过分了，因为一些明显的目标，如伍德沃德，已经同意这一点，并且有很多关于独特数学解释的有趣工作，强调了如何它们与他没有参与的因果解释形成对比并补充。但作为非因果解释示例的集合，本章为现有示例添加了新材料，特别是 Marc Lange（2016）给出的独特统计解释的示例。在第 4 章中，赖斯提出了他自己的反事实解释，并将其与其他类似的解释进行了对比。他提出了任何此类帐户都应满足的三个标准，这些标准在这些讨论中将很有用（93），即使人们不想采用赖斯自己的特定帐户。赖斯自己的描述在这里的细节似乎很压缩，如果人们只阅读这一章，很难看出这应该如何运作并成为真正的进步。后面的章节，特别是第 6、7 和 9 章，展示了该帐户在应用时如何运作，很有启发性。因此，例如，如果在研讨会上根据本书进行教学，将第 4 章与其中一章（尤其是第 6 章）配对是很有用的。第 5 章很简短，重点介绍如何将模型分解为子组件，然后单独处理这些子组件根本不适用于大多数型号。赖斯对为什么必须从整体上对待模型提出了一些非常明确的观点，巩固了他的观点，即理想化是这些模型中的扭曲，当通过分解尝试脱离上下文时，这些模型没有意义。普遍性，这里的一个艺术术语关于莱斯的其他工作（例如，参见 Rice 2018、2019；Batterman 和 Rice 2014），在第 6 章中进行了详细处理。本章列出了一些详细的案例研究，并说明了理想化中涉及的整体扭曲是如何传达或捕获的模型中的特定模态信息。第七章继续讨论莱斯在其他地方写过的主题：多尺度模型以及普遍性如何适应尺度和重正化的考虑。第八章继续考虑模型如何能够提供理解，即使它们不提供解释。赖斯的例子涉及科学家的解释不完整的情况，因此有些人可能会认为这些已经是解释，因为人们不需要要求解释完全完整才能算作解释。本章还将理解与现实主义和科学进步联系起来。理想化经常被视为现实主义的失败，据称，一个原本成功的模型被分解为多个元素，其中一些元素显然在字面上并不准确，而人们可能认为有必要成为该组件的现实主义者（天真的另一种方式）真理的对应处理潜入了科学哲学，假设模型的各个部分应该一对一地映射到世界的各个部分，并且如果存在不以这种简化方式映射的理想化，那么模型的现实主义就会失败） 。在本章的第一部分，他利用自己对事实理解的描述，提出了一种现实主义的替代方法，其中重点不是孤立的模型组件，而是模型为科学家产生的理解主体。这种理解体系再次需要整体论，可以作为关于由此理解的行为的现实主义的认知基础。最后，在第 9 章中，赖斯汇集了书中的所有主题，并提出了迄今为止最清晰的案例，说明理想化如何被用作“整体扭曲”，而不仅仅是部分它是科学的一部分，但对建模技术在提供解释和理解方面的成功做出了核心和积极的贡献。本章是将书中不同主题结合在一起的一个很好的结论。许多其他主题都是赖斯在其他地方写过的，这一章的结论有助于理解所有这些工作所适合的概要项目。如果有人用这篇文章来教学，这可能是一个很好的开始而不是结束的章节。总的来说，这本书很好地汇集了莱斯以前的工作，同时也用新的例子和解决的例子扩展了这项工作更详细地讲，并围绕整体主义和理想化作为整体模型扭曲的方向，以有凝聚力的方式连接不同的主题。这使得它成为一系列围绕解释、模型、理解和现实主义的当代讨论的一个很好的补充，也是研究生进入这些主题的一个很好的起点。