The Dynamics of Science: Computational Frontiers in History and Philosophy of Science

The Dynamics of Science

Computational Frontiers in History and Philosophy of Science

Edited by

Grant Ramsey and Andreas De Block

University of Pittsburgh Press

Published by the University of Pittsburgh Press, Pittsburgh, Pa., 15260

Copyright © 2022, University of Pittsburgh Press

All rights reserved

Manufactured in the United States of America

Printed on acid-free paper

10 9 8 7 6 5 4 3 2 1

Cataloging-in-Publication Data is available from the Library of Congress

ISBN 13: 978-0-8229-4737-0

ISBN 10: 0-8229-4737-4

Cover art from Freepik.com

Cover design by Melissa Dias-Mandoly

ISBN-13: 978-0-8229-8909-7 (electronic)

Contents

Acknowledgments

Introduction. Tools, Tests, and Data: An Introduction to the New History and Philosophy of Science

Andreas De Block and Grant Ramsey

Part I. Toward a New Logic of Scientific Discovery, Creativity, and Progress

1. Five Models of Science, Illustrating How Selection Shapes Methods

Paul E. Smaldino

2. Pooling with the Best

Justin Bruner and Bennett Holman

3. Promoting Diverse Collaborations

Mike D. Schneider, Hannah Rubin, and Cailin O’Connor

4. Using Phylomemies to Investigate the Dynamics of Science

David Chavalarias, Philippe Huneman, and Thibault Racovski

Part II. Frontiers in Tools, Methods, and Models

5. LDA Topic Modeling: Contexts for the History and Philosophy of Science

Colin Allen and Jaimie Murdock

6. The Potential of Supervised Machine Learning for the Study of Science

Krist Vaesen

7. Help with Data Management for the Novice and Experienced Alike

Steve Elliott, Kate MacCord, and Jane Maienschein

Part III. Case Studies

8. How Not to Fight about Theory: The Debate between Biometry and Mendelism in Nature, 1890–1915

Charles H. Pence

9. Topic Modeling in HPS: Investigating Engaged Philosophy of Science throughout the Twentieth Century

Christophe Malaterre, Jean-François Chartier, and Davide Pulizzotto

10. Bolzano, Kant, and the Traditional Theory of Concepts: A Computational Investigation

Annapaola Ginammi, Rob Koopman, Shenghui Wang, Jelke Bloem, and Arianna Betti

11. The Evolution of Evolutionary Medicine

Deryc T. Painter, Julia Damerow, and Manfred D. Laubichler

Notes

References

Contributors

Index

Acknowledgments

Our first debt is to the twenty-seven contributing authors, without whom this book would not have been possible. We also thank the reviewers who gave their time and expertise to provide valuable feedback. We give special thanks to the University of Pittsburgh Press staff, in particular to our editor, Abby Collier, who believed in this project and helped carry it to fruition. The seed for this book was a workshop held at the KU Leuven on the cultural evolution of science. We are grateful to the Research Foundation Flanders for their financial support for this workshop.

Introduction

Tools, Tests, and Data

An Introduction to the New History and Philosophy of Science

Andreas De Block and Grant Ramsey

In 1973, philosopher Ronald Giere published a now-famous review of a 1970 volume, edited by historian Roger Stuewer (1970), in which he reflected on history and philosophy of science and how integrated they can or should be. According to Giere, there is no intimate relationship between history of science and philosophy of science, only a marriage of convenience. In his view, normative questions constitute the field of philosophy of science, and such normative questions cannot be answered by the descriptive work of historiography.

One problem with Giere’s view is that the boundaries between the normative and the descriptive are often vague. Another problem is that how science has been done is relevant for how it should be done, much in the same way that moral psychology is relevant for normative ethics (Doris 2002). Last, Giere later wholly denounced earlier claims about the nature of philosophy of science and defended its naturalization in the sense that philosophers should be in the business of constructing a theoretical account of how science works (2011, 61). In other words, Giere later defended the view that he attacked in the 1970s. Much like history of science, Giere now believes that philosophy of science should be conceived of as a descriptive discipline, even though the two disciplines address somewhat different research questions.

Today, most philosophers of science seem to subscribe to this sort of naturalism (Schickore and Steinle 2006), and many are at least sympathetic to the view that history of science and philosophy of science can be so intertwined as to form an interdiscipline (Thorén 2015). Yet it is not always clear how the two disciplines should interact. Some fear that without this clarity, we are condemned to the dilemma between making unwarranted generalizations from historical cases and doing entirely ‘local’ histories with no bearing on an overall understanding of the scientific process (Chang 2011, 109). Indeed, philosophers of science all too often cherry-pick examples from the history of science to illustrate or support their philosophical views. This even holds for some of the philosophers with a solid background in the historiography of science, like Thomas Kuhn. Even today, philosophers and historians often engage in detailed analyses of particular discoveries, inventions, scientific debates, and developments and simply assume that these examples are representative of (the history of) science, without actually showing that the examples are paradigmatic (Scholl and Räz 2016).

Case studies do have heuristic value, and their use in pedagogy is generally undisputed, but they provide a weak inductive basis for general claims about science. Or, to put it more bluntly, anecdotal evidence is unreliable evidence. This problem is exacerbated both by the lack of historical interest in what are seen as grand narratives and by the theory-ladenness of case studies. It is quite likely that the choice and interpretation of case studies tend to be profoundly influenced by the researcher’s theoretical presuppositions, idiosyncratic preferences, values, biases, and philosophical training: Cases are often generated in a manner that does not adequately guard against biases in selection, emphasis, and interpretation (Steel, Gonnerman, and O’Rourke 2017). This problem is aggravated by the fact that the researcher’s choices are rarely made explicit, which makes an independent assessment of the conclusions or interpretations even more difficult.

In this book, we will address whether new computational tools can successfully address some of these problems. This introduction sketches the philosophical and scientific background of computational history and philosophy of science (HPS), argues for the great potential of these methods and tools, and offers an overview of the chapters that follow, each of which contributes to the realization of the promise that computational HPS holds, a promise that motivates this edited volume.

Naturalism in Philosophy of Science

Clearly, there is not as much progress in philosophy as in science. David Chalmers, for instance, claims that there has not been large collective convergence to the truth on the big questions of philosophy (Chalmers 2015, 5). Most of the progress in philosophy is negative, or so Chalmers argues. We now know better than before that some arguments of the great philosophers are deeply flawed and what knowledge, for example, is not. But we have not arrived at a widespread agreement on what knowledge—or any of the core philosophical subjects—is. Plenty of philosophers seem to think that there is little we can do about it. Philosophy, they believe, should be primarily about asking good questions and much less about finding the final answers to them.

Others hold that philosophy does make considerable progress but that this progress is obfuscated by how we conceptualize philosophy. In their view, once an important philosophical question has been answered decisively, we stop considering the question to be a philosophical one; it is expelled from the philosophical realm. As Daniel Dennett (1998) put it,

The trajectory of philosophy is to work on very fundamental questions that haven’t yet been turned into scientific questions. Once you get really clear about what the questions are, and what would count as an answer, that’s science. Philosophy no longer has a role to play. That’s why it looks like there’s just no progress. The progress leaves the field. If you want to ask if there has been progress in philosophy, I’d say, look around you. We have departments of biology and physics. That’s where the progress is. We should be very proud that our discipline has spawned all these others.

In this view, Newton’s natural philosophy became modern physics, and Wundt’s experimental philosophy can be seen as one of the first successful attempts at a genuinely scientific psychology.

Still other prominent philosophers, such as Timothy Williamson (2006), think that current academic philosophy can and must do better. But what kind of reform is necessary for such an improvement? Since Descartes, philosophers have tried to borrow scientific methods to arrive at the same kind of certainties and cumulative knowledge that the sciences deliver. In recent years, philosophy has witnessed a renewed interest in the use of scientific tools and methods to tackle philosophical issues. In many instances, this tendency is illustrated by evolutionary and formal epistemology (Callebaut and Pinxten 2012; Hendricks 2006, 2010). Ever since the early 1970s, evolutionary epistemology had been analyzing belief systems and belief change with the help of hypotheses, theories, and models that were developed to understand population dynamics in biology. And at the beginning of this century, formal epistemology really took off. It made clear that popular approaches in, for example, computer science and statistics were instrumental in getting a better grip on some of the more vexing philosophical issues.

Whereas naturalism is now mainstream in philosophy of science in that it often focuses on descriptive questions and takes the details of scientific research into account, philosophers of science have not seemed to be overly enthusiastic about embracing naturalistic methods. According to Edouard Machery, philosophers of science have surprisingly been reluctant to include these methods in their toolbox, but doing so is necessary for philosophy of science to be a genuine part of a naturalized epistemology (2016, 487).

Yet this is beginning to change. Experimental philosophy of science is now a flourishing approach (Griffiths and Stotz 2008; Wilkenfeld and Samuels 2019), and the digital or computational tools of formal epistemology are now regularly applied in philosophy of science (Leitgeb 2011), which in itself can be seen as a form of applied epistemology.

Despite the reality of this tendency in philosophy toward naturalization, it would be an oversimplification to see the rise of computational HPS as nothing but the result of this tendency. As the next section makes clear, the maturation of the science of science also played a role in it, as did developments within history of science.

HPS as a Science of Science

History and philosophy of science has always had close ties and tensions with other academic disciplines that reflect on science. This holds in particular for sociology of science and its longtime ally, the interdisciplinary science and technology studies (STS). Much like traditional HPS, STS relies mostly on qualitative methods and verbal reasoning. However, within the STS community of the 1970s, a number of scholars argued that in order to weigh optimally on science and technology policy, more quantitative approaches were necessary (Martin, Nightingale, and Yegros-Yegros 2012). To accomplish that, they regularly joined forces with scholars from scientometrics (Leydesdorff 1989), another (sub)discipline that emerged in the 1970s. It took a while, but eventually this became the precursor of a new and burgeoning field, the science of science.¹ In a recent review published in Science, Santo Fortunato and colleagues (2018, 1007) define this field as follows: The science of science . . . offers a quantitative understanding of the interactions among scientific agents across diverse geographic and temporal scales: It provides insights into the conditions underlying creativity and the genesis of scientific discovery, with the ultimate goal of developing tools and policies that have the potential to accelerate science. The science of science does not limit itself to measuring innovation and impact; it also focuses on the epistemic and ethical costs of science policy and the (changes in the) organization of science. Evolutionary biologists Gross and Bergstrom (2019), for example, used economic contest theories to assess the relative efficiency of grant proposal competitions and concluded that these competitions probably hinder science more than advance it, thus tying into recent work in philosophy of science about the epistemic merits of lotteries (Avin 2019). Similarly, quantitative and empirical work on peer review is highly relevant for philosophical reflections on its value (Heesen and Bright 2021). Consequently, papers published under the science of science banner are sometimes indistinguishable from quantitative and empirically informed articles in philosophy of science.²

Not all scientifically conducted HPS is considered computational. Yet it is not always clear how to draw lines between work that is computational and work that is not. Within history, everyone seems to agree that a computational turn has taken place, but there is disagreement on what exactly constitutes this turn. We believe the two most common names for this turn are indicative of what it means to do computational HPS. First, the term computational is used because the methods that are at the heart of this turn or revolution often involve mathematical abstractions and mathematical and other formal models (Roth 2019).³ Second, the turn is sometimes also called a digitized turn, a big data revolution, or a digital revolution (Gibson, Laubichler, and Maienschein 2019) because this revolution to a large extent hinges on a systematic engagement with digitized archives and big data.

These two strands or orientations are in line with how Grim and Singer (2020) sketch computational philosophy: Techniques employed in computational philosophy may draw from standard computer programming and software engineering, including aspects of artificial intelligence, neural networks, systems science, complex adaptive systems, and a variety of computer modeling methods. As a growing set of methodologies, it includes the prospect of computational textual analysis, big data analysis, and other techniques as well. The two orientations can go hand in hand, as some of the chapters in this volume show, but they need not. For instance, the disciplinary exchange between history and computational humanities shows that there are valid research questions for C[omputational] H[umanities] regardless of the size of the data, in particular in the domain of knowledge representation and reasoning (Piotrowski and Fafinski 2020, 175).

Of course, the new tools and methods should not just be naively accepted and applied. A good fit between research question and method is necessary, and computational tools are not well suited to address all the central questions of HPS. Hence, traditional HPS should not be completely replaced by computational HPS. There is also more to HPS than answering questions, and non-computational methods certainly have great heuristic and pedagogic value. Another reason to temper a naive enthusiasm for the use of computational methods in HPS is the steep learning curve. Many historians and philosophers simply lack sufficient training in mathematics and computer science for conducting computational HPS on their own, though they can outsource the more technical work to others who have this training (Gibson, Laubichler, and Maienschein 2019).

We think the computational developments in historiography, the naturalistic turn in philosophy, and the rise of the science of science will inevitably lead to more interdisciplinary approaches and collaborations. A full grasp of the nature and dynamics of science requires a huge toolbox and an academic community that harbors the abilities to use all the tools in that toolbox. Philosophers and historians of science will surely play an important role in that community, but so will data and computer scientists, cognitive scientists, evolutionary theorists, and statisticians. Philosophers of science, even those who are not well versed in computational methods, can bring to this interdisciplinary community their well-trained capacities for conceptual rigor, methodological reflection, and synthesis. These capacities complement the computational skills because philosophical considerations and decisions are necessary for a reliable and valid collection, processing, and interpretation of the data, for instance by identifying the specific, extant conjectures about the processes of scientific change to be tested (Laudan et al. 1986, 143).

Let’s now consider the promises that computational tools and methods hold for history and, especially, for philosophy of science. We will consider the potential of digitized material as well as the prospects of models and simulations.

How Not to Be Selective

In 2006, an estimated 1.35 million scientific articles were published—almost four thousand per day (Bjork, Roos, and Lauri 2009)—and this number is doubling about every nine years (Van Noorden 2014). Thus, while there was a time when we could keep up with broad swaths of the scientific literature, it is now difficult to stay abreast of advances within even the smallest subdisciplines. Traditional approaches to the study of science—such as HPS—involve closely reading a relatively small set of books, journal articles, and other documents. The historian may be studying a specific moment in the history of science, trying to uncover what happened and why. The philosopher may be trying to unravel the meaning of particular concepts or to reconstruct forms of scientific inference or explanation. In doing so, each researcher can examine but a single drop in the ocean of literature—their studies may involve depth, but they lack breadth.

As we already mentioned, some studies require this depth and don’t greatly suffer from a lack of breadth. Very specific claims about local phenomena are often best addressed through only a handful of case studies. In that sense, the continued use of case studies in philosophy of science is not necessarily a problem. What is worrisome, though, is the still rather widespread reliance on case studies as evidence for general claims about science (Mizrahi 2020). And as Laudan and colleagues already noted in the 1980s, many of the avowed case studies are not ‘tests’ of the theory in question at all; rather, they are applications of the theory to a particular case (Laudan et al. 1986, 158). In fact, there are many questions about the nature and history of science that would clearly benefit from—or even require—casting a larger net. If we want to know whether most scientific change is gradual or revolutionary, or what the key sources of scientific novelty are, then it cannot be close reading—the careful reading of text by humans—or a single case study that serves as the only data. Instead, it would be ideal to have a digital database full of scientific literature and to equip computers with algorithms capable of helping us answer questions like these.

Fortunately, the past few decades have seen a massive effort to digitize the academic literature. This literature is thus accessible in ways that it has never been and can now be subject to distant reading—machine reading involving automated textual analysis (Moretti 2013). Having a rich source of data—an ocean of scientific literature—at our disposal can get us partway to answering fundamental questions about the nature and history of science. Importantly, digital tools are not useful just because they can engage with larger corpora. They have additional advantages. For instance, they can detect fine-grained linguistic patterns that would be lost to a (single) human reader and can increase the breadth of research questions (Pence and Ramsey 2018).

Despite the surfeit of digital data from the scientific literature, this resource is a largely unexplored frontier. Some researchers have begun to employ digital approaches to study the scientific literature, but the techniques are still highly experimental. In order to understand the motivation for exploring the computational HPS frontier, let’s consider three published studies of the scientific literature. These highlight the increasing interest in textual analysis, as well as the relative absence of appropriate analysis tools.

Example 1. In 2007, Antonovics et al. published a paper titled Evolution by Any Other Name: Antibiotic Resistance and Avoidance of the E-word (Antonovics et al. 2007). Antonovics and his team manually tallied terms in thirty articles from medical and biological journals to test the hypothesis that the term evolution is used more rarely in medicine than in biology in the context of antibiotic resistance. Their results supported this claim, but they had to manually read through the papers, limiting their sample size to only thirty. Such a study suffers from a lack of automation. If they had been able to use even the most basic automated techniques—simple searches for evolution across a larger corpus—they could have considerably broadened and strengthened their study. Is the difference they detected historically recent, or is it long-standing? Is it true of all Anglophone journals or only those from the United States? Do medical journal articles that use the e-word exhibit different patterns of citation? Answering these questions could lead to a much richer understanding of how and why the term evolution appears to be avoided in medical journals. In chapter 11 of this volume, Painter, Damerow, and Laubichler use automated techniques to touch upon a few of these issues and show how biologists seem to be more attracted to evolutionary medicine than medical scholars.

Example 2. In 2013, Overton performed an analysis of how the term explain is used in the scientific literature. His paper (Overton 2013) used a set of 781 articles from the journal Science. Although he obtained interesting results, his study was limited in scope (a single journal over a limited span of time) and used only the most basic of techniques (counting the frequency of n-grams, i.e., strings of letters). Science is a highly selective prestigious general science journal, and this poses questions about the degree to which we can generalize from these results. Is the concept of explanation deployed differently in a journal like this than in narrower, lower-impact journals? How and why has usage changed over time? Is the concept used differently when new phenomena are explained than when old ones are? These are the sorts of questions one could answer only with more comprehensive data analysis tools at hand.

Example 3. In 2014, Dietrich, Ankeny, and Chen (2014) published an article titled Publication Trends in Model Organism Research. They examined the rates of citations for articles referencing specific model organisms. They were interested in the pattern of citations, especially in the wake of the National Institute of Health specifying in 1990 a list of organisms as official model organisms. Their study used the Web of Science database, but this involved a significant limitation: it was not until 1992 that the Web of Science began to include not only titles but also abstracts. Thus, Dietrich, Ankeny, and Chen (2014) could use titles only, which underreport model organism usage, and the data were likely considerably noisier than they would have been using abstracts or whole texts. Had they used a tool that provides the full text of journal articles, they may have had a much richer dataset and may have been able to go beyond word tallies to examine word associations. For example, what words modify or occur near the names of model organisms?

Models and Simulations

Which models—from cultural evolution, game theory, or other sources—can best be used to understand scientific change is still an open question. In 1988, David Hull published Science as a Process, in which he tried to give an evolutionary account of the dynamics of science. To substantiate his claims, he relied on several sources, such as interviews with pupils of successful scientists, information about patterns of journal article publication, and data on the age and number of scientists who accepted Darwin’s thesis on the mutability of species between 1859 and 1869. Yet the data are quite limited and were extracted by hand. Moreover, Hull (1988) does not rely at all on mathematical models of social evolution, and he barely mentions the existence of such methods and how they can be applied to cultural phenomena. This is especially strange, since foundational work in this area had been published just a few years before Hull’s book—two important examples being Boyd and Richerson’s (1985) Culture and the Evolutionary Process and Cavalli-Sforza and Feldman’s (1981) Cultural Transmission and Evolution: A Quantitative Approach.

Suppose Hull is right and certain aspects of science are shaped by the same (or very similar) evolutionary forces that shape organisms (as Popper 1968 and Campbell 1974 also contend). It seems obvious then that the study of scientific change could benefit from using the mathematical tools and software programs that evolutionary biologists use to model and understand such forces. First, these tools can help to evaluate or strengthen the arguments made about the dynamics and nature of science. After all, it is well known that formalizing arguments helps us to detect errors and to add rigor, albeit sometimes at the expense of comprehensibility. Second and relatedly, formal tools can also help us better model and simulate the dynamics of science. Agent-based models, for example, can be used to explore the epistemic effects of different sorts of communication between scientists. Quite a few simulation studies have now shown that cognitive division of labor and cognitive diversity bring clear epistemic advantages (Thoma 2015). Some of the findings of such agent-based modeling, however, are much less intuitive. For example, Zollman (2010) has shown that the best functioning communication networks are not the fully connected networks, a finding further explored by Holman and Bruner in chapter 2 of this volume.

Of course, many useful simulations and models are not intrinsically linked to evolutionary explanations of the change of scientific knowledge. Paul Thagard, for instance, was a staunch opponent of evolutionary epistemology (Thagard 1980) but contributed enormously to the computational study of scientific networks. His seminal article on this topic, Explanatory Coherence (Thagard 1989), has now been cited more than twelve hundred times. That said, many philosophers keep on highlighting and exploring the similarities between evolutionary dynamics and the dynamics of science. Epistemic landscapes are regularly likened to, and even inspired by, fitness landscapes (Alexander, Himmelreich, and Thompson 2015). Likewise, evolutionary theory–inspired terms such as the cultural red king effect have been coined to refer to effects that simulations of scientific communication bring to the fore (O’Connor 2017; see also chapter 3 in this volume where Schneider, Rubin, and O’Connor dig deeper into the cultural red king and related effects).

Structure

This book is structured in three parts. Each part treats different aspects of the history and philosophy of science with model- or algorithm-based approaches.

Part I, Toward a New Logic of Scientific Discovery, Creativity, and Progress, comprises the first four chapters of the book. It deals with questions that have been central to philosophy of science and explores how the new techniques can elucidate them. Although many insights in philosophy of science have been gained through traditional philosophical methodologies, the new computational methods explored here can shed fresh light on old questions and may be instrumental in deciding debates at the center of the field.

In chapter 1, Smaldino shows that the classic hypothesis testing model of science is flawed and discusses a series of Darwinian models free from the flaws of the classic model. These models treat science as a population-level process. Each of these models highlights different incentives and suggests that the current dynamics often results in science that lacks robustness and reproducibility. On the other hand, the models also suggest how selection pressures on individual scientists and their work, instantiated by publishing and funding practices, can be altered in such a way that they structurally advance methodological rigor.

In chapter 2, Bruner and Holman investigate how norms and incentives shape scientific group inquiry. Social structure matters, and as a result, many philosophers of science have investigated how norms and incentives can lead to more or less successful inquiries. More specifically, Bruner and Holman use agent-based modeling with realistic assumptions to study whether less competent scientists will come to interact with the most competent researchers. They find that scientists will eventually pool with the best, but that this will not always result in a communication structure that is optimal for the reliability of science. They conclude that the social structures that emerge from the simulations are suboptimal. This result has implications for understanding these structures but also provides some guidance for how to improve them.

In chapter 3, Schneider, Rubin, and O’Connor use agent-based modeling to explore how underrepresentation of minority groups within science affects diverse collaborations. Arguably, such collaborations are epistemically beneficial and morally desirable. Hence, many initiatives have been taken to promote diverse collaborations. Yet Schneider, Rubin, and O’Connor’s modeling indicates that many of these initiatives will not improve the diversity of collaborative groups, in part because they have unintended negative consequences.

In chapter 4, Chavalarias, Huneman, and Racovski describe how the evolution of science involves both historical patterns and particular processes. They argue that, traditionally, philosophers have based their account of patterns almost exclusively on a study of the processes. In their chapter, they show how this unidirectional approach can be replaced by two-way reasoning that explicitly draws on phylogenetic methods. The phylomemetic approach that Chavalarias et al. present in their chapter allows for better reconstructions of the dynamics of science than the methods traditionally used in philosophy of science.

Part II, Frontiers in Tools, Methods, and Models, contains three shorter, method-focused chapters. These chapters are intended to provide the reader with insight into the cutting edge of computational HPS. This will inform the readers of the state of the field but will also serve as a kind of instruction manual for future work.

In chapter 5, Allen and Murdock use latent Dirichlet allocation (LDA) topic modeling to understand and quantify conceptual similarity and conceptual change, the sensitivity of meanings to context, and pathways of intellectual influence. They show how this method is relevant for HPS researchers and that its relevance goes beyond its use as a mere search and retrieval tool. According to them, many of the pitfalls of LDA can be avoided by conceiving of topics in a topic model as representing contexts, not just as buckets of words.

In chapter 6, Vaesen scrutinizes the promises and pitfalls of different machine learning techniques to track and analyze disciplinary change in fields like economics, sociology, and cognitive science. He focuses on how many of the shortcomings of unsupervised techniques can be overcome by human supervision. He argues that, although the potential of supervised machine learning techniques is hitherto relatively unexplored within HPS, this family of techniques is able to deal with relevant classification features such as semantics, grammar, and style, features that mostly escape the unsupervised tools. Compared to unsupervised techniques, supervised techniques suffer less from information loss and can incorporate more classification features.

In chapter 7, Elliott, MacCord, and Maienschein give useful advise on how to manage data. HPS researchers are typically not very well trained in managing and storing data. With the advent of digital humanities, however, more and more researchers must overcome the data hurdle. They present different principles to manage data and illustrate the use of those principles with two digital projects from the history of science.

Part III, Case Studies, focuses on the application of models and automated textual analyses to gain new insights into the history of science. While parts I and II focus on theoretical and epistemological aspects of computational tools, part III considers what fruits can come from applying these tools and methods to specific historical moments and episodes in science. These case studies can further clarify the strengths and limitations of the new techniques for studying the dynamics of science.

In chapter 8, Pence uses digital tools to sketch a network of discourse for