The Cognitive Science of Science: Explanation, Discovery, and Conceptual Change

© 2012 Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher.

For information about special quantity discounts, please email [email protected].

Library of Congress Cataloging-in-Publication Data

Thagard, Paul.

The cognitive science of science : explanation, discovery, and conceptual change / Paul Thagard ; in collaboration with Scott Findlay . . . [et al.].

   p.  cm.

Includes bibliographical references and index.

ISBN 978-0-262-01728-2 (hardcover : alk. paper)

ISBN 978-0-262-30097-1 (retail e-book)

1. Science—Philosophy. 2. Cognitive science. I. Findlay, Scott. II. Title.

Q175.T478 2012

501—dc23

2011039760

10 9 8 7 6 5 4 3 2 1

d_r1

To the pioneers of the cognitive science of science, especially Herbert Simon

Contents

Preface

Acknowledgments

I    Introduction

1    What Is the Cognitive Science of Science?

II    Explanation and Justification

2    Why Explanation Matters

3    Models of Scientific Explanation

with Abninder Litt

4    How Brains Make Mental Models

5    Changing Minds about Climate Change: Belief Revision, Coherence, and Emotion

with Scott Findlay

6    Coherence, Truth, and the Development of Scientific Knowledge

III    Discovery and Creativity

7    Why Discovery Matters

8    The Aha! Experience: Creativity through Emergent Binding in Neural Networks

with Terrence C. Stewart

9    Creative Combination of Representations: Scientific Discovery and Technological Invention

10    Creativity in Computer Science

with Daniel Saunders

11    Patterns of Medical Discovery

IV    Conceptual Change

12    Why Conceptual Change Matters

13    Conceptual Change in the History of Science: Life, Mind, and Disease

14    Getting to Darwin: Obstacles to Accepting Evolution by Natural Selection

with Scott Findlay

15    Acupuncture, Incommensurability, and Conceptual Change

with Jing Zhu

16    Conceptual Change in Medicine: Explanations of Mental Illness from Demons to Epigenetics

with Scott Findlay

V    New Directions

17    Values in Science: Cognitive-Affective Maps

18    Scientific Concepts as Semantic Pointers

References

Index

Preface

This book is a collection of my recent essays on the cognitive science of science that illustrate ways of combining philosophical, historical, psychological, computational, and neuroscientific approaches to explaining scientific development. Most of the chapters have been or will be published elsewhere, but the introductions are brand new (chapters 1, 2, 7, 12), as are the last two chapters, which take the cognitive science of science in new directions related to values and concepts. The reprinted chapters reproduce the relevant articles largely intact, but I have done some light editing to coordinate references and remove redundancies. Origins of the articles and coauthors are indicated in the acknowledgments.

Early in my career, I wandered into the cognitive science of science through a series of educational accidents, and have enthusiastically pursued research that is variously philosophical, historical, psychological, computational, and neurobiological. In high school, I did very well in physics and chemistry, but only because I was adept at solving math problems, not because I found science very interesting. As an undergraduate at the University of Saskatchewan, I avoided serious science courses, although I did get a good sampling of mathematics and logic. My interest in science was sparked during my second undergraduate degree at Cambridge University, where the philosophy course I took required a paper in philosophy of science. Through lectures by Ian Hacking and Gerd Buchdahl, along with books by Russell Hanson and Thomas Kuhn, I started to appreciate the value of understanding the nature of knowledge by attention to the history of science. I was struck by how much more rich and interesting the scientific examples of knowledge were compared to the contrived thought experiments favored by epistemologists working in the tradition of analytic philosophy. Accordingly, my Ph.D. work at the University of Toronto focused on scientific reasoning enriched by historical case studies.

My move into cognitive science was also serendipitous. In 1978, in the second term of my first teaching job at the University of Michigan–Dearborn, I decided to sit in on a graduate epistemology course taught by Alvin Goldman at the main Michigan campus in Ann Arbor. It turned out that this course was coordinated with one on human inference taught by the social psychologist Richard Nisbett. The combination of these two courses was amazing: Goldman was pioneering an approach to epistemology that took experimental research on psychology seriously, and Nisbett was presenting a draft of his path-breaking book with Lee Ross, Human Inference. I started reading avidly in cognitive psychology, which quickly led me to the field of artificial intelligence. I was attracted by the theoretical ideas of visionaries such as Marvin Minsky, and also by the prospects of a new methodology—computer modeling—for understanding the structure and growth of scientific knowledge.

Accordingly, I did an MS in computer science at Michigan and started building my own computational models of various aspects of scientific thinking. My early models of analogical thinking were somewhat crude, but became much more powerful when my collaborator Keith Holyoak came up with the idea of modeling analogy using connectionist ideas about parallel constraint satisfaction. I quickly realized that theory choice based on the explanatory power of competing theories could also be simulated using neural networks.

My interest in neuroscience also came about indirectly. After I moved to Waterloo in 1992, one of my graduate students, Allison Barnes, was investigating empathy as a kind of analogy, which led me to general concern with emotions. The work of Antonio Damasio revealed how crucial neuroscience was to understanding emotions, and since I was already building artificial neural network models, it was natural to try to undertake more realistic neural models of emotion and decision making. Happily, this line of work has turned back around to scientific applications, described in some of the chapters below.

I remain convinced that understanding the growth of knowledge requires the kind of interdisciplinary approach found in cognitive science. I hope this collection will appeal to anyone interested in the structure and growth of scientific knowledge, including scientists, philosophers, historians, psychologists, sociologists, and educators.

Acknowledgments

While I was writing and revising this work, my research was supported by the Natural Sciences and Engineering Research Council of Canada. I am grateful to the coauthors of essays included in this collection: Scott Findlay (who made major contributions to three chapters), Abninder Litt, Daniel Saunders, Terry Stewart, and Jing Zhu. Please note that Daniel is first author of our joint article. Chris Eliasmith’s exciting ideas about theoretical neuroscience contributed to several chapters. For comments or suggestions for particular chapters, I am indebted to him, William Bechtel, Chris Grisdale, Lloyd Elliott, Robert Hadley, Phil Johnson-Laird, Kostas Kampourakis, Eric Lormand, Elijah Millgram, Daniel Moerman, Nancy Nersessian, Eric Olsson, Robert Proctor, Peter Railton, David Rudge, Daniel Saunders, Cameron Shelley, and Terry Stewart. CBC Radio 2 provided the accompaniment.

I am grateful to my coauthors and to the respective publishers for permission to reprint the following essays:

Thagard, P., & Litt, A. (2008). Models of scientific explanation. In R. Sun (Ed.), The Cambridge handbook of computational psychology (pp. 549–564). Cambridge: Cambridge University Press. © Cambridge University Press.

Thagard, P. (2010). How brains make mental models. In L. Magnani, W. Carnielli & C. Pizzi (Eds.), Model-based reasoning in science and technology: Abduction, logic, and computational discovery (pp. 447–461). Berlin: Springer. © Springer.

Thagard, P., & Findlay, S. D. (2011). Changing minds about climate change: Belief revision, coherence, and emotion. In E. J. Olsson & S. Enqvist (Eds.), Belief revision meets philosophy of science (pp. 329–345). Berlin: Springer. © Springer.

Thagard, P. (2007). Coherence, truth, and the development of scientific knowledge. Philosophy of Science, 74, 28–47. © University of Chicago Press.

Thagard, P., & Stewart, T. C. (2011). The Aha! experience: Creativity through emergent binding in neural networks. Cognitive Science, 35, 1–33. © Cognitive Science Society.

Thagard, P. (forthcoming). Creative combination of representations: Scientific discovery and technological invention. In R. Proctor & E. J. Capaldi (Eds.), Psychology of science. Oxford: Oxford University Press. © Oxford University Press.

Saunders, D., & Thagard, P. (2005). Creativity in computer science. In J. C. Kaufman & J. Baer (Eds.), Creativity across domains: Faces of the muse (pp. 153–167). Mahwah, NJ: Lawrence Erlbaum. © Taylor and Francis.

Thagard, P. (2011). Patterns of medical discovery. In F. Gifford (Ed.), Handbook of philosophy of medicine (pp. 187–202). Amsterdam: Elsevier. © Elsevier.

Thagard, P. (2008). Conceptual change in the history of science: Life, mind, and disease. In S. Vosniadou (Ed.), International handbook of research on conceptual change (pp. 374–387). London: Routledge. © Taylor and Francis.

Thagard, P., & Findlay, S. (2010). Getting to Darwin: Obstacles to accepting evolution by natural selection. Science & Education, 19, 625–636. © Springer.

Thagard, P., & Zhu, J. (2003). Acupuncture, incommensurability, and conceptual change. In G. M. Sinatra & P. R. Pintrich (Eds.), Intentional conceptual change (pp. 79–102). Mahwah, NJ: Lawrence Erlbaum. © Taylor and Francis.

Thagard, P., & Findlay, S. (forthcoming). Conceptual change in medicine: Explanations of mental illness from demons to epigenetics. In W. J. Gonzalez (Ed.), Conceptual revolutions: From cognitive science to medicine. A Coruña, Spain: Netbiblo. © Netbiblo.

Finally, I am grateful to Judith Feldmann for skillful editing and to Eric Hochstein for help with the index.

I

Introduction

1

What Is the Cognitive Science of Science?

Explaining Science

Science is one of the greatest achievements of human civilization, contributing both to the acquisition of knowledge and to people’s well-being through technological advances in areas from medicine to electronics. Without science, we would lack understanding of planetary motion, chemical reactions, animal evolution, infectious disease, mental illness, social change, and countless other phenomena of great theoretical and practical importance. We would also lack many valuable applications of scientific knowledge, including antibiotics, airplanes, and computers. Hence it is appropriate that many disciplines such as philosophy, history, and sociology have attempted to make sense of how science works.

This book endeavors to understand scientific development from the perspective of cognitive science, the interdisciplinary investigation of mind and intelligence. Cognitive science encompasses at least six fields: psychology, neuroscience, linguistics, anthropology, philosophy, and artificial intelligence (for overviews, see Bermudez, 2010; Gardner, 1985; Thagard, 2005a). The main intellectual origins of cognitive science are in the 1950s, when thinkers such as Noam Chomsky, George Miller, Marvin Minsky, Allan Newell, and Herbert Simon began to develop new ideas about how human minds and computer programs might be capable of intelligent functions such as problem solving, language, and learning. The organizational origins of cognitive science are in the 1970s, with the establishment of the journal Cognitive Science and the Cognitive Science Society, and the first published uses of the term cognitive science (e.g., Bobrow & Collins, 1975).

Cognitive science has thrived because the problem of understanding how the mind works is far too complex to be approached using ideas and methods from only one discipline. Many researchers whose primary backgrounds are in psychology, philosophy, neuroscience, linguistics, anthropology, and computer science have realized the advantages of tracking work in some of the other fields of cognitive science. Many successful projects have fruitfully combined methodologies from multiple fields, for example, research on inference that is both philosophical and computational, research on language that is both linguistic and neuroscientific, and research on culture that is both anthropological and psychological.

Naturally, cognitive science has also been used to investigate the mental processes required for the practice of science. The prehistory of the cognitive science of science goes back to philosophical investigation of scientific inference by Francis Bacon, David Hume, William Whewell, John Stuart Mill, and Charles Peirce. Modern cognitive science of science began only in the 1980s when various psychologists, philosophers, and computer scientists realized the advantages of taking a multidisciplinary approach to understanding scientific thinking. Pioneers include: Lindley Darden, Ronald Giere, and Nancy Nersessian in philosophy; Bruce Buchanan, Pat Langley, and Herbert Simon in computer modeling; and William Brewer, Susan Carey, Kevin Dunbar, David Klahr, and Ryan Tweney in experimental psychology. Extensive references are given in the next section. The earliest occurrence of the phrase cognitive science of science that I have been able to find is in Giere (1987), although the idea of applying cognitive psychology and computer modeling to scientific thinking goes back at least to Simon (1966).

This chapter provides a brief overview of what the component fields of cognitive science bring to the study of science, along with a sketch of the merits of combining methods. It also considers alternative approaches to science studies that are often antagonistic to the cognitive science of science, including formal philosophy of science and postmodernist history and sociology of science. I will argue that philosophy, history, and sociology of science can all benefit from ideas drawn from the cognitive sciences. Finally, I give an overview of the rest of the book by sketching how the cognitive science of science can investigate some of the most important aspects of the development of science, especially explanation, discovery, and conceptual change.

Approaches to the Cognitive Science of Science

It would take an encyclopedia to review all the different approaches to science studies that have been pursued. Much more narrowly and concisely, this section reviews what researchers from various fields have sought to contribute to the cognitive science of science.

My own original field is the philosophy of science, and I described in the preface how concern with the structure and growth of scientific knowledge led me to adopt ideas and methods from psychology and artificial intelligence, generating books and articles that looked at different aspects of scientific thinking (e.g., Thagard, 1988, 1992, 1999, 2000). Independently, other philosophers have looked to cognition to enhance understanding of science, including Lindley Darden (1983, 1991, 2006), David Gooding (1990), Ronald Giere (1988, 1999, 2010), and Nancy Nersessian (1984, 1992, 2008). Andersen, Barker, and Cheng (2006), Magnani (2001, 2009), and Shelley (2003) also combine philosophy of science, history of science, and cognitive psychology. Collections of work on philosophical approaches to the cognitive science of science include Giere (1992) and Carruthers, Stich, and Siegal (2002).

Philosophy of science is not just a beneficiary of cognitive science but also a major contributor to it. Since the 1600s work of Francis Bacon (1960), philosophers have investigated the nature of scientific reasoning and contributed valuable insights on such topics as explanation (Whewell 1967), causal reasoning (Mill 1970), and analogy (Hesse 1966). Philosophy of science was sidetracked during the logical positivist era by (1) a focus on formal logic as the canonical way of representing scientific information and (2) a narrow empiricism incapable of comprehending the theoretical successes of science. Logical positivism was as inimical to understanding scientific knowledge as behaviorism was to understanding thinking in general.

In response to logical positivism, Russell Hanson (1958), Thomas Kuhn (1962), and others spurred interest among philosophers in the history of science, but there was a dearth of tools richer than formal logic for examining science, although Hanson and Kuhn occasionally drew on insights from Gestalt psychology. In the 1980s, when philosophers looked to cognitive science for help in understanding historical developments, we brought to the cognitive science of science familiarity with many aspects of high-level scientific thinking. The method that philosophy of science can most valuably contribute to the cognitive science of science consists in careful analysis of historical case studies.

Most psychologists concerned with scientific thinking adopt a very different method—behavioral experiments. Such experimentation is a crucial part of cognitive science, providing data about many different kinds of thinking that theories aim to explain. Professional scientists are rarely available for psychological experiments, but participants can be recruited from among the modern-day lab rats of cognitive psychologists—university undergraduates. Much of the valuable work on scientific thinking has been motivated by an attempt to understand how children can develop an understanding of science, a worthy enterprise that is part of both developmental and educational psychology.

Experimental and theoretical work on the development of scientific knowledge has been conducted by many psychologists (e.g., Carey, 1985, 2009; Dunbar, 1997, 2001; Dunbar & Fugelsang, 2005; Gentner et al., 1997; Klahr, 2000; Schunn & Anderson, 1999; Tweney, Doherty & Mynatt, 1981; Vosniadiou & Brewer, 1992). Like all cognitive scientists, psychologists can contribute to the development of theories about scientific thinking, but their main methodological contribution consists in behavioral experiments, although some psychologists such as Dedre Gentner and Ryan Tweney also undertake historical studies. Useful collections of work on the psychology of science include Crowley, Schunn, and Okada (2001), Gholson et al. (1989), Gorman et al. (2005), and Proctor and Capaldi (forthcoming). Other works in the psychology of science tied less closely to experimental cognitive psychology include Feist (2006), Simonton (1988), and Sulloway (1996). The introductory chapters below for parts II, III, and IV provide further references to work in the psychology of science on the more specific topics of explanation, discovery, and conceptual change.

In addition to philosophical/historical studies and behavioral experiments, the cognitive science of science has made extensive use of computational models, which have been theoretically and methodologically important since the 1950s. The theoretical usefulness comes from the fruitfulness of the hypothesis that thought is a kind of computation: thinking consists in applying processes to representations, just as computing consists in applying algorithms to data structures (see Thagard 2005a for a review). This hypothesis was far more powerful than previous attempts to understand the mind in terms of familiar mechanisms such as clockwork, vibrating strings, hydraulic systems, or telephone switchboards.

Moreover, computer modeling provides theorizing about the mind with a novel methodology—writing and running computer programs. Beginning with the seminal work on problem solving by Newell, Shaw, and Simon (1958), computer modeling has provided an invaluable tool for developing and testing ideas about mental processes (Sun, 2008b). Computational models of scientific thinking have been developed both by researchers in the branch of computer science called artificial intelligence, and by philosophers and psychologists who have adopted computer modeling as part of their methodological toolkit. There are many notable examples of computer simulations of different aspects of scientific thinking (e.g., Bridewell et al., 2008; Bridewell & Langley, 2010; Kulkarni & Simon, 1988, 1990; Langley et al., 1987; Lindsay et al., 1980; Shrager & Langley, 1990; Thagard, 1992; Valdes-Perez, 1995). The next section gives a more detailed discussion of how computational modeling contributes to the cognitive science of science.

Experimental neuroscience has so far made little contribution to understanding scientific thinking, even though it is becoming increasingly important to cognitive psychology and other areas such as social, developmental, and clinical psychology. The role of neuroscience in cognitive science has increased dramatically over the past two decades because of new technologies for observing neural activity using brain scanning tools such as functional magnetic resonance imaging (fMRI). The complementary theoretical side of neuroscience is the development of computational models that take seriously aspects of neural processing such as spiking neurons and interconnected brain areas that were neglected in the connectionist models of the 1980s (see, e.g., Dayan & Abbott, 2001; Eliasmith & Anderson, 2003). There has not yet emerged a distinct enterprise one would call the neuroscience of science, although some of the kinds of thinking most relevant to scientific thought such as analogy, causal reasoning, and insight are beginning to receive experimental and theoretical investigation. What neuroscience can contribute to the understanding of science is knowledge about the neural processes that enable scientists to generate and evaluate ideas. Some of my own most recent work in chapters 3, 4, 8, and 19 employs ideas from theoretical neuroscience.

To complete this review of how the different fields of cognitive science contribute to the understanding of science, I need to include linguistics and anthropology. Unfortunately, I am not aware of much relevant research, although I can at least point to the work of Kertesz (2004) on the cognitive semantics of science, and to the work of Atran and Medin (2008) on folk concepts in biology across various cultures. Let me now return to why computer modeling is important for the cognitive science of science.

Methodology of Computational Modeling

What is the point of building computational models? One answer might come from the hypothetico-deductive view of scientific method, according to which science proceeds by generating hypotheses, deducing experimental predictions from them, and then performing experiments to see if the predicted observations occur. On this view, the main role of computational models is to facilitate deductions. There are undoubtedly fields such as mathematical physics and possibly economics where computer models play something like this hypothetico-deductive role, but their role in the cognitive sciences is much larger.

The hypothetico-deductive method is rarely applicable in biology, medicine, psychology, neuroscience, and the social sciences, where mathematically exact theories and precise predictions are rare. These sciences are better described by what I shall whimsically call the mechanista view of scientific method. Philosophers of science have described how many sciences aim for the discovery of mechanisms rather than laws, where a mechanism is a system of interacting parts that produce regular changes (e.g., Bechtel, 2008; Bechtel & Richardson, 1993; Bunge, 2003; Craver, 2007; Darden, 2006; Machamer, Darden, & Craver, 2000; Thagard, 2006a; Wimsatt, 2007). Biologists, for example, can rarely derive predictions from mathematically expressed theories, but they have been highly successful in describing mechanisms such as genetic variation and evolution by natural selection that have very broad explanatory scope. Similarly, I see cognitive science as primarily the search for mechanisms that can explain many kinds of mental phenomena such as perception, learning, problem solving, emotion, and language.

Computer modeling can be valuable for expressing, developing, and testing descriptions of mechanisms, at both psychological and neural levels of explanation. In contemporary cognitive science, theories at the psychological level postulate various kinds of mental representations and processes that operate on them to generate thinking. For example, rule-based theories of problem solving, from Newell and Simon (1972) to Anderson (2007), postulate (1) representations of goals and if-then rules and (2) search processes involving selection and firing of rules. The representations are the parts and the processes are the interactions that together provide a mechanism that explains mental changes that accomplish tasks. Other cognitive science theories can also be understood as descriptions of mechanisms, for example, connectionist models that postulate simple neuronlike parts and processes of spreading activation that produce mental changes (Rumelhart & McClelland, 1986). Computational neuroscience now deals with much more biologically realistic neural entities and processes than connectionism, but the aim is the same: to describe the mechanisms that explain neuropsychological phenomena.

Expressing and developing such theoretical mechanisms benefits enormously from computational models. It is crucial to distinguish between theories, models, and programs. On the mechanista view, a theory is a description of mechanisms, and a model is a simplified description of the mechanisms postulated to be responsible for some phenomena. In computational models, the simplifications consist of proposing general kinds of data structures and algorithms that correspond to the parts and interactions that the theory postulates. A computer program produces a still more specific and idealized account of the postulated parts and interactions using data structures and algorithms in a particular programming language. For example, the theory of problem solving as rule application using means-ends reasoning gets a simplified description in a computational model with rules and goals as data structures and means-ends search as interactions. A computer program implements the model and theory in a particular programming language such as LISP or JAVA that makes it possible to run simulations. Theoretical neuroscience uses mathematically sophisticated programming tools such as MATLAB to implement computational models of neural structures and processes that approximate to mechanisms that are hypothesized to operate in brains.

Rarely, however, do computer modelers proceed simply from theory to model to program in the way just suggested. Rather, thinking about how to write a computer program in a familiar programming language enables a cognitive scientist to express and develop ideas about what parts and interactions might be responsible for some psychological phenomena. Hence the development of cognitive theories, models, and programs is a highly interactive process in which theories stimulate the production of programs and vice versa. It is a mistake, however, to identify theories with programs, because any specific program will have many details arising from the peculiarities of the programming language used. Nevertheless, writing computer programs helps enormously to develop theoretical ideas expressed as computer models. The computer model provides a general analogue of the mechanisms postulated by the theory, and the program provides a specific, concrete, analogical instantiation of those mechanisms.

In the biological, social, and cognitive sciences, descriptions of mechanism are rarely so mathematical that predictions can be deduced, but running computer programs provides a looser way of evaluating theories and models. A computer program that instantiates a model that simplifies a theory can be run to produce simulations whose performance can be compared to actual behaviors, as shown in systematic observations, controlled behavioral experiments, or neurological experiments.

There are three degrees of evaluation that can be applied, answering the following questions about the phenomena to be explained:

1. Is the program capable of performing tasks like those that people have been observed doing?

2. Does the behavior of the program qualitatively fit with how people behave in experiments?

3. Does the behavior of the program quantitatively fit numerical data acquired in experiments?

Ideally, a computer program will satisfy all three of these tests, but often computer modeling is part of a theoretical enterprise that is well out in front of experimentation. In such cases, the program (and the model and theory it instantiates) can be used to suggest new experiments whose resulting data can be compared against the computer simulations. In turn, data that are hard to explain given currently available mechanisms may suggest new mechanisms that can be simulated by computer programs whose behaviors can once again be compared to those of natural systems. The three questions listed above apply to models of psychological behavior, but analogous questions can be asked about computational simulations of neural data.

The general interactive process of mechanism-based theory development using computational models is shown in figure 1.1, which portrays an interactive process with no particular starting point. Note that the arrows between mechanisms and models, and between models and simulations, are symmetrical, indicating that models can suggest mechanisms and programs can suggest models, as well as vice versa. In one typical pattern, experimental results prompt the search for explanatory mechanisms that can be specified using mathematical–computational models that are then implemented in computer programs. Simulations using these programs generate results that can be compared with experimental results. This comparison, along with insights gained during the whole process of generating mechanisms, models, and simulations, can in turn lead to ideas for new experiments that produce new experimental results.

Figure 1.1

The role of computer models in developing and testing theories about mechanisms. Lines with arrows indicate causal influences in scientific thinking. The dashed line indicates the comparison between the results of experiments and the results of simulations.

Unified Cognitive Science Research

I have described philosophical, psychological, computational, and neuroscientific contributions to the understanding of science, but cognitive science at its best combines insights from all of its fields. We can imagine what an ideal research project in the cognitive science of science would be like, one beyond the scope of any single researcher except perhaps Herbert Simon. Consider a team of researchers operating with a core set of theoretical ideas and multiple methodologies. Let ASPECT stand for some aspect of scientific thinking that has been little investigated. We can imagine a joint enterprise in which philosophers analyze historical cases of ASPECT, psychologists perform behavioral experiments on how adults and children do ASPECT, neuroscientists perform brain scans of people doing ASPECT, and computational modelers write programs that can simulate ASPECT. Linguists and anthropologists might also get involved by studying whether ASPECT varies across cultures. Representatives of all six fields could work together to generate and test theories about the mental structures and processes that enable people to accomplish ASPECT. My own investigations into the cognitive science of science do not have anything like the scope of this imaginary investigation of ASPECT, but they variously combine different parts of the philosophical, historical, psychological, computational, and neuroscientific investigation of scientific thinking.

Unified investigations in the cognitive science of science can be normative as well as descriptive. It is sometimes said that philosophy is normative, concerned with how things ought to be, in contrast to the sciences which are descriptive, concerned with how things are. This division is far too simple, because there are many applied sciences, from engineering to medicine to clinical and educational psychology, that aim to improve the world, not just to describe it (Hardy-Vallée & Thagard, 2008). Conversely, if the norms that philosophy seeks to develop are to be at all relevant to actual human practices, they need to be tied to descriptions of how the world, including the mind, generally works. I have elsewhere defended the naturalistic view that philosophy is continuous with science, differing in having a greater degree of generality and normativity (Thagard, 2009, 2010a). This book assumes the priority of scientific evidence and reasoning over alternative ways of fixing belief such as religious faith and philosophical thought experiments, but I argue for that assumption in Thagard (2010a, ch. 2).

The cognitive science of science can take from its philosophical component and also from its applied components a concern to be normative as well as descriptive. An interdisciplinary approach to science can aim not only to describe how science works, but also to develop norms for how it might work better. The methodology is captured by the following normative procedure (adapted from Thagard, 2010a, p. 211):

1. Identify a domain of practices, in this case ways of doing scientific research.

2. Identify candidate norms for these practices, such as searching for mechanisms.

3. Identify the appropriate goals of the practices in the given domain, such as truth, explanation, and technological applications.

4. Evaluate the extent to which different practices accomplish the relevant goals.

5. Adopt as domain norms those practices that best accomplish the relevant goals.

The descriptive side of cognitive science is essential for all of steps 1–4, but description can quickly lead to normative conclusions via the assessment shown in steps 4–5. My concern in the cognitive science of science is primarily descriptive, but normative issues will arise in chapters on climate change (ch. 5), truth (ch. 6), and values (ch. 17).

Other Approaches to Studying Science

Cognitive science is not the only way to study the practices and results of science, and there are alternative approaches that are antagonistic to it. Cognitive science is scorned by some philosophers, historians, and sociologists who view it as fundamentally inadequate to understand the process of science and other important aspects of human life. I will now concisely review some of these alternatives, and describe why I think their opposition misses the mark.

Within philosophy, the cognitive science of science exemplifies naturalism, the view that philosophical deliberations should be tied to scientific evidence. Naturalistic philosophy has a venerable history, with practitioners such as Aristotle, Epicurus, Bacon, Locke, Hume, Mill, Peirce, Dewey, Quine, and many contemporary philosophers of science and mind. But philosophy also has a strong antinaturalistic strain, which challenges the relevance of science to philosophy from various directions. One prominent challenge seeks philosophical truths from reason alone, independent of scientific evidence; such truths are pursued by Plato, Kant, Frege, Husserl, and contemporary philosophers who try to use thought experiments to arrive at conceptual truths (for critiques of this approach, see Thagard, 2009, 2010a). This reason-based approach to philosophy tends to be antagonistic to cognitive science on the grounds that mind, like everything else, can be understood most deeply by methods that are a priori—independent of sense experience.

The antipsychologistic tendency of philosophy is also evident in contemporary work in the philosophy of science that employs formal methods such as symbolic logic, set theory, and probability theory. All of these tools are potentially relevant to the cognitive science of science, but formal philosophy of science uses them to the exclusion of many other tools (including the varied computational ones mentioned above) that cognitive science can bring to the examination of scientific knowledge. Formal philosophy of science follows in the tradition of the logical positivists in assuming that scientific theories are best viewed as abstract structures rather than as mental representations. Such abstractions are of limited use in understanding the actual practice of science and the details of the growth of scientific knowledge. It is particularly odd that the philosophy of science should ignore branches of science such as psychology that are highly relevant to understanding how science works, but the continuing influence of the Fregean, antipsychologistic strain of analytic philosophy is large.

A very different challenge to naturalistic philosophy comes from a more nihilistic direction that is generally skeptical of scientific and philosophical claims to achieve knowledge. Philosophers such as Nietzsche, Heidegger, Derrida, Foucault, and Lyotard rejected Enlightenment values of evidence, rationality, and objectivity. From a postmodernist perspective, science is just another human enterprise beset by power relations, whose discourse can be investigated by the same hermeneutic means that apply to other institutions. Cognitive science is then merely an attempt by scientists and science-oriented philosophers to exaggerate their own importance by privileging one style of thinking. In contrast, chapter 6 below provides a defense of scientific realism, the view that science aims to achieve truth and sometimes succeeds.

The postmodernist rejection of science as a way of knowing the world has infected much work in the history and sociology of science. Around the same time that the cognitive science of science was taking off, an alternative movement arose that managed to take over science studies programs at many universities. Sociologists of science produced a research program called the Sociology of Scientific Knowledge that abandoned the normative assessment of science in favor of purely sociological explanations of how science develops (e.g., Barnes, Bloor & Henry, 1996). Latour and Woolgar (1986) even called for a ten-year moratorium on cognitive explanations of science until sociologists had had a chance to explain all aspects of scientific development. That moratorium has long expired, and sociologists have obviously left lots of science to be explained. Moreover, some prominent proponents of postmodern sociology of science have made the shocking discovery that science and technology might even have something to do with reality (Latour, 2004).

In contrast to the imperialism of sociologists who think they can explain everything about scientific development, the cognitive science of science is friendly to sociological explanations. Power relations are undoubtedly an important part of scientific life, from the local level of laboratory politics to the national level of funding decisions. Like some analytic philosophers, some sociologists suffer from psychophobia, the fear of psychology, but cognitive approaches to science are compatible with the recognition of important social dimensions of science. For example, in my study of the development and acceptance of the bacterial theory of ulcers, I took into account social factors such as collaboration and consensus as well as psychological processes of discovery and evaluation (Thagard, 1999). Other works in the cognitive science of science have similarly attended to social dimensions (e.g., Dunbar, 1997; Giere, 1988). The cognitive and the social sciences should be seen as complements, not competitors, in a unified enterprise that might be called cognitive social science. Anthropology, sociology, politics, and economics can all be understood as requiring the integration of psychological and social mechanisms, as well as neural and molecular ones (Thagard, 2010d, forthcoming-c). Novel kinds of computer models are needed to explore how the behavior of groups can depend recursively on the behavior of individuals who think of themselves as members of groups. Agent-based models of social phenomena are being developed, but they are only just beginning to incorporate psychologically realistic agents (Sun, 2008a; Thagard, 2000, ch. 7, presents a cognitive-social model of scientific consensus). The aim of these models is not to reduce the social to the psychological and neural, but rather to show rich interconnections among multiple levels of explanation. My hope is that future work