1
Editorsâ Introduction
Michael E. Gorman
University of Virginia
Ryan D. Tweney
Bowling Green State University
David C. Gooding
University of Bath
Alexandra P. Kincannon
University of Virginia
At the turn of the 21st century, the most valuable commodity in society is knowledge, particularly new knowledge that may give a culture, a company, or a laboratory an adaptive advantage (Christensen, 1997; Evans & Wurster, 2000; Nonaka & Takeuchi, 1995). Turning knowledge into a commodity poses two dangers. One is to increase the risk that one culture, company, or group can obtain an unfair advantage over others. The other is to impose a one-dimensional, goal-driven view of something that, as the chapters in this volume will show, is subtle, complex, and diverse as to its motivations and applications. To be sure, knowledge about the cognitive processes that lead to discovery and invention can enhance the probability of making valuable new discoveries and inventions. However, if made widely available, this knowledge could ensure that no particular interest group âcorners the marketâ on techno-scientific creativity. It would also facilitate the development of business strategies and social policies based on a genuine understanding of the creative process. Furthermore, through an understanding of principles underlying the cognitive processes related to discovery, educators can use these principles to teach students effective problem-solving strategies as part of their education as future scientists.
A special focus of this volume is an exploration of what fine-grained case studies can tell one about cognitive processes. The case study method is normally associated with sociology and anthropology of science and technology; these disciplines have been skeptical about cognitive explanations. If there is a well-established eliminativism among neuroscientists (Churchland, 1989), there is also a social eliminativism that seeks to replace cognitive accounts with sociological accounts (Woolgar, 1987). In these socio-anthropological case studies, interactions, inscriptions, and actions are made salient. The private mental processes that underlie public behavior have to be inferred, and most anthropologists and sociologists of science are trained to regard these cognitive processes as epiphenomenal. By contrast, intellectual or internalist studies by historians of science go into reasoning processes in detail (Drakes, 1978; Mayr, 1991; Westfall, 1980). Historians of science have produced a large number of studies that describe processes of experimentation, modeling, and theory construction. These studies can be informative about reasoning and inference in relation to declarative knowledge, experiential knowledge and experimental data (Galison, 1997; Gooding, 1990; Principe, 1998), visualization (Rudwick, 1976; Tweney & Gooding, 1991; Wise, 1979), and the dynamics of consensus formation through negotiation and other forms of personal interactions (Rudwick, 1976). However, historians generally have no interest in identifying and theorizing about general as opposed to personal and culture-specific features of creative processes. Thus, very few historical studies have combined historical detail with an interest in general features of the creative process (for exceptions, see Bijker, 1995; Carlson & Gorman, 1990; Giere, 1992; Gooding, 1990; Gruber, 1974; Law, 1987; Miller, 1986; Tweney & Gooding, 1991; Wallace & Gruber, 1989).
Despite the usefulness of fine-grained case studies, cognitive psychologists have traditionally lamented their lack of rigor and control. How can one identify general features, let alone develop general principles of scientific reasoning, from studies of a specific discovery, however detailed they may be? One answer is to develop the sorts of computational models preferred by cognitive scientists (Shrager & Langley, 1990). One classic example of this kind of modeling is, of course, Kulkarni and Simonâs (1988) simulation of a historical account by Larry Holmes (1989) dealing with Hans Krebsâs discovery of the ornithine cycle (see also Langley, Simon, Bradshaw, & Zytkow, 1987). These models can be abstracted from historical cases as well as current, ethnographic ones. However, it is important to remember that such models are typically designed to suit the representational capabilities of a particular computer language. Models derived from other domainsâsuch as information theory, logic, mathematics, and computability theoryâcan become procrustean beds, forcing the territory to fit the researcherâs preconceived map (Gorman, 1992; Tweney, 1990).
The tension among historical, sociological, and cognitive approaches to the study of science is given thorough treatment in the chapter 2, by Nersessian. She distinguishes between good old-fashioned artificial intelligence, represented by computer programs whose programmers claim they discover scientific laws, and social studies of science and technology, represented by detailed or âthickâ descriptions of scientific practice. Proponents of the former approach regard cognition as the manipulation of symbols abstracted from reality; proponents of the latter see science as constructed by social practices, not reducible to individual symbol systems. Nersessian describes a way for cognitive studies of science and technology to move beyond these positions, taking what she calls an environmental perspective that puts cognition in the world as well as in the brain. Ethnography can be informative about cognitive matters as well as sociological ones, as Nersessianâs study of biomedical engineering laboratories shows, and historical research helps make the cognitive practices intelligible.
In order to ground cognitive theory in experimental data, psychologists have conducted reasoning experiments, mainly with college students but also with scientists. These experiments allow for controlled comparisons, in which all participants experience the same situation except for a variable of interest that is manipulated. A psychologist might compare the performance of scientists and students on several versions of a task. Consider, for example, a simple task developed by Peter Wason (1960). He showed participants in his study the number triplet â2, 4, 6â and asked them to propose additional triplets in an effort to guess a rule he had in mind. In its original form, the experimenterâs rule was always âany three increasing numbers,â which proved hard for most participants to find, given that the starting example suggests a much more specific rule (e.g., âthree even numbersâ). Each triplet can be viewed as an experiment, which participants used to generate and test hypotheses. For Wason, participants seemed to manifest a confirmation bias that made it hard to see the need to disconfirm their own hypotheses. Later research suggested a much different set of explanations (see Gorman, 1992, for a review). Tasks such as this one were designed to permit the isolation of variables, such as the effect of possible errors in the experimental results (Gorman, 1989); however, they are not representative of the complexities of actual scientific practice. The idealization of reasoning processes in scientific and technological innovation (Gorman, 1992; Tweney, 1989), and of scientific experiments in computer models (Gooding & Addis, 1999), is a critical limitation of this kind of research.
What makes models powerful and predictive is their selectivity. A good model simplifies the modeled world so as to make it amenable to oneâs preferred methods of analysis or problem solving. However, selectivity can make oneâs models limited in scope and, at worst, unrealistic. Insofar as this is a problem, it is often stated as a dichotomy between abstraction and realâ ism, as for example by the mathematician James Gleick: âThe choice is always the same. You can make your model more complex and more faithful to reality, or you can make it simpler and easier to handleâ (Gleick, 1987, p. 278). David Gooding illustrated this problem at a workshop with a joke that could become a central metaphor for science and technology studies:
A millionaire with a passion for horse racing offered a large prizeâenough to buy a few Silicon Graphics machinesâto anyone who could predict the outcome of any horse race. Three scientists took up the challenge, a physiologist, a geneticist and a theoretical physicist. One year later the three scientists announced their results. Hereâs what each reported:
The Physiologist: âI have analysed oxygen uptake, power to weight ratios, dietary intake and metabolic rates, but there are just too many variables. I am unable to predict which horse will win.â
The Geneticist: âI have examined blood lines, breeding programs and all the form books, but there are just too many uncertainties. I cannot predict who will win any race.â
The Physicist: âI have developed a theoretical model of the dynamics of horse racing, and have used it to write a computer program that will predict the outcome of any horse race to 7 decimal places. I claim the prize. Butâthere is one proviso. The model is only valid for a perfectly spherical horse moving through a vacuumâ1
Experimental simulations of scientific reasoning using tasks like Wasonâs (1960) 2â4â6 task are abstractions just like the spherical horse: They achieve a high degree of rigor and control over participantsâ behavior but leave out many of the factors that play a major role in scientific practice. Gooding (in press) argues that in the history of any scientific field there is a searching back and forth between models and real world complexities, to achieve an appropriate level of abstractionânot overly simple, capturing enough to be representative or valid, yet not so complex as to defeat the problem-solving strategies of a domain. Goodingâs chapter for this volume (chap. 9) shows how scientists use visualization in creating models that enable them to negotiate the tension between simplicity and solvability on the one hand and complexity and real world application on the other. He argues that, like any other scientific discipline, cognitive studies of science and technology must find appropriate abstractions with which to describe, investigate, model, and theorize about the phenomena it seeks to explain.
To compare a wide range of experiments and case studies conducted in different problem domains, one needs a general framework that will establish a basis for comparison. As Chris Schunn noted in the workshop on cognitive studies of science and technology that inspired this volume (see Preface), models, taxonomies, and frameworks are like toothbrushesâno one wants to use anyone elseâs. In science and technology studies, this has been the equivalent of the ânot invented hereâ syndrome. This usually reflects the methodological norms of a discipline, such as the sociological aversion to cognitive processes, which is reminiscent of behavioral psychologyâs rejection of mental processes as unobservables. One strategy for achieving de facto supremacy is to assume, even if one cannot demonstrate it, that oneâs own âtoothbrushâ is superior to any other (Gorman, 1992).
Discovery
Sociological and historical studies of the resolution of scientific controversies have shown that the supremacy of a particular theory, technology, or methodological approach involves negotiation. Because no method is epistemologically neutral, this negotiation often focuses on the validity of the method (s) of establishing facts and of making inferences from them (Galison, 1997). Therefore, rather than promoting the investigative potential of a single method, we advocate approaches that address both Goodingâs problem of abstraction and Schunnâs problem of shareable frameworks. Dunbar and Fugelsang develop one such approach in their contribution (chap. 3) to this volume. This approach combines experiments, modeling and case studies in a complementary manner. They develop the distinction (first made by Bruner, Goodnow, & Austin, 1956) between in vitro studies of scientific thinking (which involve abstract tasks like Wasonâs [1960] 2â4â6 task) and in vivo studies (which involve observing and analyzing scientific practice). Dunbar (1999) used an in vitro task to study how participants reasoned about a genetic control mechanism and conducted in vivo studies of molecular biology laboratories. In chapter 3, Dunbar and Fugelsang label four more approaches in the same style:
- Ex vivo research, in which a scientist is taken out of her or his laboratory and investigated using in vitro research, by presenting problems similar to those he or she would use in his or her research.
- In magnetico research, using techniques such as magnetic resonance imaging to study brain patterns during problem solving, including potentially both in vitro and in vivo research.
- In silico research, involving computational simulation and modeling of the cognitive processes underlying scientific thinking, including the good old-fashioned artificial intelligence work cited by Nersessian and alternatives.
- Sub specie historiae research, focusing on detailed historical accounts of scientific and technological problem solving. These in historico studies can serve as data for in silico simulations.
Later chapters offer a variety of examples of sub specie historiae and in vivo studies, with references to the other types of research noted earlier.
In chapter 4, Klahr takes a framework that he was involved in developing and stretches it in a way that makes it useful for organizing and comparing results across chapters in this volume. His idea is that discovery involve searches in multiple problem spaces (Klahr & Dunbar, 1988; Simon, Langley, & Bradshaw, 1981). For example, a scientist may have a set of possible experiments she might conduct, a set of possible hypotheses that might explain the experimental results, and a set of possible sources of experimental error that might account for discrepancies betwe...