1. SOLOMON'S CONCERN: DYSFUNCTIONAL GROUPS

Solomon notes that deliberation and critical discussion have long been celebrated as the most effective means to realizing our epistemic goals. A number of influential philosophers, including John Stuart Mill (see Mill Reference Mill1956), and philosophers of science, notably Karl Popper and Helen Longino (see Solomon Reference Solomon2006: 28), argue that critical discussion provides the most promising path to the truth.Footnote ² These philosophers claim critical discussion leads people to refine and strengthen their views, altering them as needed in response to the criticisms of others (see Solomon Reference Solomon2006: 38).

Solomon, though, believes that “there is room for skepticism about the results of criticism, since sometimes deliberation makes things worse rather than better” (Solomon Reference Solomon2006: 38). She notes that studies in organizational research, often done by scholars at business schools, suggest that group deliberation is remarkably ineffective, often leading to suboptimal outcomes (see Solomon Reference Solomon2006: 31). For example, Solomon notes that “group deliberation often produces worse decisions than can be obtained without deliberation” and that “a group of nonexperts often produces better decisions on a topic than does an expert about that topic” (Solomon Reference Solomon2006: 31).Footnote ³

Most importantly, Solomon suggests that when groups deliberate “pressure to reach a consensus can … lead (1) to phenomena such as groupthink and (2) to the suppression of relevant data” (Solomon Reference Solomon2006: 28; numerals added). These concerns are related. When a group deliberates in their efforts to solve a problem, individual members of the group may be led to suppress information they have that may threaten to undermine a consensus that is emerging in the group. They may be compelled by peer pressure or intimidated by a forceful group leader to withhold information that would stall the group in reaching a consensus. But the specific information that they are suppressing may be just the sort of information that would have led the group to reach a different decision than the one they reached.

Solomon thinks that a more promising path to reaching a rational consensus is through judgment aggregation (see Solomon Reference Solomon2006: 31). When we aggregate judgments, we solicit each individual's judgment independently, thus preventing some individuals from adversely influencing the judgments of others in the group. Individuals are thus less likely to suppress the information they have, information that conflicts with the views of the most vocal members of the group. In this way, valuable information is less likely to be lost. Indeed, there is a growing body of literature on the wisdom of crowds that cites impressive examples where aggregation leads to effective solutions. For example, Francis Galton famously noted that the mean of the estimates of the weight of an ox of “approximately 800 people visiting a fair in Plymouth” was within one pound of the real weight of the ox (see Solomon Reference Solomon2006: 34; also Surowiecki Reference Surowiecki2004: xi–xiii). Solomon notes that unlike deliberation which can lead to groupthink, “aggregation without deliberation preserves information” (Solomon Reference Solomon2006: 36; emphasis in original). Hence, the resulting view is reached on the basis of a consideration of all the available data.

Solomon also argues that philosophers have exaggerated the value of consensus (see Solomon Reference Solomon2001). As far as she is concerned, it can be an impediment to reaching the truth. Dissent in a community helps preserve information. As long as dissenting views are heard, then valuable information is less likely to be lost. Solomon's defense of dissent is similar to Mill's defense of the protection of free thinking and free speech in On Liberty. Mill argues that when people are permitted to express dissenting opinions, the views we accept are kept alive, rather than being accepted uncritically and becoming dead dogmas (see Mill Reference Mill1956: Ch. 2). When dissenting views are suppressed, on the other hand, people quickly forget why the accepted view is a rational view. Unlike Solomon, though, Mill does believe that deliberation is an effective means to getting at the truth.Footnote ⁴

2. WHICH GROUPS?

The precise target of Solomon's analysis is far from clear. At times she seems to be concerned with social epistemology in general, and any sort of situation in which we might encounter a group deliberating about a topic with the aim of determining a course of action. At other times, though, she seems to have science and scientific inquiry specifically in mind. She does cite Longino (Reference Longino1990) and explicitly refers to Popper, and distinguishes her own view from their views. In what follows, I will assume her analysis was intended as a contribution to the social epistemology of science. I assume her aim is to demonstrate the relevance of both groupthink and judgment aggregation to the epistemology of science.

There is a further ambiguity about the target of Solomon's analysis. It is not exactly clear what types of groups of scientists she is concerned about. Given the authors she explicitly addresses, and her own past writings, I assume that Solomon is concerned that a research community as a whole may be prone to groupthink.Footnote ⁵ Specifically, it seems that her concern is that a research community or scientific specialty may settle on a theory prematurely because individual scientists in the community may withhold information that would challenge the theory about which a consensus is emerging either because of peer pressure or unwarranted pressure from community leaders, those in the research field who command power over resources.Footnote ⁶

In the remainder of this section, I want to explain why the threat of groupthink in this context is unfounded, and also why judgment aggregation is equally irrelevant to understanding this dimension of the culture of science. Solomon, I argue, is focusing on the wrong sorts of groups.

Scientific specialties are not the sorts of groups that deliberate. This is a point I have discussed at length elsewhere (see Wray Reference Wray2007). The scientists working in a specialty do not gather together to determine when a theory should be accepted, or whether in fact a long-accepted theory should be replaced. Instead, when there is a consensus in a scientific research community it is a consequence of many individual scientists accepting the same theory. Different scientists will be compelled by different evidence, given their training and past experiences, and no individual scientist will have access to all the relevant evidence. Some scientists may even be persuaded to accept a new theory by the leaders in their field, rather than as a result of their own assessment of the evidence (see Wray Reference Wray2011: Ch. 11). But as a matter of fact sometimes scientific communities do reach a consensus. Geologists, for example, came to accept the theory of plate tectonics in the mid-1960s and early 1970s. But geologists did not need to deliberate as a group in order to reach a consensus. Groupthink, though, only affects groups that reach a consensus by means of deliberation. Groups prone to groupthink must aim to settle on a view that will stand as the group's views. Hence, given that scientific specialties are not the sorts of groups that aim to let a view stand as the view of the group, they are not prone to groupthink.

There is another reason for thinking that Solomon is mistaken in claiming that scientific specialty communities could be prone to groupthink. It seems that such groups do not have key features that the sorts of groups that have been identified as prone to groupthink have.

There is now a vast body of literature on groupthink, much of it concerned with understanding when and what types of groups are prone to the phenomenon (see, for example, McCauley Reference McCauley1989; Aldag and Fuller Reference Aldag and Fuller1993; Esser Reference Esser1998). Irving Janis, the psychologist who first introduced the concept, originally appealed to it as a means to explain decision-making fiascos in international policy contexts (see Janis Reference Janis1971; also Aldag and Fuller Reference Aldag and Fuller1993). For example, Janis examined the decision making that led to the United States' support for the Bay of Pigs invasion in the early 1960s, a key event in the cold war. He also examined the decision making that led to the United States' decision to engage in a war in Vietnam. Other researchers have suggested that groupthink may occur in other contexts. Indeed, psychologists have conducted numerous laboratory studies aimed at determining what sorts of contexts and social groups give rise to the threat of groupthink. The various laboratory studies of groupthink assign tasks to groups of research subjects, tasks that are far removed from the sorts of situations that initially inspired Janis to invoke the concept. In principle any sort of deliberating group could be prone to groupthink and determining which groups are in fact prone to it requires empirical investigation.

Janis is quite explicit about the sorts of features that characterize the groups that are prone to groupthink.

1. (1) Such groups stick “with the policies to which the group has already committed itself, even when those policies are obviously working out badly”;

2. (2) They make unrealistic appraisals “of alternative courses of action”;

3. (3) The “members of decision-making groups … avoid being too harsh in their judgments of their leaders' or their colleagues' ideas” (Janis Reference Janis1971: 43).

It seems clear that scientific specialties lack these features. They do not make policies. They do not make appraisals of courses of action. And it is not clear that the members of specialties avoid being harsh in their judgments of colleagues' ideas. Further, Solomon provides no evidence that specialties are similar in the relevant respects to the sorts of groups that Janis identified as prone to groupthink. In fact, she does not discuss the features that allegedly characterize groups prone to groupthink. And, to repeat what I have argued above, scientific specialties are not decision-making bodies. The consensus that does emerge in such communities is not a consequence of deliberation.Footnote ⁷ The exchange of ideas that plays out in the relevant scientific journals is not deliberation in the relevant sense. The various publications that are part of this exchange are not directed toward the same end. Deliberation, though, is directed toward an end, a plan of action. It is in the efforts to realize a shared goal that groups of individuals are led to act in ways that result in the fiascos associated with groupthink.

Judgment aggregation seems equally inapplicable to scientific specialties. There is no designated group or individual assigned the job of determining what the consensus or view of a research community should be. Philip Kitcher, in his modeling of the division of cognitive labor in science, has invoked the idea of a powerful science planner, a philosopher-monarch, with the power and responsibility to determine who works on what research programs (Kitcher Reference Kitcher1993: 305). But even Kitcher knows no such figure exists. He invokes the idea as a heuristic. He was concerned to evaluate the effectiveness of the current division of labor in science, a division of labor that tends to lead to an excessive concentration of scientists on some research problems and strategies at the cost of neglecting other research problems and strategies.

Indeed, there is a deeper problem with Solomon's appeal to judgment aggregation. Solomon seems to assume that we all know what is involved in aggregating judgments. Her appeal to Galton's ox example makes it look like a straightforward procedure. In that example, one merely calculates the mean of the various individual judgments of the ox's weight. But judgment aggregation is a far more complicated affair, as is evident from the vast technical literature addressing the issue. Indeed, there are lively debates about how we should aggregate the judgments of a group. And things can be complicated further when we consider aggregating judgments that may involve chains of reasoning. For example, there is a famous paradox that arises when we consider a group's judgments of a set of premises and the group's judgment of a conclusion that is deductively entailed by the premises. Though the majority of the members of the group may accept a set of premises, the majority may not accept a conclusion that is deductively entailed by the premises (see Dietrich Reference Dietrich and Kaldis2013: 512–13). And there is no universally accepted way to resolve such problems. Hence, judgment aggregation is not, at least not always, the straightforward procedure that Galton's ox example makes it out to be.

It would seem that the sort of judgment aggregation that would be relevant in the context of science would be the complex sort, and not merely the sort that takes the mean of various individual judgments on an issue. In scientific contexts where groups may be tempted to aggregate their judgments the groups are dealing with complex chains of reasoning.

There is a general methodological point to which I want to draw attention. Philosophers are encouraged to appeal to whatever conceptual and theoretical resources they find useful in their efforts to understand the epistemic culture of science. The science of science, including the philosophy of science, is no different than other areas of science. Methodological pluralism can be beneficial. But the onus is on the philosopher to show that the concepts and theoretical frameworks that were developed in one context, for example, international policy making, are applicable to the study of the culture of science. In this instance I do not think Solomon has made the case that the concept “groupthink” or the theoretical framework of judgment aggregation illuminates the issue at hand. Neither shed light on how or when a consensus is or should be reached in a scientific specialty.

Let me be clear about what I am not claiming. I am not objecting to Solomon's appeal to psychological research as a means to shed light on the culture of science. Indeed, there is a long tradition of naturalizing the epistemology of science by drawing on insights from research in psychology. Thomas Kuhn (Reference Kuhn1962/2012) did it to great effect, drawing on the anomalous playing card experiments of Bruner and Postman, the experiments involving inverted visual images, and other studies as well. The objection I raise is with respect to the specifics of Solomon's appeal to groupthink. So far it seems that the notion of groupthink is irrelevant to science. Certainly when examining the dynamics of scientific specialty communities and the erosion and emergence of consensus in them, it seems that groupthink is not a relevant concept. James Esser (Reference Esser1998) has rightly noted that “the risk of casually characterizing poor group decisions as groupthink without careful and thorough analysis, is to make groupthink a convenient label with little explanatory or predictive value” (Esser Reference Esser1998: 126). If the concept is to be useful in illuminating group dynamics, then it must be used with precision, and applied with restraint. Concepts drawn from psychology, especially those that have been popularized, like “the tipping point” and “wisdom of crowds,” risk becoming facile clichés.

3. COLLABORATION AND GROUPTHINK

There is another social context in science to which the concept of groupthink may be relevant. Specifically, I am thinking of collaborative research teams. Research teams, unlike scientific specialties, do aim to reach a consensus on an issue. Some sort of consensus seems necessary if the group is to publish the results of their research in a co-authored article. The group must, for example, resolve the issues of how to interpret their data, or which methods to use in their experiments. And deliberation plays a crucial role in the determination of the group's view. Hence, such groups are at least in principle vulnerable to groupthink.

It is worth noting that the types of groups that engage in collaborative research in science are a heterogeneous lot. There are collaborations involving two scientists, collaborations involving a team working in a laboratory together, and enormous collaborations that are tied to permanent and costly research facilities, like the research teams that work at CERN or Fermilab. So we need to be mindful of these differences in determining which sorts of scientific groups are prone to groupthink, if in fact any sorts of groups are. In the remainder of the paper, I will be concerned with groups of scientists that produce coauthored articles, articles that present the view of the group as a whole.

Let us first consider what insights we might gain from the scholarly literature on groupthink that has developed since the publication of Janis' original paper on the topic, specifically the laboratory studies of groupthink. Aldag and Fuller (Reference Aldag and Fuller1993) and Esser (Reference Esser1998) provide useful summaries of the findings from the laboratory studies of groupthink. The various laboratory studies involve bringing together experimental subjects to work as groups on solving a problem. A number of variables have been manipulated in these experiments. Some have included compatriots of the experimenters, planted in the group to determine the effects of a forceful leader (Esser Reference Esser1998: 131). Other studies have involved time constraints, for it appears that when a group must make a decision in a timely fashion they are prone to a variety of dysfunctional behaviors, including groupthink. In some studies, experimental groups were given misinformation about the cohesiveness of the group (see Esser Reference Esser1998: 127). Roughly, cohesiveness is a measure of the extent to which the group has “esprit de corps and they work together on a continuing basis” (Flowers Reference Flowers1977: 888).Footnote ⁸

Let me briefly summarize some of the key findings of these laboratory studies.

1. (1) One study found that “open leadership style produced more suggested solutions” to the problem the group faced and that such groups use more of the “available facts than did [groups with a] closed leadership style” (see Aldag and Fuller Reference Aldag and Fuller1993: 536).

2. (2) Another study found that “contrary to the groupthink prediction, members of noncohesive groups exhibited more self-censorship than did members of cohesive groups” (see Esser Reference Esser1998: 127).

3. (3) Another study, though, found that “cohesive groups were more discouraging of dissent than noncohesive groups. However … cohesive groups reported less self-censorship and generated more alternatives than noncohesive groups” (Esser Reference Esser1998: 130).

4. (4) Another study found that “groups with directive leaders used less available information, suggested fewer solutions, and rated their leader as more influential in the decision process than groups with nondirective leaders” (Esser Reference Esser1998, 131).

Some of these findings might have some bearing on how we organize or manage research teams in science. But the findings are rather general and provide no concrete recommendations for organizing and managing a research team. Indeed, in order to draw specific recommendations one would need to compare the social structure of the collaborative research team and the constraints on their decision making with the social structure and constraints on the specific groups in the laboratory studies. Some studies may thus be relevant or insightful to some collaborative research teams, but not relevant or insightful to others. So far we are not yet in a position to make specific recommendations to scientists working collaboratively.

4. PROBLEM-SOLVING AND COLLABORATIVE RESEARCH

There may be a more fruitful way to draw on the research on groupthink to make advances in the social epistemology of collaborative research. Aldag and Fuller provide a general model of group problem-solving, a framework that may be useful for understanding the challenges that collaborative research teams are apt to face.

Aldag and Fuller note that problem-solving involves a number of distinct stages:

1. (1) problem identification,

2. (2) alternative generation,

3. (3) alternative evaluation and choice,

4. (4) decision implementation, and

5. (5) decision control. (Aldag and Fuller Reference Aldag and Fuller1993: 541; numerals added)

They argue “that groupthink defects focus primarily on the first three stages of the problem-solving process” (Aldag and Fuller Reference Aldag and Fuller1993: 541). That is, groupthink threatens (1) the problem identification stage, (2) the alternative generation stage, and (3) the alternative evaluation and choice stage. Importantly, Aldag and Fuller also argue that

Aldag and Fuller were not concerned with research teams. Rather, they were concerned with developing a general model of group problem-solving and decision-making. But let us apply their insights to the context of scientific research teams. It seems that when a laboratory team is seeking to generate alternative solutions to a problem, interaction may lead members of the team to suppress their ideas, and thus lead to groupthink. That is, at the stage of generating alternative solutions, team members may be inclined to suppress their ideas, especially if they sense any sort of resistance from others in the group. But once the various individual members of the team have generated a number of alternative solutions, an interactive discussion with the others is crucial to effectively evaluate the various proposed solutions. That is, the group becomes an asset once alternative solutions are generated. The group becomes a valuable source of criticism, and the differences between the individual members ensure that the various alternatives will be scrutinized from multiple perspectives.

It may be tempting to think that we can rehabilitate the distinction between the context of discovery and the content of justification as a means to make sense of the findings on groupthink. In this framework the lessons learned from the groupthink research could be summarized as follows. In the context of discovery when alternative hypotheses are needed deliberation can be an impediment. But in the context of justification matters are quite different. There, we are choosing between competing alternatives, and deliberation is a useful resource. Deliberation is thus restricted to the context of justification. In the context of discovery deliberation appears to be unnecessary, even counterproductive.

But this analysis is a bit too simplistic. It fails to take into account the fact that there are many stages in a research project where we need to encourage the development of alternatives, and where premature deliberation may stifle creativity and the advancement of science. At the stages in the research project where the team needs to generate multiple possible interpretations of the data, or multiple plausible explanatory hypotheses, group members may need a chance to work alone, shielded from the critical scrutiny of the other team members. Indeed, we may also want to encourage the generation of numerous alternatives when scientists are trying to determine what methods or instruments to employ. Premature deliberation about such matters could stifle the creativity of scientists and suppress interesting and novel ideas. But in order to glean the full benefits of working with a group the various competing interpretations or explanations should be scrutinized by different people. Hence, at the hypothesis evaluation stage deliberation seems to be a valuable resource afforded by teamwork. Similarly, after the individual members of the research team have generated ideas about which methods to employ, deliberation would aid the group in determining which proposal is apt to be most effective.

Clearly, this sort of information about group problem-solving seems relevant to the way we should structure and manage research teams in science.Footnote ¹⁰

5. JUDGMENT AGGREGATION

It is worth briefly reflecting on the relevance of judgment aggregation to scientific research teams. I doubt that the sort of work that scientists do when they work in research teams would benefit from the straightforward sort of judgment aggregation that Galton discusses in the determination of the weight of the ox at the country fair. That is, I doubt that the interests of a research team can be served by merely calculating the mean of a series of individual judgments of some value. Science is far more complex than that.

Where judgment aggregation is relevant to the work of a research team, it would be at Stage 3 in the process described by Aldag and Fuller, the evaluation of alternatives. But scientists are apt to be reluctant to relinquish control over what views they accept as a research team to a mechanical aggregation process. But given Solomon's conception of judgment aggregation, that is what judgment aggregation requires. For judgment aggregation to work effectively, we need to “preserve” the information that the various individuals in a group have. And this requires that each individual member's judgment is independent of the other members' judgments. But this would amount to having each scientist work alone, and merely gather together to aggregate their judgments. Such an approach misses the point of collaborative research. Scientists collaborate in order to access resources, skills and knowledge they do not individually have.

Perhaps one scientific context where this straightforward sort of judgment aggregation might be useful and appropriate is with respect to committees of scientists commissioned to issue policy recommendations. Such an example is discussed at length by John Beatty (Reference Beatty2006). In the 1950s in the United States a number of biologists were asked to make recommendations about safe levels of radiation exposure for humans. Some of those involved questioned whether scientists had adequate knowledge about radiation and its effects to make a warranted assessment and recommendation. But the biologists involved also knew that if they did not provide a recommendation someone else would be asked to provide one, and that someone else may have been even less qualified than the biologists to do so. Given that the biologists involved in this process disagreed quite extensively, judgment aggregation may have provided an expedient and reasonable solution to the problem at hand. Importantly, the situation these biologists faced is similar in some important respects to the situation that the political advisors faced in the examples discussed by Janis. The biologists had a pressing problem on their hands, and there were significant time constraints, as the government was determined to issue a policy statement about the safe levels of radiation exposure. The government was not going to wait until more research on the topic was completed. Consequently, we should not be surprised if groupthink is a genuine threat in contexts in which scientists are pressed to make policy recommendations. The conditions are very similar to those that led to the fiascos in international policy-making studied by Janis.

A more complex form of judgment aggregation probably does have a role to play in scientific research teams. Research teams, after all, must reach agreement on what views are expressed in their co-authored publications. Where agreement cannot be easily reached, it would make sense to rely on some mechanism or other to enable the various members of the group to transcendent their differences and settle on a view that will stand as the view of the group. But in this context, all the problems that arise in the technical literature on judgment aggregation are apt to arise, for there is no indisputably correct way to resolve the differences of opinion of a group. Resolving the larger question about what the proper means is to aggregate judgments is beyond the scope of this paper.

One proposal that has a prima facie plausibility is that it is probably best if different scientific research teams use different methods of judgment aggregation. In this way, collectively the scientific research community benefits from a diverse portfolio of methods, and in the long run scientists can study the success rates of different methods of aggregation. This proposal is in the spirit of naturalized epistemologies, for it recognizes that many questions in epistemology are empirical questions, and can only be adequately answered by appeal to empirical data.

6. CONCLUDING REMARKS

My aim has been to evaluate the extent to which we could learn something about how we should be conducting collaborative research in science from the research on groupthink. Solomon seems to have set us in the wrong direction, failing to recognize that the consensus in scientific specialties is not the result of deliberation, at least not in the sense relevant to the phenomenon of groupthink. Nonetheless, the attention to the structure of problem-solving that has emerged in the groupthink research conducted by psychologists can help us see when deliberation could lead to problems for a research team. The analysis of problem-solving developed by Aldag and Fuller provides a useful framework for understanding at which stages in the research process a research team is most prone to groupthink. Whenever we need to generate alternative solutions or proposals, groupthink is a genuine threat, and research teams would be wise to allow individuals opportunities to work alone. The benefits of team work emerge when scientists seek to evaluate the various proposals generated, and determine a course of action. Then the group is less prone to groupthink, and the interaction of group members can be an epistemic asset.