The Epistemic Benefit of Transient DiversityKevin J.S. Zollman∗Carnegie Mellon UniversitySeptember 29, 2009AbstractThere is growing interest in und ers tanding and eliciting division of labor withingroups of scientists. This paper illustrates the need for this division of labor througha historical example, and a formal model is presented to better analyze s ituations ofthis type. Analysis of this model reveals that a division of labor can be maintained intwo different ways: by limiting information or by endowing the scientists with extremebeliefs. If both features are present however, cognitive diversity is maintained indef-initely, and as a result agents fail to converge to the truth. Beyond the mechanismsfor creating diversity suggested here, this shows that the real epistemic goal is notdiversity but transient diversity.∗The author would like to thank Brian Skyrms, Kyle Stanford, Jeffrey Barrett, Bruce Glymour, SamHillier, Samir Grover, Kevin Kelly, Teddy Seidenfeld, Michael Strevens, Michae l Weisber g, Ryan Muldoon,several contributors at the Wikipedia reference desk, and the anonymous referees for their ass istance. Codefor the simula tio ns can be obtained from the author’s website: http://www.andrew.cmu.edu/users/kzollman/1The Epistemic Benefit of Transient DiversityA striking social feature of science is the extensive division of labor. Not o nly are differentscientists pursuing different problems, but even those working on t he same problem willpursue different solutions to that problem. This diversity is to be applauded because, in manycircumstances one can simply not determine a priori if a general theoretical or methodologicalapproach will succeed without first attempting to apply it, study the effects of its application,and develop additional auxiliary theories to assist in its application.1The value of diversity presents a problem for more traditional approaches to scientificmethodology, since if everyone employs the same standards for induction and ha s access tothe same info r matio n, we ought to expect them to all adopt the theory which at the currenttime looks most pro mising. Kuhn suggests t his problem represents a failure of the traditionalapproach,Before the group accepts [a scientific theory], a new theory has been tested overtime by research of a number of [people], some working within it, others within itsmore traditional rival. Such a mode of development, however, require s a decisionprocess which permits rational men t o disagree, and such disagreement wouldbe barred by the shared algorithm which philosophers have generally sought. Ifit were at hand, all conforming scientists would make the same decision at thesame time (Kuhn, 197 7, 332) .Kuhn’s approach to the problem, allowing diversity in standards for induction, is supportedby others who offer similar solutions (Hull, 1988; Sarkar, 1983; Solomon, 19 92, 20 01).Alternatively, Philip Kitcher (1990; 1993; 2002) and Michael Strevens (2003a; 2003b)have both suggested that homogeneity in inferential strategy can still produce diversity ifthe scientific reward system has an appropriate structure and scientists are appropriately1When defending alternatives to the classic interpretation of quantum mechanics, Feyerabend succinctlydeclares, “It takes time to build a good theory” (1968, 150 ).2motivated. In a similar vein, Paul Thagard (1993) has suggested that a uniform method butdifferentia l access to information can be of some assistance in maintaining this diversity.Just as traditional epistemologists ignored the benefit of diversity, many of these contem-porary champions of diversity ignore the metho d by which ultimate consensus is achieved.2They do however point to an important learning situation faced by scientists. In these sit-uations, information about the effectiveness of a theory or method can only be gathered byscientists actively pursuing it. But scientists also have some interest in pursuing a theorywhich turns out to be right, since effort developing an inferior theory is often regarded as awaste. This is precisely the circumstance described by Kuhn.Rather than focusing on diversity directly, we will consider this type of learning situa-tion as a problem in social epistemology. We will begin this investigation by presenting animportant episode fr om the history of science which illustrates the need for diversity. Afterpresenting the history of the investigation of peptic ulcer disease in Section 1, we will inves-tigate a model which captures some of the central features of this episode in the history ofscience.This model represents one type of learning situation discussed by the contemporarychampions of diversity, and so presents a situation where the benefit of diversity can beexplicitly analyzed. The analysis of this model demonstrates a consistent theme, that acertain amount of diversity provides some benefit to the community. One way to a tt ain thisdiversity is by limiting the amount of information available to the scientists. This is achievedby arranging them so that they only see a proper subset of the total experiments performed,using different subsets for each agent. This preserves the diversity present in the agents’priors which helps facilitate exploration.Diversity can be maintained in the models in a second way, extreme priors. If agents havevery extreme beliefs, this has the same effect as limiting information. But, this solution offers2Feyerabend is perhaps the most extreme in this reg ard. He says, “This plurality of theories must notbe regarded as a preliminary stage of knowledge that will at some time in the future be replaced by the’one true theory.”’ (1965, 149). But even in papers of this e r a (1965; 1968), he does not advocate hold ontoinferior theories indefinitely.3an opportunity to see the downside to diversity. If agent s are both extreme in their initialbeliefs and limited in their access to information, the initial diversity is never abandoned.This shows that in these learning situation diversity is not an indep endent virtue whichought to be maintained at all costs, but instead a derivative one that is only beneficial for ashort time.3Beyond simply maintaining diversity, we also find that in some cases apparently irrationalindividual behavior may paradoxically make communities of individuals more reliable. Inthis case we have shown that the socially optimal structure is one where scientists haveaccess to less information. But, before loo