The PI will use computational models of the social structure of science to study several issues associated with the division of cognitive labor. Those issues include the following. How do individual scientists allocate their cognitive labor given research costs and limited access to information? How does drawing on the cognitive resources of the scientific community enhance or retard the discovery of significant truths? How should the scientific community as a whole optimally allocate its cognitive labor?

The notion of the division of cognitive labor may be characterized as follows. Scientists are not lone agents, cut off from the outside world, responding only to information generated in their laboratories. Rather, they make decisions about what to investigate by integrating what they discover for themselves with what they learn from others. They also take into account external factors such as grants, prizes, and prestige. This feedback leads scientists to divide their resources among differing approaches to studying phenomena of interest. While this coordination is neither planned nor explicit, it seems to enhance the ability of scientific communities to discover significant, true things about the world.

Although the results of this research project will primarily contribute to the philosophy of science literature, it has three major areas of potential broader impact. First, it will involve training undergraduate students in the use of computational modeling to investigate philosophical problems. Second, while the models developed in this project are not appropriate for making direct policy advice, they can investigate sets of incentives that would change the division of cognitive labor in more epistemically productive ways. Finally, the project will generate materials that can be reformulated for use in secondary and higher education with the goal of making citizens more informed consumers of science, especially in fields such as evolution and climate modeling where critics prey on public misunderstanding of how theories are tested.

Project Report

Scientists are not lone agents, cut off from the outside world, responding only to information generated in their own laboratories. Rather, they make decisions about what to investigate by integrating what they discover for themselves with what they learn from others. They also take into account external factors such as grants, prizes, and prestige. These sources of feedback lead scientists to coordinate and divide their resources among differing approaches to the research domain. This coordination seems to enhance the success of scientific communities, but this coordination is neither planned nor explicit. Philosophers have called this phenomenon the division of cognitive labor. Although the division of cognitive labor is a widely recognized phenomenon, relatively little is known about the connection between these cognitive and social facts about science and the research productivity of scientific communities. More specifically, we know little about different divisions of cognitive labor and how effective these divisions are for achieving scientific goals. We also do not know how restrictions of information and resources affect the division of labor, especially what kinds of incentives might allow the scientific community to achieve more productive divisions of cognitive labor. This project was a sustained investigation of these questions through the use of computer simulations. These simulations represented scientists as agents exploring an "epistemic landscape," which is a highly simplified representation of potential approaches for investigating a research question. To begin answering the questions posed above, I investigated two extreme versions of divisions of cognitive labor: mavericks, who try to avoid doing what others are doing, and followers, who always try to imitate the most successful approaches. An especially surprising result of this research has to do with relative performance of mavericks and followers. It would seem that scientific communities composed of followers would be effective researchers — they could build on what each other are doing, using their collective knowledge to advance research quickly. Mavericks, on the other hand would seem to make poor researchers, because each scientist would have to potentially "reinvent the wheel." This, however, is not what the simulations show. Mavericks vastly outperform followers because followers easily fall into herding behavior. Their central tendency is to follow each other, not to branch out into new horizons. These initial results suggested that a challenge for scientific communities, and those that incentivize communities, is to find ways to break this herding behavior. Many of the models investigated were aimed at finding such incentives, at least in the abstract. Two effective strategies for improving follower behavior are adding some number of mavericks into the community, and allowing the followers to be led, in part, by these mavericks. An even more effective strategy is to allow followers to take cost-free risks. Although herding behavior remains, these cost-free risks can break some of the epistemic traps that followers create for themselves. Future work in this area will aim at making the models more realistic, drawing on sociological and citation data. With more realistic models in hand, one might rethink the norms governing research and the policies governing the funding of research, with an eye to making more effective knowledge generating communities.

Agency
National Science Foundation (NSF)
Institute
Division of Social and Economic Sciences (SES)
Application #
0957189
Program Officer
Frederick M Kronz
Project Start
Project End
Budget Start
2010-07-01
Budget End
2013-06-30
Support Year
Fiscal Year
2009
Total Cost
$155,861
Indirect Cost
Name
University of Pennsylvania
Department
Type
DUNS #
City
Philadelphia
State
PA
Country
United States
Zip Code
19104