By Rob Mitchum // February 26, 2013
The march of science is stumbling and easily sidetracked, fraught with bias, fads and dead ends. A new research initiative based at the University of Chicago and the Computation Institute will use the latest computational tools to scrutinize this imperfect path and better understand how knowledge was and is created. Such understanding could transform the process of research, calling out past missteps while revealing unanticipated new directions for the future.
With a $5.2 million grant from the John Templeton Foundation, the new Metaknowledge Network brings together social scientists, computer scientists and domain experts from several disciplines to explore how knowledge emerges, thrives, evolves and dies out. The lessons learned can be used to accelerate discovery across fields, as scientists develop a deeper understanding of why we have the knowledge we have—and why certain promising questions were left unasked or unanswered.
“We have an opportunity to create a really rich science of science, one that builds on novel computational tools to exploit the increasingly widespread digital traces of the research process,” said James Evans, director of the center, associate professor of sociology, and Computation Institute Fellow.
Metaknowledge means “knowledge about knowledge”—the study of how different scientific questions and ideas appear, mature and potentially take root. The idealistic view of research is that it proceeds in unbiased, empirical steps. But scientists faced with an almost limitless number of potential questions may also choose a research path based on non-empirical factors such as available resources or equipment, professional and educational networks, access to previous findings and the biases and trends of their field.
Until now, the fingerprints of these external influences have been difficult to detect with the naked eye.
“Most of what we know comes from putting science under the microscope, from deep historical or ethnographic study,” said Jacob Foster, assistant professor of sociology and a member of the Metaknowledge Network.
But the current explosion of digitally available text, including journal publications, books, patents and news articles, makes it possible for the first time in history to study the dynamics that shape scientific research at scale, as the latest computational tools can capture some of the richness of these insights.
“The central idea is, how can we take these huge data resources associated with science today—all the publications, preprints and data that are floating around—and use that to figure out why people ask the questions that they ask?” Evans said. “And how can this knowledge lead us to ask better questions?”
Initial core projects within the Metaknowledge Network will examine why some theories are more popular than others, what strategies are most likely to produce groundbreaking research and how deeply held notions and assumptions shape what scientists study.
For example, preliminary work by network collaborators found that award-winning scientists were more likely to attempt riskier projects, which were subsequently cited more often by their peers. Other work identified biases in the production and analysis of data from many fields, and demonstrated how previous findings reported in the biomedical literature can distort how scientists interpret their own independent results, a kind of publication peer pressure.
In a time of tight research budgets, these insights can help direct funding to the most promising scholars and projects, rather than merely the most popular or prominent. A combination of text-mining and machine-learning tools also can potentially provide a continuously refreshed “sentinel” on a body of research, updating as new articles are published and suggesting new high-impact hypotheses to be explored.
“For science policy and the business of innovation, we could be much more rational about how we search through the enormous space of questions,” Evans said. “And that includes being more creative—asking new, unexpected questions.”
To support the network’s research, a Knowledge Lab will be established at the Computation Institute on the UChicago campus for researchers to collaborate with computer programmers. The network also will build an online Metaknowledge Portal for sharing software, data and publications. Databases and results generated by the projects will be made open source and available to the public where legally possible.
The inaugural members of the Metaknowledge Network include researchers from the University of Chicago, Argonne National Laboratory, Stanford University, Northwestern University, the University of California, the University of Washington, the University of Wisconsin, Princeton University and Harvard University.