By Rob Mitchum // February 13, 2017
What does the field of science look like? Is there a metaphor that can accurately describe millions of scientists in hundreds of countries, simultaneously collaborating, competing, and crawling towards new discoveries? In his talk at the Computation Institute on February 3rd, UCLA’s Jacob Foster proposed one humble comparison — the complex communities of ants.
Ant colonies are complex systems that coordinate large-scale activity through simple chemical signals, allowing thousands or millions of the tiny creatures to act as a kind of super-organism. Science can sometimes resemble this system, said Foster, and understanding how the field works as a whole towards discoveries — and how it could do much better — was the subject of his research and talk, part of the Computational Social Science and Public Policy Colloquium.
Foster, a former Computation Institute researcher and frequent collaborator with Knowledge Lab, proposes a theory of science as “the social production of collective intelligence” — a complex system designed to generate discovery. Foster used text mining and network theory techniques on millions of papers to explore how this system performs on three major challenges for collective intelligence: organizing attention, not getting stuck, and communicating efficiently. The answers can guide scientists and policymakers on how to encourage and speed up important discoveries.
Much of science can be boiled down to the tension between tradition and innovation — the pressure to stay with the pack versus striking out into riskier territory. To examine this balance, Foster and CI Senior Fellow James Evans built a network of “possible wanderings” between roughly 181,000 chemicals, with links representing when two chemicals were studied in the same paper. Studying the network revealed that 85 percent of studies make conservative connections, with only the small remainder trying riskier, long-distance combinations.
Furthermore, these numbers have remained stable despite a huge explosion in the number of available chemicals over the last two decades, suggesting a strong preference for safety over risk. However, ambitious studies were more likely to gain citations and prizes, creating incentives for scientists to break from the pack. Scientific institutions could encourage more of this innovative behavior, Foster said, by funding individuals instead of projects and reducing the role of productivity in job security.
Other sections of the talk looked at whether science currently uses the most efficient methods for exploring the possible space of experiments (it doesn’t) and whether the publication process is the most efficient form of communication between scientists and fields (nope). Foster also described research with Evans on the “map of science” created by studying the use of jargon across different scientific disciplines.
To apply these insights to improving the pace of discovery in science, Foster encouraged increased publication of negative results, to prevent wasted time on experiments that have already been tried and found unsuccessful, and the use of less assumed knowledge and jargon in scientific papers to facilitate interdisciplinary research. But these strategies could have potential unintended consequences, such as false negatives that misdirect the progress of a given scientific pursuit or a dilution of the diversity of ideas present in modern science, isolated as some fields may be.
In order to further explore some of these ideas, Foster has moved into game theory experiments to study how different incentive systems might motivate scientists to participate or avoid a given scientific question. In a draft paper called “Why Scientists Chase Big Problems,” Foster and collaborators Carl Bergstrom and Yangbo Song compred how academic, industrial, and open science motivate — or impede — discovery. Their model provided new ideas about how to organize science for particular situations, such as the response to a disease outbreak.
“I think one of the most interesting things is the scientific response to crises,” Foster said. “The way we deal with crises currently is not a good idea; having a funding competiton to study ebola when the crisis is unfolding is not a good allocation of people’s efforts. There are other ways to think about that, within the framework of how people respond to incentives, such as having an established, funded backup group of people who, when something happens, they’re already funded and ready to go. That’s the kind of way we should be thinking about crises.”
The Computational Social Science and Public Policy Colloquium, co-organized by the Knowledge Lab and the Center for Data Science and Public Policy, is a monthly series featuring exciting work at the intersection of these two fields. For information on future talks in the series, visit the DSaPP events page.