Concepts of collective intelligence or the so-called "wisdom of crowds" play increasingly fundamental roles in economic and political forecasting, as well as in addressing public policy disputes and panel decisions about creative merit. Diversity of information and expertise among crowd members has been identified as a key condition for the emergence of collective decisions that are superior to any individual’s personal judgment. However, diversity is hard to attain: there are systemic difficulties in mobilizing minority groups on top of which additional factors related to social influence such as herding, groupthink, and conformity tend to further reduce the variety of point of views. This project aims to design network-aware machine learning tools that elicit useful diversity, counter herding and homophily effects in restraining the wisdom of crowds, and improve the accuracy of collective forecasting. The proposed research will train graduate and involve undergraduate students in state-of-the-art data science as they develop and deploy novel research algorithms and software for effectively harnessing collective intelligence at the intersection of modern computational and sociological research. The work will ultimately deliver new technical tools that optimize the wisdom of networked crowds, helping designers of online systems and policy makers leverage the benefits of collective intelligence. The project aims to develop theoretical and analytical models as well as algorithmic tools for diversity evaluation and collective intelligence enhancement under social influence. In specific, the team will build 1) a model based on the exponential random graph formalization to evaluate the role of homophily in restraining diversity and 2) a herding model using observational learning that disentangles factors that lead towards lowest common denominator information in a crowd. A better understanding of these two effects will inform feature design for a comprehensive machine learning framework that predicts crowd performance, helps identify robust collective intelligence signals, and aids the development of opinion aggregation mechanisms that efficiently capitalize on diversity. These research goals are related to the PI’s prior work on predicting outcomes in online crowdlending and -investment systems. The planned work places a specific focus on developments that will make collective intelligence detection tools practical by providing early warnings of collective misconceptions. With its theoretical, algorithmic, and software components, the proposed research will produce deliverable assets that are intended to benefit computer scientists, as well as the broader community of researchers and practitioners in the social and economic sciences who are interested in harnessing the wisdom of crowds in web-based systems. Research outputs will be accessible via the project web site (http://). The tools and results will also be integrated into courses on complex networks and data science at the PI’s institution, benefiting students from a variety of research groups and departments. Due to the project’s focus on diversity and its explicit component about gender-related differences, the proposed work is likely to attract students from underrepresented groups. The grant will allow the PI to provide research opportunities to these undergraduates.
|Effective start/end date||3/15/18 → 2/28/21|
- National Science Foundation (IIS-1755873)
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.