Methods for Large Scale Distributed Decision Making
Abstract
Methods for Large-Scale Distributed Decision-Making Sharad Goel The last decade has given rise to a global information environment, characterized in part by fast, reliable, and widely available communication devices. This relatively new ability to communicate across physical and social boundaries has opened the door to large-scale, distributed collaboration and information integration. For example, crowdsourcing has proven to be an effective approach for collectively solving diverse tasks, such as answering technical computing questions on Stack Overflow, and compiling an encyclopedia of human knowledge on Wikipedia. Yet, perhaps surprisingly, we have little understanding of how such processes work. I propose to develop and assess methods for enabling large, distributed groups of individuals to make collective decisions. Below, I describe three projects that address the theory and application of distributive decision-making. Simple aggregation of individual opinions has been shown to be a surprisingly powerful technique for prediction, inference, and decision-making. Over the last century this “wisdom of crowds” effect has been observed in a variety of settings, ranging from answering general knowledge questions to predicting sports outcomes. However, given the diversity of experimental designs, subject pools, and analytic methods employed, it is difficult to compare studies and extract general principles. It is thus unclear whether these documented examples are representative of a larger collection of tasks that exhibit a wisdom of crowds phenomenon, or conversely, whether they are highly specific instances of an interesting, though ultimately limited occurrence. By designing and conducting a large-scale experiment, my objective with this first project is to determine which types of problems exhibit a wisdom of crowds effect, how best to aggregate individual responses, and how the size and composition of the crowd affects performance. One limitation of aggregating independent judgments as described above is that individuals cannot interact and learn from one another. Consequently, there is no opportunity for participants to refine their assessments based on input from others. Free-form discussion, however, can be difficult to manage, especially at scale, and reaching consensus in diverse groups is difficult. Thus, in the second project, I plan to investigate methods for structured discussion and decision-making, a middle ground between no interaction and interaction without constraint. Finally, I propose to apply ideas from distributed collaboration and decision-making to develop a prototype system for online education based on crowd-contributed content. By aggregating diverse explanations, viewpoints, and teaching styles, I envision the platform would cater to each learner’s background, preferences, and needs. By enlisting an international group of contributors, I seek to promote greater global access to online education, bypassing existing language and cultural barriers. My hope is to lay the foundation for an engaging and responsive platform that motivates members of the community to share their knowledge and help one another learn. More broadly, I view this effort as a testbed for novel methods for large-scale information integration. In the years ahead, distributed decision-making will almost certainly become more prevalent, with applications in a broad range of areas spanning government, industry, and academia. I believe the collection of projects I have outlined above help provide both a theoretical and empirical foundation for such work.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512711
Entities
People
- Sharad Goel
Organizations
- Office of Naval Research
- Stanford University
- United States Navy