Crowd Architectures     


Elizabeth M. Daly photoWendy Kellogg photo

Crowd Architectures - overview

Four types of crowds.

In the beginning – back in the late '90's – the Social Computing Group focused on online conversation among small to medium sized groups, most notably with its Babble and Loops projects. In the ensuing years, it has become possible for increasingly large numbers of people to not only talk but to collectively produce coherent products and events. We are interested in studying systems that succeed at this, understanding the many possible ways that the structure of crowds affords coherent work, and in using those findings to advance our ability to design new systems.

This project intersects with our work on Social Infrastructures for Smarter Cities.


Mobile Crowdsensing. Crowdsensing is the enlistment of large numbers of ordinary people in gathering data supports societal goals such as ameliorating pollution, managing public resources or planning development. This line of work began with an OCR (for Open Collaborative Research) project involving IBM Research, University of Illinois, and University of Minnesota. The work ranged from the development of a middleware platform for mobile devices to prototyping end user crowdsensing applications. The application level continues as a collaboration between IBM Research and the University of Minnesota with a focus on designing a platform called Citizen Sense that allows individuals to propose, design, and manage distributed crowdsensing campaigns. Work on this is ongoing.

Crowd Games. How can we use games to tap the power of the crowd? In this project, we are designing games to elicit natural language semantics, especially with respect to words that are used in metaphorical or idiomatic ways. For example, if a sentence reads "The bill died in the House" does that mean that the bill was not passed or that a funeral will be held? And under what circumstances does the mapping of "died" to "not passed" hold true? This requires complex human judgement. Using crowdsourcing to unpack semantic issues like this can aid in developing systems that better understand natural language.

Cyclopath. This is not our project, and not supported by IBM, but we are big fans and one of us spends some time working with this cutting edge research group at the University of Minnesota. Cyclopath enables people to find bicycle-friendly routes around the city. What makes Cyclopath interesting is that data sources it draws on – road maps, bike route maps, and topographic data – lack much of the information required to compute a bicycle-friendly route: Is the pavement smooth? Are there a lot of parked cars along the curb? Is it scenic? Does it flood when it rains? Cyclopath relies on its users to add data and rate routes, and this in turn means that to function Cyclopath has to do ‘social work’: it has to attract users, engage them so that they will contribute data, and provide mechanisms (its a wikified map) for users to evaluate one anothers contributions. Cyclopath is in its third year of use in Minneapolis-St Paul, and the information its users have added have demonstrably improved its ability to find routes. Cyclopath, besides providing a valuable service, serves as a platform for further research and design. You can step through a demo of cyclopath at, or see Cyclopath itself at

Avatar-mediated Interaction in 3D Worlds. While the Social Computing Group's work on Second Life is in abeyance, this genre of interaction continues to hold our interest as a way that provides unique affordances for synchronous interaction among large groups. Of most relevance here is our study of a 3-day conference of 500+ people held entirely in Second Life, and our observations of and framework for thinking about synchronous interaction within crowds: Synchronous Interaction Among Hundreds: An Evaluation of a Conference in an Avatar-Based Virtual World). Related work, albeit at the scale of groups rather than crowds, explores creating games for team building (See: Games for Virtual Team Building). While much of the world has lost interest in 3D Worlds, we believe that advances in fluid interaction, wearable sensors, and a generation or two of Moore's law, may lead to another wave of interest.

Foundational work. Threading through the various lines of work described here are efforts to develop conceptual frameworks for understanding various facets of crowd architecture. These include work on incentive mechanisms carried out through a workshop at Group 2005, and Some Thoughts on a Framework for Crowdsourcing prepared for a CHI 2010 workshop, and a discussion of the foundations of systems like, the ESP Game and Wikipedia in an article on Social Computing for the Encyclopedia published in 2011.