Login
Computer Science @ Rochester
Wednesday, April 17, 2013
10:00 AM
CSB601
Ph.D. Thesis Proposal
Walter Lasecki
University of Rochester
Crowd Agents Interactive Collective Intelligence
Crowdsourcing has been shown to be an effective means of solving problems that automated approaches cannot yet handle. Crowdsourcing leverages human computation in the form of large groups of non-expert workers of varying reliability, each contributing a small unit of work towards a larger goal. Many interesting tasks such as image classification, document editing, and visual question answering have been performed by the crowd. However, nearly all existing crowdsourcing systems rely on workers completing work in an offline, batch processing fashion, limiting most uses of crowdsourcing to large-scale projects that aim to collect results for use at a later point. The introduction of ubiquitous access to the web has presented the opportunity to recruit workers from online marketplaces within seconds. This fundamentally changes the nature of work, and can be used to complete tasks that human computation would otherwise not be practical for, such as getting on-demand answers to visual questions.

This thesis introduces two new models of crowdsourcing that aim to elicit responses from the crowd as if they were a human assistant. The first is continuous crowdsourcing, in which groups of workers are recruited in real-time and remain engaged in a task for longer periods of time in order to complete continuous tasks. Building on this model, we introduce the crowd agents model, which leverages multiple crowd workers simultaneously in order to complete tasks as if they were a single, well-trained, worker. This is made possible by intelligent mediation of the input of each worker, and allows the crowd to maintain consistent interactions with users, perform difficult psycho-motor tasks in real-time, and even exercise autonomy when asked to complete a complex task. In fact, these agents can not only support users via collective intelligence in the short term, but can also be used in unison with artificial intelligence to provide responses quicker and cheaper, helping create robust intelligent systems.