Leveraging crowdsourcing to collect data is becoming more common. Human Computation, in particular, has looked at how to use artificial intelligence on data collected from people playing games, to validate that useful data has been collected on a very large scale. This work will investigate a new form of artificial-intelligence based crowdsourced games called Computational Gaming, in which questions will be posed without knowing what the answers are beforehand. Questions that require human judgment will be posed in the context of a game, and machine learning will be used to determine what questions to pose to which players and how to determine whether the responses are valid.
Intellectual Merit. This project will demonstrate the validity of Computational Gaming through two examples in text and image labeling, delineating a set of guiding design principles for building and evaluating future Computational Gaming designs, and producing a toolkit that supports and encourages the use of these design principles for building Computational Gaming systems.
Potential Broader Impacts. The project will create, both more quickly and more cheaply, databases of human-labeled data; it will also do so for a wider variety of problems than currently exists. The framework and toolkit for Computational Gaming will be valuable for game designers, for researchers in many domains that need labeled data, and for the users for whom the research is being conducted.
