AI-MIX: Using Automated Planners to Steer Human Workers Towards Better Crowdsourced Plans

One subclass of human computation applications are those directed at tasks that involve planning (e.g. tour planning) and scheduling (e.g. conference scheduling). Existing literature on these systems show that even primitive automated oversight on human contributors help in significantly improving the effectiveness of humans/crowd. In this paper, we explain how the automated oversight used in such systems can be viewed as a primitive automated planner, and outline several opportunities for more sophisticated automated planning in effectively steering crowdsourced planning. Adapting the current planning technology requires overcoming the mismatch between the capabilities of human workers and automated planners. To overcome this mismatch, two important challenges are to be addressed. They are (i) interpreting the inputs of human workers and (ii) steering or critiquing the plans produced by the human workers provided the incomplete domain and preference models. We discuss these challenges and describe our initial attempt at tackling them. Specifically, we describe our ongoing work AI-MIX, a tourplan generation system that uses automated checks and alerts to help improve the quality of plans created by human workers.