Using informative behavior to increase engagement in the tamer framework

摘要: In this paper, we address a relatively unexplored aspect of designing agents that learn from human training by investigating how the agent's non-task behavior can elicit feedback higher quality and quantity. We use TAMER framework, which facilitates human-generated reward signals, i.e., judgements actions, as foundation for our investigation. Then, propose two new interfaces to increase active involvement in process thereby improve task performance. One provides information on uncertainty, other its Our results 51-subject user study show these induce trainers train longer give more feedback. The performance, however, increases only response addition performance-oriented information, not sharing uncertainty levels. Subsequent analysis suggests organizational maxim about behavior, "you get what you measure" - metrics with people causes them focus maximizing or minimizing those while de-emphasizing objectives also applies agents, providing powerful guiding principle human-agent interface design general.