Agents Teaching Humans in Reinforcement Learning Tasks

This paper extends our existing teacher-student framework to allow a knowledgeable agent to teach human students. An agent teacher instructs a human student by suggesting actions the student should take as it learns. This paper extends previous algorithms, used for agents teaching other agents, to develop several new algorithms for agents teaching humans. Our results in the Pac-Man domain show that our new approaches can indeed be effectively used to improve human learning. Moreover, some of these human-teaching approaches perform better than some of the original algorithms when one agent teaches another agent.