Impact of Interaction Design on Human Satisfaction Teaching Reinforcement Learning Agents in Partially Observable Domains

Fri 10/7/2020 - 3:36 pm
Presenter(s): Project:
Download this presentation

Interactive machine learning involves humans interacting with agents during their learning process. As this field grows, it is pertinent that non-expert users are able to have a satisfying experience in teaching the agents in order to retain usage of the agent and minimize frustration for the user. Previous work has investigated which factors contribute to the user’s experience when teaching agents in a fully observable domain. In this paper, we investigate how four different interaction methods affect agent performance and teacher experience in partially observable domains. As the domain in which the agent is learning becomes more complex, it accumulates less reward overall and needs more advice from the teacher. It is found that the most salient features that affect teacher satisfaction are whether or not the agent complies with input, how fast the agent responds, how much overall instruction is needed, and the degree to which the agent’s behavior is probabilistic. It is suggested that machine learning algorithms incorporate a short time delay in the agent’s response and minimize probabilistic actions. The need to generalize advice over time to reduce the amount of instruction needed varies depending on the presence of penalties in the environment.
A video to this presentation can be viewed here.

Map of Cognitive Engineering Center

Cognitive Engineering Center (CEC)
Georgia Institute of Technology
270 Ferst Drive
Atlanta GA 30332-0150