Actively Adaptive Agent for Human-Agent Collaborative Task

Abstract

Active interface is one of critical characteristics of agents who have to interact with human users to achieve human-agent collaboration. This characteristic is especially important in beginning phase of human-agent interaction when an ordinary human user begins to interact with an adaptive autonomous agent. In order to investigate principal characteristics of an active interface, we developed a human-agent collaborative experimental environment named WAITER. Two types of experiment: WOZ agent experiment and autonomous agent experiment were conducted. Objective of the experiment is to observe how human users change their instructions when interacting with adaptive agents with different degree of freedom. Experimental results indicate that participants can recognize changes of agent’s actions and change their instruction methods accordingly. It infers that changes of instruction method depend not only on waiter agent’s reactions, but also on human manager’s cognitive models of the agent.