Abstract
As human-computer interfaces become more sophisticated, people expect computational agents to behave more like humans. However, humans interacting make assumptions about mutual conceptual understanding that they may not make when interacting with a computational agent, where spatial cues in the environment affect their assumptions about the agent’s knowledge. In this paper, we examine an interaction between human subjects and a virtual embodied avatar displayed on a screen, wherein a surface displayed on the screen is either “continued” in the real world by a physical surface or not. Subjects are, with minimal instruction, asked to indicate objects displayed in the shared environment to the agent in the course of a collaborative task. We then examine the subjects’ adaptations, in aggregate, to the different configurations.