Human-Agent Interactions: Does Accountability Matter in Interactive Control Automation?

Abstract

In human-automation systems, where high situation awareness is associated with better decision-making, understanding accountability may be crucial to preventing automation complacency. In supervisory control automation, there is some evidence that accountability increases human-automation performance; however, with increasingly intelligent automated agents, human-agent work relationships may resemble more interactive control compared to supervisory control. We investigate the effects of social accountability in a simulated joint task environment and hypothesize that people under an accountability condition would cooperate more with an automated agent than people under a non-accountability condition, in a shared cognitive task. Results from our study support this hypothesis. However, for the accountability group, people’s performance in terms of units processed was lower, and this group also self-reported lower performance and attentional control, with higher frustration. These findings indicate that accountability may slow the decision-making process through added pressure, with some costs to short term efficiency.