Exploring AI Teammates’ Initiative in Supporting Teamwork Processes in Human-AI Teams

Abstract

This study explores the impact of designing AI teammates to support teamwork processes in Human-AI Teams (HATs), comparing reactive and proactive support’s influence on teamwork experiences. HATs achieve shared objectives through taskwork and teamwork processes. Teamwork processes between humans are well documented, such as goal setting, plan development, and coordination. However, it is less clear how AI teammates can be designed to support these processes in HATs. Furthermore, more attention is needed on how initiative levels of the AI teamwork support would impact the HATs dynamics. Two levels of initiative of a Wizard of Oz AI teammate (RoboAnt) were compared in a gamified testbed HiveMind. Reactive RoboAnt responded to teamwork requests when prompted. Proactive RoboAnt initiated and engaged in teamwork processes. The findings show significantly higher counts of teamwork processes with proactive AI across preparation, execution, and evaluation phases and better team performance.