Can an AI agent lead human teams?

Abstract

The use of artificial intelligent (AI) agents to assist or even augment humans performing a task has been an area of research in the field of AI. Most recently, there has been growing interest in developing assistive agents for team applications using Large Language Models (LLMs). However, little seems to be known about whether humans would want to be led by a conversational agent, and what impact such agents might have on team performance and communication dynamics. Here we report the findings which examine this question using a multiplayer video game involving a collaborative search task, where teams of players were led by either a human drone operator or one of two AI drone operators: a simple heuristic (rule) based model and one based on a fine-tuned large language model. While teams led by either AI agent did not perform as well as those led by a human drone operator overall, subtleties in team performance measures observed. Specifically, teams led by the LLM operator had comparable containment performance as those led by the human operator, yet were unsuccessful in completing trials in the presence of fog. Taken together, our findings suggest that AI agents hold the potential of leading human teams engaged in a collaborative tasks and provide guidelines for the development of future collaborative conversational agents adopting team leadership roles.