Would you trust your AI teammate in wargame competitions? The impact of levels of automation, human control, and reward uncertainty on trust in agents

The widespread use of unmanned weapons in modern warfare highlights trust in collaborative agents in war competitions to achieve collaboration efficiency. Previous research has generally indicated that agent-, human-, and environment-related factors are interrelated and influence human−agent trust relationships. Therefore, the present study examined the impact of levels of automation, human control, and reward uncertainty on trust in agents, as well as the mechanism of sense of control in wargame competitions through two experiments. Participants collaborated with an agent to fight against another artificial intelligence (AI) agent while their eye movement behaviors were recorded. Results revealed that low-automation agents were more trustworthy than high-automation agents when individuals maintained high control. For individuals with high control, high-reward-uncertainty situations reduced trust in high-automation agents. A stronger sense of control over low-automation agents than high-automation ones further led to the trust differences between the two types of agents, and this effect was stronger in higher-reward-uncertainty situations. These findings might help address theoretical gaps in the understanding of automation−trust dynamics, that is, how human, agent, and environmental factors collectively shape human−agent trust in competitive environments. Practically, they may offer valuable insights for optimizing human−agent collaboration systems, training protocols, and operational guidelines to enhance human−agent collaborative efficiency.