Addressing the role of context on trust in human-AI teams: the influence of team role and violation type in high-risk tasks
通过军事高风险任务实验，发现AI伦理违规的类型（诚信vs能力）和人类角色（依赖程度）显著影响信任与伦理感知，但信任修复策略效果不显著。
The current paper reports on an experiment examining how contextual factors influence trust, perceived ethicality, and performance in human-AI teams undertaking a high-risk, action-based task within a military setting. The study examined the impact of team role and trust violation framing on trust, perceived ethicality, and the efficacy of four trust repair strategies when an AI teammate commits an unethical action. Results indicated that trust and perceived ethicality of the AI team member were significantly higher when ethical violations were framed as integrity-based violations rather than competency-based violations. Additionally, those in the Ground role, who relied more on the AI for their safety, also had higher trust and ethicality ratings for the AI. However, trust repair strategies did not significantly impact trust in the AI team member after an ethical violation. These results highlight the significance of context in determining trust in response to AI ethical violations.