Conversational agents have the potential to support intergroup relations when psychological or linguistic barriers prevent direct interaction. Based on intergroup contact theory, we propose GroupEnvoy, a conversational agent that represents outgroup perspectives during ingroup discussions, grounded in transcripts from outgroup-only sessions. To evaluate this approach and derive design principles, we conducted a mixed-methods, between-subjects study with university students, where host-country students formed the ingroup and international students formed the outgroup. Ingroup students performed a collaborative task, receiving outgroup perspectives via GroupEnvoy (experimental) or reading written transcripts (control). Compared to the control group, the experimental group showed greater reduction in intergroup anxiety and greater improvement in perspective-taking. Qualitatively, AI-mediated contact enhanced outcome expectancies, whereas passive exposure fostered future contact intentions. The two conditions also elicited empathy toward distinct targets: outgroup evaluations of the ingroup versus outgroup lived experiences. These findings validate AI-mediated contact as a promising paradigm for improving intergroup relations.

翻译：对话代理有潜力在心理或语言障碍阻碍直接互动时支持群际关系。基于群际接触理论，我们提出了GroupEnvoy，一种对话代理，在内群体讨论中代表外群体视角，该视角基于仅外群体的对话记录生成。为评估该方法并推导设计原则，我们进行了一项混合方法的组间研究，以大学生为被试，其中东道国学生构成内群体，国际学生构成外群体。内群体学生执行协作任务，并通过GroupEnvoy（实验组）接收外群体视角或阅读书面记录（对照组）。与对照组相比，实验组在群际焦虑方面表现出更大程度的降低，并在观点采择方面有更大改善。定性分析表明，AI中介接触增强了结果预期，而被动暴露促进了未来接触意图。两种条件还引发了针对不同目标的共情：外群体对内群体的评价与外群体的生活经历。这些发现验证了AI中介接触作为改善群际关系的一种有前景范式。