Simulating Emergency Teamwork Behavior with Multi-Agent LLMs: Reproducing Human Training Patterns Using Persona-Based Agents

团队合作 培训(气象学) 计算机科学 工作(物理) 应急响应 人工智能 知识管理 工程类 过程管理 人机交互 过程(计算) 动作(物理) 模拟 自动化 人机交互 钥匙(锁)
作者
Atsushi Shimoda,Rei Kobori,Keito Takahashi,Zhang Yatong
出处
期刊: [Institute of Electrical and Electronics Engineers]
卷期号:: 201-208 被引量:1
标识
DOI:10.1109/icmsci67830.2026.11469339
摘要

Emergency response teams such as Computer Security Incident Response Teams (CSIRTs) must operate under intense time pressure, rapidly evolving threats, and incomplete information. Team effectiveness in these environments depends strongly on the interaction among individual skill proficiency, leadership style, and collective coordination. Traditional human emergency training exercises are costly, difficult to repeat under controlled conditions, and limited in the diversity of team configurations that can be systematically examined. Recent advances in large language models (LLMs) and multi-agent conversational frameworks raise a new question: Can LLM-driven multi-agent simulations reproduce human teamwork behavior observed in real emergency training? This study investigates the extent to which persona-based LLM agents can replicate behavioral patterns documented in a publicly available emergency response training dataset. We construct triad teams of LLM agents, assign personas reflecting human skill and leadership profiles, and expose them to an incident scenario parallel to that used in the human exercises. Each conversation log is evaluated using the same 11-item teamwork behavior rubric applied in the human study. Experimental results show that approximately 60% of the behavioral tendencies observed in human teams-such as skillbalanced coordination or follower-dominant communication— are reproduced by the LLM multi-agent simulations. These findings provide the first empirical evidence that LLM-based multi-agent simulations can partially approximate human teamwork behavior in emergency contexts. The proposed approach offers a low-cost and reproducible framework for analyzing team dynamics, exploring hypothetical team structures, and supporting digital-twin-based training tools for CSIRT and other emergency response organizations.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
学术雷锋发布了新的文献求助10
刚刚
1秒前
Juvenilesy的应助被HONGZ采纳,获得10
1秒前
1秒前
xing发布了新的文献求助10
1秒前
Yuenna发布了新的文献求助10
1秒前
2秒前
wanci的应助被kk采纳,获得10
2秒前
2秒前
DOC_XIONG的应助被寒冷书兰采纳,获得10
2秒前
拉拉啊了完成签到,获得积分10
2秒前
3秒前
3秒前
罗鑫发布了新的文献求助10
4秒前
小马甲的应助被shanguren采纳,获得10
4秒前
江边拔草完成签到,获得积分10
4秒前
曹牛牛发布了新的文献求助10
6秒前
完美世界的应助被咩了个咩采纳,获得10
6秒前
可耐的稀发布了新的文献求助10
6秒前
6秒前
ddddd完成签到,获得积分10
6秒前
后悔体验生活的北极兔完成签到,获得积分20
6秒前
lin发布了新的文献求助10
6秒前
6秒前
molihuakai的应助被舒适念真采纳,获得10
7秒前
嘻嘻哈哈发布了新的文献求助10
7秒前
烟花的应助被Yuenna采纳,获得10
7秒前
7秒前
7秒前
拉拉啊了发布了新的文献求助10
8秒前
顾矜的应助被抗体小王采纳,获得10
8秒前
顾矜的应助被赵晋杰采纳,获得10
8秒前
Qq发布了新的文献求助10
8秒前
Alex完成签到,获得积分10
8秒前
见见完成签到,获得积分10
9秒前
KKH发布了新的文献求助10
9秒前
xiaogu1013完成签到,获得积分10
10秒前
寒冷书兰完成签到,获得积分10
10秒前
酷酷向真完成签到,获得积分10
10秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1500
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
Analyses de reconnaissances de dettes passées devant les échevins d'Ypres (1249-1291) 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8161320
求助须知:如何正确求助?哪些是违规求助? 9556074
关于积分的说明 21732157
捐赠科研通 7751456
什么是DOI,文献DOI怎么找? 3427537
关于科研通互助平台的介绍 2552583
邀请新用户注册赠送积分活动 2471992