Case study of Work-As-Large Language FRAM models for safety-oriented analysis

The case study is imagined but plausible and concerns a hazardous chemical spill in the chemical laboratory of a hospital. A newly hired chemist at the hospital accidentally knocks over a bottle containing a hazardous chemical, causing it to spill onto the floor. Although the chemist had undergone training on how to handle emergency situations, in this real circumstance, he is unsure about the correct procedure to follow. In an effort to seek immediate guidance, he decides to request assistance from a large language model chatbot. However, the chatbot provides a slightly different procedure than what should have been followed, leading to confusion in incident management. Following FRAM, models of the Work-as-Prescribed (what a normative prescribes to handle the emergency above) and of the Work-as-Large Language model (what a chatbot suggests to the chemist) are provided. These can be used to analyze the differences in the two perspectives on the same process to the purpose of resilience analysis.本案例研究为虚构但符合现实逻辑的场景，主题为医院化学实验室发生危险化学品泄漏事故。该院新入职的化学技术员不慎碰倒一瓶危险化学品，致使化学品洒落至实验室地面。尽管该技术员曾接受过应急处置相关培训，但在此次实际场景中，他却不确定应遵循的正确操作流程。为寻求即时指导，他决定向大语言模型（Large Language Model，LLM）聊天机器人求助。然而，该聊天机器人给出的操作流程与标准处置流程存在细微差异，导致事故处置过程中出现混乱。 本研究遵循功能共振分析方法（FRAM），构建了两类模型：一类为规定作业流程（Work-as-Prescribed）模型，即规范文件中针对上述应急场景的标准处置流程；另一类为大语言模型生成作业流程（Work-as-Large Language Model）模型，即聊天机器人向化学技术员提供的处置建议。上述两类模型可用于剖析同一流程的两种认知视角间的差异，以开展韧性分析相关研究。