A NLP Approach to Quantify Resilience in Cyber-Socio-Technical Systems with LLM Agents

Incorporating cyber artifacts into Cyber-Socio-Technical Systems (CSTSs) poses new challenges to human safety due to their increasingly unpredictable behavior. Resilience is the ability to adapt, recover, and bounce back from challenges, setbacks, or adversity. Quantifying the resilience of CSTSs requires the identification of leading or lagging indicators. Among the leading indicators, allostatic load measures the level of systemic tension accumulated from misalignments in the perspectives of system actors regarding how work should be performed. In this paper, we propose a novel approach based on Natural Language Processing (NLP) to measure allostatic load. This approach involves lightweight modelling of process perspectives, extraction of token vectors from process function descriptions, and computing vector similarity by using the Dice similarity algorithm. Then, allostatic load is defined as the complement to one of the similarity value. An example application concerning a chemical spill in a hospital laboratory demonstrates the method's practical use.