Tuning into whispered frequencies: Harnessing Large Language Models to detect Weak Signals in complex socio-technical systems

Complex socio-technical systems performance varies in daily operations due to endogenous and exogenous disturbances. This variability demands continuous adaptation by individuals and organizations. Stories of variability are usually ignored in case the operation does not have large consequences, either in terms of safety or in terms of productivity. A specific elicitation protocol called Structured Exploration of Complex Adaptation (SECA) has been proposed to foster data gathering about everyday variability as performed by front-line operators. The obtained information, however, remains challenging to be studied systematically by a human analyst, since they suffer from being linguistic, large in volume, and potentially fragmented. This paper leverages SECA via a dedicated approach to applying Grounded Theory by means of Large Language Models (LLMs). The ultimate purpose is to obtain methodological support by Artificial Intelligence (AI) that allows converting the elicited tacit information into useable explicit knowledge, i.e. identifying weak signals, otherwise hidden. A use case in the Air Traffic Management domain is presented to showcase the applicability of the method, as it has been iterated for 50 selected questionnaires.