Beyond the hype: The nuances of using Natural Language Processing to uncover pilot adaptations in aviation

Human adaptation to unexpected situations is critical in modern socio-technical systems. Aviation illustrates this need, where controllers’ and pilots’ adaptations are both normal and necessary especially when considering challenges such as single-pilot operations or airspace congestion. Leveraging Artificial Intelligence, this chapter investigates adaptations as reported by aviation pilots via open-ended surveys. Starting from a large corpus of data collected by the European Cockpit Association (ECA), Latent Dirichlet Allocation (LDA) and k-means have been tested with the aim of identifying underlying patterns in surveyed data. While some common themes can be retrieved by AI, they remain at an abstract level, and they still demand for more sophisticated algorithms to be run on structured surveys, or more traditional thematic analysis, possibly leveraging human automation teaming also in the analysis of the data themselves.