| dc.description.abstract | Long-duration spaceflight requires astronauts to live in isolated, confined, and extreme environments while physically and socially separated from support systems for up to years at a time. This thesis explores how intelligent, autonomous tools may augment behavioral health management when real-time, Earth-based support is inaccessible in deep space. Such intelligent psycho-social support must be able to: 1) characterize individual risks, 2) assess health state accurately, and 3) deliver appropriate interventions or countermeasures. To augment these system functions, techniques from statistical modeling, natural language processing, and conversational AI are investigated across three case studies of isolation: wide-scale isolation during the COVID-19 pandemic, isolation in a space analog habitat, and social isolation in the general public. To explore risk characterization, statistical models were constructed on trait-based, behavioral, and social environment factors in relation to mood and anxiety state in isolation. Models of civilian risk under isolation were developed to inform automated risk characterization for future private astronauts. To explore augmenting psychological assessment, a feasibility analysis of natural language processing (NLP) for automated affect classification was conducted. Transformer-based NLP techniques were tested against lexicon-based and other machine learning (ML)-based techniques on affect classification of personal journal text from analog astronauts. Transformer-based models demonstrated improved detection of negative affect classes, but overall, lexicon-based models were still comparable to ML-based models. Finally, to explore augmenting countermeasures, a study of engagement and disclosure in AI-augmented reflection was conducted. In this study, participants shared more content volume and spent more time in a hybrid reflection condition (reflecting alone first before chatting with a prompted chatbot) than in separated conditions (journaling alone or reflecting only with the bot). Higher loneliness was associated with lower comfort and engagement, but behavioral benefits were similar across lonely and non-lonely users, suggesting further tailoring is needed to support isolated individuals. Through these case studies, this work presents a point of departure towards a vision of intelligent psychosocial support systems that are not meant to replace human connection, but to augment behavioral healthcare for individuals in isolated settings on Earth and in space. | |
