Adaptive AI : Personalization of a Patient-Centered Companion for the Emergency Department Waiting Area

Sammanfattning

This thesis investigates how a personality-adaptive conversational AI companion can be designed and evaluated for use in Emergency Department (ED) waiting rooms. The aim was to explore whether personality adaptation influences perceived reassurance and usability, and to generate design-relevant knowledge for patient-centred conversational AI in emergency care contexts. A personality-adaptive AI companion was developed using a Large Language Model (LLM) and personality adaptation based on the Five-Factor Model (FFM). The system was evaluated through a between-groups user study involving ten participants, comparing an adaptive and a non-adaptive version of the companion. Quantitative measures of reassurance and usability were combined with qualitative interviews. The results showed that the non-adaptive version was perceived as significantly more reassuring than the adaptive version, while no statistically significant difference in usability was observed. Participants associated reassurance primarily with clear explanations, structured guidance, and professional communication. Although the adaptive version produced more conversational and personalised interactions, it was also perceived as more verbose and cognitively demanding. The findings suggest that personality adaptation alone does not necessarily improve user experience in emergency care settings. Instead, clarity, brevity, and structured communication appear to be key factors influencing reassurance and usability. As an exploratory study conducted in a simulated environment, the findings are intended to generate design insights rather than generalizable conclusions. The study contributes design recommendations for the development ofpatient-centred conversational AI systems in healthcare.