Human-Centred Artificial Intelligence: Pilot Design of a Virtual Patient

Abstract

Virtual patient (VP) avatars driven by artificial intelligence (AI) could enhance healthcare training and offer flexible learning. Recent advances have developed large language models (LLMs) for more natural and contextual responses, but research into LLMs for VPs is still emerging. This research aimed to co-create prompts and training for a conversational AI system supporting a VP. The objective was to pilot the processes of prompt engineering and AI training in line with iterative development and human-centred design (HCD). Co-creation underpinned by design thinking was employed, involving collaboration between the university as User Experience designers and industry AI Engineers. Fifty-nine participants from student, academic, clinical, and technology groups were recruited for six co-creation workshops. Four versions of the VP prompts were developed and tested, with nine training samples included in the reinforcement fine tuning data set. This research adopts an HCD approach, outlining various techniques for co-creating optimal prompts that consider the VP's personality, life context and clinical situation. A definition of VPs as it relates to human-centred AI (HCAI) development and a framework for the process is discussed.

Ähnliche Arbeiten