Effects of information relevance communication on novices’ experience and performance with a cooperative ultrasound diagnosis system
Mon-H5-Talk 1-1005
Presented by: Tim Schrills
Novices may particularly benefit from medical artificial intelligence (AI) in diagnostic tasks, e.g., ultrasound-based diagnosis. However, novices often have limited domain knowledge, which hampers their ability to understand how a medical AI processes information and could negatively affect cooperation. The present research investigates the effects of communication of diagnostic information between humans and medical AI. Based on an initial experiment with N = 30 laymen, we designed interactions highlighting the users' influence on information (i.e., short ultrasound video recording) used by the system. To this end, we identified information that 1) is under the user's control and 2) is where the system cannot monitor the quality. In a second between-subject experiment with N = 40, medical novices were either instructed to apply wrong pressure levels within the ultrasound examination proactively and to observe effects on the system information processing or only received basic instructions to conduct the examination. We examined Subjective Information Processing Awareness and the relation between reported confidence and performance regarding diagnostic quality. Results indicated that instruction to manipulate the information provided by novices' actions explicitly and to observe effects on the system impacts the development of diagnostic quality.
Keywords: Human-AI Interaction, Medical AI, Automation