Psychological aspects of AI-mediated communication in health care (VERIKOM)
Center for Psychosocial Medicine, Institute of Medical Psychology
March - October 2022
ECG, Cortisol, Alpha-amylase
Background and aim
The EU ethical guidelines for trustworthy artificial intelligence (AI) call for the development of concrete recommendations for dealing with vulnerable people, which include patients in a medical context. Current developments in AI will have a significant impact on doctor-patient interactions, and it is still unclear how patients will react to AI-mediated communication in specific situations. Since it would not be ethical to study real patients, this experiment will be conducted with healthy volunteers. The aim of the overall project is to make recommendations for the use of digitalised and AI-based technologies in the healthcare sector and to focus on psychological aspects and sensitive patients that have received little attention to date.
For this purpose, a standardised interaction sequence was designed in which we simulate a genetic counselling session. This genetic counselling will take place in four different conversation formats to which the test persons will be randomly assigned: There will be a face-to-face conversation, which will serve as a baseline. The next stage is a video conversation with a doctor and is thus the first digitised form of the conversation. The two forms of conversation that now follow depict interaction with an AI in the form of a chatbot (written exchange) and an avatar (video conversation with a virtual doctor).
The stress reaction of the test persons that they will show due to the conversation serves as the dependent variable. We assume that the subjects will feel more stress the more impersonal the interaction is. We will record the stress of the test persons on several levels: In addition to a subjective assessment, we will take saliva samples and analyse the hormones cortisol and alpha-amylase in our in-house wet laboratory. In addition, we are also interested in continuous stress measures in the form of HRV and skin conductance in order to be able to take a look at the components of the conversation and stress in the course of the conversation. In addition to stress, we are also interested in the variables of memory performance, user acceptance of the digital formats and the doctor-patient relationship as perceived by the patient. All these variables are captured through subjective assessments.