講演情報
[5O2-IS-5a-01]Empathy as a Controllable Relational Design Variable in Human–AI Interaction: Experimental Evidence from a Large-Scale Randomized Trial in an Antidepressant Digital InterventionEmpathy: Decorative Element or Therapeutic Factor?
〇mirai so1 (1. Ichikawa General Hospital, International University of Health and Welfare)
regular
キーワード：
NLP、depression、Internet Cognitive Behavior Therapy、randomized controlled trial、empathy
This study focused on empathy as a controllable interaction component in human–AI interaction and experimentally evaluated its effects. A three-arm parallel randomized controlled trial (N = 1187) was conducted using a fully self-guided internet-based cognitive behavioral therapy (iCBT) program. Participants were assigned to (1) an iCBT group with NLP-based empathic AI feedback, (2) an otherwise identical iCBT group without AI feedback, or (3) a waitlist control group. The primary clinical outcome was depressive symptom severity measured by the PHQ-9, and the primary behavioral outcome was weekly participation rate. Intention-to-treat analyses showed no significant between-group differences in depressive symptom improvement, whereas empathic AI feedback significantly enhanced engagement (OR = 1.23). Furthermore, exploratory analyses in an engagement-adherent sample showed that only the AI condition demonstrated a reduction in the proportion of participants meeting the PHQ-9 score threshold of 10 or higher, representing clinically relevant depressive severity, suggesting a potential role of empathic AI feedback in preventing symptom deterioration.