The initial psychiatric interview centers on patients’ chief complaints, symptoms, and functional impairments, forming the basis of diagnostic impressions. In real clinical practice, however, interviews are constrained by limited time and the unpredictability of patient responses, making it difficult to secure essential information efficiently. While prior conversational agents have focused on conversationalizing validated instruments or advancing interview systems in general medical domains, little research has addressed the distinctive challenges of initial psychiatric history-taking from clinicians’ perspective. We present a flexible psychiatric interviewer that dynamically adapts question flow and prioritizes clinically essential information within time constraints, with a clinical dashboard for efficient review. We evaluated the system through 1,440 simulated patient dialogues and follow-up interviews with 19 clinicians. Results show that it captures essential information within a limited time while preserving conversational flexibility and empathy, highlighting design implications for coachable and responsible AI interviewers that align with clinical practice.
ACM CHI Conference on Human Factors in Computing Systems