Psychiatrists See Promise in Generative AI for Mental Health Care, but Not as a Replacement for Clinicians

(서울=뉴스1) 조유리 기자 = South Korean psychiatrists view generative artificial intelligence as a useful tool for supporting mental health care but remain skeptical of its ability to replace human clinicians, according to a new study examining physicians' real-world experiences with the technology.

Korea University Anam Hospital said Tuesday that a research team led by Dr. Chul-Hyun Cho of the Department of Psychiatry and Dr. Dooyoung Jung of the Korea Advanced Institute of Science and Technology (KAIST) analyzed psychiatrists' experiences, interpretations and recommendations for the safe adoption of generative AI in clinical practice. The findings were published in a leading international journal in the field of digital health.

The researchers surveyed 408 members of the Korean Neuropsychiatric Association, including practicing psychiatrists and trainees, between Oct. 27 and Dec. 26, 2024. Responses from 311 participants who provided substantive open-ended answers were included in the qualitative analysis.

The survey focused on three areas: clinicians' experiences with generative AI tools such as chatbots and diagnostic assistants in practice, the advantages and limitations of generative AI compared with human therapists, and the most urgent requirements for safely integrating the technology into mental health care.

The study concluded that generative AI should be viewed as a "clinically ambivalent technology," meaning its benefits and risks vary depending on how it is used, the intensity of use and the vulnerability of individual patients.

Physicians reported positive cases in which patients used generative AI as a low-barrier tool for emotional processing, self-management and entry into treatment. One psychiatrist described a patient who used AI not simply to generate comforting messages but to organize and articulate emotions that had previously been difficult to express, leading to symptom improvement.

At the same time, doctors reported significant risks associated with excessive use or use in high-risk situations. These included reinforcement of delusional beliefs, increased social withdrawal, overdependence on AI tools and cases linked to self-harm or suicide risk, including medication overdoses.

Researchers also found that some patients began comparing AI-generated responses with physicians' assessments, creating new challenges for therapeutic relationships, treatment adherence and trust in clinical diagnoses.

Survey participants described generative AI as a consistent and standardized tool that does not suffer from fatigue. However, they also noted that the technology remains "relationally thin" in areas requiring deeper therapeutic engagement.

According to respondents, generative AI struggles to interpret nonverbal cues and emotional nuances. They also warned that AI systems often validate users' statements without sufficient critical evaluation, potentially providing reassurance while inadvertently reinforcing distorted beliefs.

When asked about implementation priorities, psychiatrists emphasized that safety measures should take precedence over rapid adoption.

Key requirements identified by respondents included stronger governance and accountability frameworks, safety infrastructure for crisis situations and vulnerable populations, rigorous technical and clinical validation before widespread deployment, and enhanced education, oversight and institutional support.

While physicians broadly acknowledged the value of generative AI for administrative tasks, information organization, patient education and support before and after appointments, they drew a clear distinction when it came to direct patient care.

Respondents said AI should not replace clinicians in areas involving treatment decisions, crisis assessment or the development of therapeutic relationships.

The researchers suggested that psychiatrists may increasingly take on new responsibilities related to generative AI, including helping patients interpret AI-generated information, guiding appropriate use and monitoring potential risks.

"This study goes beyond measuring abstract attitudes and captures how psychiatrists are actually encountering generative AI in clinical settings," Cho said. "The findings suggest that AI applications in mental health require more layered oversight and diagnosis-sensitive safeguards than many other forms of medical AI."

The study was conducted as part of a project led by the Future Strategy Committee of the Korean Neuropsychiatric Association.

Myeongseong Kim, a doctoral student at the Ulsan National Institute of Science and Technology (UNIST), and Dr. Yoo-Seok Ahn of Seoul National University College of Medicine and the National Traffic Rehabilitation Hospital served as co-first authors. Cho and Jung were co-corresponding authors.

Separately, Korea University Anam Hospital recently announced plans to renovate its psychiatry ward as part of a broader initiative to strengthen advanced care for critically ill patients.

ur1@news1.kr