Family Caregivers’ Perspectives on AI-Assisted Mental Health Care in China: Acceptance, Concerns, and Implementation Conditions

Technical Summary: Family Caregivers' Perspectives on AI-Assisted Mental Health Care in China

Problem Statement
The rapid integration of Artificial Intelligence (AI), particularly generative AI and Large Language Models (LLMs), into psychiatry presents unique challenges compared to other medical specialties. Unlike somatic medicine, psychiatric practice relies heavily on language-based communication, subjective experiences, and behavioral observation, raising concerns regarding empathy, ethical boundaries, and the therapeutic alliance. While AI shows promise in diagnostic support, risk prediction, and service efficiency, its adoption in mental health has been slower due to the complex psychosocial dimensions of disorders and the subjective nature of care.

Crucially, family caregivers (FCGs) bear the primary burden of daily supervision and emotional support for patients with mental disorders, often experiencing psychological distress comparable to or exceeding that of the patients themselves. Despite their central role in the recovery trajectory and the potential for AI to alleviate workforce shortages and caregiving burdens, there is a significant gap in understanding how these non-professional caregivers perceive, accept, or resist AI technologies. Existing literature lacks qualitative exploration of the inner experiences and specific influencing factors shaping FCGs' attitudes toward AI in the Chinese healthcare context, where family members are the primary caregivers.

Methodology
This study employed a qualitative descriptive design using semi-structured in-depth interviews to explore the attitudes of primary caregivers toward AI applications in clinical settings.

• Setting and Participants: Data were collected between January and March 2026 at the Henan Mental Health Center, a large national-level psychiatric hospital in China. The study recruited 18 primary caregivers (10 female, 8 male) of hospitalized patients with mental disorders. Inclusion criteria required participants to be aged 18–60, immediate family members with at least three months of caregiving experience, and willing to provide informed consent. Participants were excluded if they had no prior exposure to AI or suffered from severe physical/mental illness.
• Sampling: Purposive sampling was used to ensure diversity across five departments: Child and Adolescent Psychiatry, General Psychiatry, Addiction Psychiatry, Early Intervention Psychiatry, and Geriatric Psychiatry. Data collection continued until saturation was achieved (no new themes emerged after the 18th interview).
• Data Collection: Face-to-face interviews were conducted in private rooms by two experienced researchers (a senior psychiatric nurse and an academic nursing researcher). Interviews lasted 30–40 minutes, were audio-recorded, and supplemented with notes on non-verbal cues. An interview guide, developed through literature review and expert consultation, covered expectations, concerns, and suggestions regarding AI.
• Data Analysis: Transcripts were analyzed using Braun and Clarke's thematic analysis approach via NVivo 12.0. An inductive coding strategy was employed, allowing themes to emerge naturally from the data. Rigor was ensured through member checking, consensus coding among researchers, and researcher reflexivity to bracket preconceived assumptions.

Key Results
Thematic analysis of the interview data revealed four core themes reflecting caregivers' multifaceted perspectives:

1. Perceived Clinical Benefits: Caregivers expressed high expectations for AI's functional utility, particularly in:

   • Monitoring and Safety: A strong desire for AI-driven early warning systems to detect emotional fluctuations, predict self-harm/suicide risks, and monitor cognitive impairment (e.g., wandering). Caregivers viewed this as a shift from passive response to proactive intervention.
   • Treatment Management: Expectations included AI-assisted medication adherence (reducing confrontational dynamics), objective tracking of symptom progression to compensate for subjective reporting biases, and improved access to specialist expertise through digital integration.

2. Emotional Concerns and Human Connection: Despite acknowledging potential benefits, caregivers held deep-seated anxieties regarding the human element of care:

   • Irreplaceability of Human Warmth: AI was frequently described as a "cold machine" incapable of genuine emotional variability. Caregivers feared that over-reliance on AI would erode the therapeutic alliance and lead to social withdrawal or pathological dependence in patients.
   • Functional Decline: Concerns were raised that excessive reliance on AI could diminish the independent thinking and problem-solving abilities of patients (particularly adolescents) and shift the caregiver's burden from physical tasks to sustained psychological strain (e.g., anxiety over false alerts).

3. Trust and Adoption Drivers: Acceptance was not solely technical but socio-psychological, structured hierarchically:

   • Physician Authority: Trust was primarily anchored in the endorsement of treating physicians. Caregivers deferred to clinical judgment over algorithmic output and trusted hospital-integrated systems more than private apps due to perceived regulatory oversight and accountability.
   • Social Discourse and Experience: Negative media coverage regarding data leaks or algorithmic errors acted as a significant barrier. Conversely, positive prior experiences with AI tools (e.g., for information seeking) facilitated acceptance.

4. Barriers to Implementation: Significant practical and systemic obstacles were identified:

   • Privacy and Data Governance: Intense resistance existed regarding the collection of sensitive audio/video data due to fears of stigma affecting future employment and social integration. However, some caregivers indicated a willingness to compromise privacy for critical safety needs if strict confidentiality guarantees were provided.
   • Cost and Accountability: Financial constraints were a major barrier, with caregivers expressing that AI devices should be covered by medical insurance or government subsidies. Furthermore, the ambiguity of legal liability in cases of AI failure (misdiagnosis, technical errors) was a primary concern, with caregivers uncertain whether responsibility lay with manufacturers, hospitals, or developers.

Significance and Contributions
This study provides one of the first qualitative investigations into the attitudes of Chinese family caregivers toward AI in psychiatric care. Its significance lies in:

• Contextual Insight: It offers specific insights into the Chinese healthcare model, where family members are the primary caregivers, highlighting how cultural and systemic factors influence technology acceptance.
• Balanced Perspective: The findings reveal a "cautiously accepting" stance, moving beyond binary views of AI as purely beneficial or harmful. It underscores that while caregivers value AI for risk monitoring and efficiency, they prioritize the preservation of humanistic care and emotional connection.
• Implementation Guidance: The study identifies that successful implementation requires more than technical advancement; it necessitates robust institutional frameworks. Specifically, it calls for clear legal accountability mechanisms, data governance protocols that address stigma, and policy support (subsidies/training) to ensure accessibility for financially strained families.
• Human-AI Collaboration: The results suggest that AI should be positioned as a supportive tool for augmentation rather than a replacement for human interaction, aligning with the need to maintain the therapeutic alliance while addressing the chronic burden of mental illness care.

The authors conclude that future AI deployment in psychiatric nursing must be caregiver-centered, balancing technological effectiveness with ethical safeguards and humanistic values to ensure real-world adoption and efficacy.