Investigating How Clinicians Form Trust in an AI-Based Mental Health Model: Qualitative Case Study

Area of Science:

• Mental Health Technology
• Artificial Intelligence in Healthcare
• Human-Computer Interaction

Background:

• Trust in artificial intelligence (AI) is a significant barrier to its adoption in mental health care.
• This study investigates trust formation in an AI mental health model and its interface among clinicians.

Purpose of the Study:

• To explore clinicians' perspectives on building trust in an AI-supported mental health screening model.
• To identify factors influencing trust formation in AI mental health tools.

Main Methods:

• Qualitative case study utilizing semistructured interviews with clinicians.
• Thematic analysis was employed to identify key trust formation factors.

Main Results:

• Clinicians' trust evolved sequentially: sense-making, risk appraisal, and conditional reliance.
• Model explainability (visualizations, feature attribution, pseudo-sumscores) was crucial for trust.
• Trust was context-bound to low-risk scenarios and safety protocols; diverse data types and ongoing evaluation are needed for broader trust.

Conclusions:

• Clinician trust in AI is contextually and sequentially built, depending on model performance and clinical reasoning alignment.
• Interpretability features, presented in clinically familiar formats, are essential for intrinsic trust.
• Responsible AI deployment requires rigorous evaluation data and clinically relevant data integration.