Investigating How Clinicians Form Trust in an AI-Based Mental Health Model: Qualitative Case Study
Bibliographic record
Abstract
Background: Trust in artificial intelligence (AI) remains a critical barrier to the adoption of AI in mental health care. This study explores the formation of trust in an AI mental health model and its human-computer interface among clinicians at a web-based mental health clinic in the Region of Southern Denmark with national coverage. Objective: This study aims to explore clinicians' perspectives on how trust is built in the context of an AI-supported mental health screening model and to identify the factors that influence this process. Methods: This was a qualitative case study using semistructured interviews with clinicians involved in the pilot of a mental health AI model. Thematic analysis was used to identify key factors contributing to trust formation. Results: Clinicians' initial attitudes toward AI were shaped by prior positive experiences with AI and their perception of AI's potential to reduce cognitive load in routine screening. Trust development followed a sequential pattern resembling a "trust journey": (1) sense-making, (2) risk appraisal, and (3) conditional decision to rely. Trust formation was supported by the explainability of the model, particularly through (1) visualization of confidence and uncertainty through violin plots, aligning with the clinicians' expectations of decision ambiguity; (2) feature attribution for and against predictions, which mirrored clinical reasoning; and (3) use of pseudo-sumscores in the AI model, which increased interpretability by presenting explanations in familiar clinical formats. Trust was contextually bounded to low-risk clinical scenarios, such as preinterview patient screening, and contingent on safety protocols (eg, suicide risk flagging). The use of both structured and unstructured patient data was seen as key to expanding trust into more complex clinical contexts. Participants also expressed a need for ongoing evaluation data to reinforce and maintain trust. Conclusions: Clinicians' trust in AI tools is contextually and sequentially constructed, influenced by both model performance and alignment with clinical reasoning. Interpretability features were essential in establishing intrinsic trust, particularly when presented in ways that resonate with clinical norms. For broader acceptance and responsible deployment, trust must be supported by rigorous evaluation data and the inclusion of clinically relevant data types in model design.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.028 | 0.060 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.013 | 0.013 |
| Scholarly communication | 0.006 | 0.007 |
| Open science | 0.003 | 0.009 |
| Research integrity | 0.003 | 0.006 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".