The rapid advancement of artificial intelligence (AI) presents a revolutionary opportunity across various sectors, including healthcare. AI's capacity to process vast datasets and identify complex patterns has led to significant transformations in fields like radiology, pulmonology, and dermatology. However, applying AI to mental healthcare, particularly for individuals with serious mental illness (SMI), presents unique challenges. The subjective and nuanced nature of mental health conditions, coupled with historical societal stigma and mistrust stemming from past practices, creates a complex landscape for AI integration. While AI offers the potential for improved access to services, early intervention, and personalized treatment, ethical and regulatory considerations must be carefully addressed. Currently, there's a scarcity of evidence regarding AI's effectiveness in mental healthcare for individuals with SMI, necessitating a deeper understanding of stakeholder perceptions to guide responsible development and implementation. This study aims to explore how various mental health stakeholders perceive the opportunities and challenges of integrating AI to support mental health care for persons with SMI.

Existing literature highlights the potential of AI in mental healthcare, acknowledging its capacity to process large datasets and identify patterns relevant to mental health conditions [1, 2]. However, the unique challenges posed by mental health, particularly concerning SMI, are also discussed extensively. Studies emphasize the ethical and regulatory aspects of integrating AI in this sensitive field [5, 6], acknowledging the importance of responsible implementation to avoid unintended negative consequences. A significant gap highlighted in previous research is the limited evidence on the effectiveness of AI in mental healthcare for those with SMI [7]. This study directly addresses this gap by investigating the perspectives of mental health stakeholders, adding valuable insights to the ongoing discussion.

This study employed a qualitative research design utilizing individual interviews with a purposive sample of diverse mental health stakeholders. Participants were recruited from government, hospital, municipality, university/research institution, health industry/cluster, and user organization sectors. Inclusion criteria focused on experience with digital health in the field of mental health. Informed consent was obtained from all participants, who were assured of voluntary participation and the right to withdraw at any time. Data collected from the corresponding author's PhD study (previously unpublished) was used. Ethical approval for the PhD study was obtained from the Norwegian Agency for Shared Services in Education and Research (SIKT) [reference no. 269350]. Interviews, conducted via Microsoft Teams, explored participants' views on AI's potential to support persons with SMI, specifically its application in digital mental health solutions and its potential to improve the quality of life for individuals with SMI. Data analysis involved thematic analysis [8] using NVivo 14 and Microsoft Excel, following a six-step process: familiarization, generating initial codes, generating themes, reviewing themes, defining and naming themes, and writing up. Sentiment analysis was also conducted using Autocode Wizard in NVivo 14 to gauge the overall tone of responses (positive, moderately positive, moderately negative, or negative). It is important to note that Autocode Wizard analyzes sentiment at the word level, not classifying entire content based on sentiment or a Likert scale.

Twenty-two informants participated, predominantly male (55%) and aged 40-59 (82%). A majority (59%) had healthcare backgrounds (psychiatry, psychology, nursing, social education). Sentiment analysis revealed that 75% of responses expressed moderately negative or negative sentiment towards AI use in supporting persons with SMI. Thematic analysis yielded four main themes: 1. **When AI meets serious mental illness:** This theme explored the potential benefits and risks of AI. Some informants believed AI could interrupt negative thought patterns and empower individuals, while others expressed concern that it could exacerbate social isolation and lead to harmful interactions if not carefully managed and monitored. The potential for misuse, particularly regarding individuals seeking confirmation of harmful beliefs, was highlighted. 2. **Human-centered AI for humanity:** This theme stressed the importance of a human-centered approach to AI development and implementation. Informants emphasized the need for personal adaptation, safety, and the preservation of human connection. The vulnerability of specific groups, such as older adults with SMI, to the potential pitfalls of misinterpreting AI interaction was discussed. The importance of addressing the digital divide and ensuring accessibility for diverse users was emphasized. 3. **AI to improve service delivery for serious mental illness:** This theme explored AI's potential to enhance clinical decision support and resource management. Informants saw the potential for AI to assist in timely identification of escalating symptoms, provide support during waiting periods for treatment, and aid in monitoring physical parameters relevant to SMI. AI's potential for reducing administrative burdens on healthcare professionals was highlighted. 4. **Building AI competence to support serious mental illness:** This theme focused on the need for increased AI literacy and competence among all stakeholders. The scarcity of evidence regarding AI's effectiveness in mental healthcare, particularly for SMI, was acknowledged. The importance of upskilling mental health professionals and empowering individuals with SMI to use AI responsibly and effectively was emphasized.

This study reveals a complex interplay of opportunities and challenges surrounding AI integration in mental health for persons with SMI. While AI shows promise in enhancing efficiency and improving service delivery, concerns regarding emotional support, social isolation, and the potential for misuse remain. The discrepancy between the generally positive tone in qualitative findings and the predominantly negative sentiment analysis results highlights the heterogeneity of stakeholder perspectives. This emphasizes the need for a careful, holistic approach that considers the unique vulnerabilities and needs of this population, with specific attention paid to ethical considerations, user accessibility, and the crucial role of human interaction. The need for robust ethical guidelines and regulatory frameworks, particularly regarding data privacy, algorithmic bias, and transparency, is paramount. Bridging the digital divide and ensuring equal access to AI-driven tools are also crucial aspects of responsible implementation.

This research underscores the significant potential of AI to support persons with SMI while highlighting the crucial need for careful consideration of ethical, practical, and accessibility challenges. Future research should focus on developing and rigorously testing AI tools that effectively address these challenges, ensuring the safety, equity, and human-centered nature of AI's integration into mental healthcare. Further investigation into the specific needs of vulnerable populations and the development of effective strategies for enhancing AI literacy are essential to maximize the benefits and minimize the risks of AI in this field.

The study's findings are based on qualitative data from a specific context (Norway), potentially limiting the generalizability of results to other settings. The purposive sampling method may have introduced bias, and the sentiment analysis, limited to individual words rather than comprehensive content, might not fully capture the nuances of stakeholder opinions. Further research with larger, more diverse samples and employing both qualitative and quantitative methods is recommended to validate these findings and explore the broader implications of AI integration in mental healthcare.