PRELIMINARY QUALITATIVE STUDY ON AI AND BURNOUT: ​diagnostic potential of vocal biomarkers​

bjective Although the extant literature indicates a growing interest in the field of AI vis-a-vis depression and stress, research on how technological advances support burnout assessment is lacking. Therefore, this paper aims to provide insights on whether by analyzing vocal biomarkers, AI can support early diagnosis of burnout syndrome. Methods We conducted semi-structured interviews to explore our research objective. Our interviewee's sample includes professors, PhD students, and psychotherapists - 20 experts in burnout assessment. Collected interviews were transcribed and analyzed through word embeddings and thematic analysis. Three investigators inductively coded every sentence. Results Primary results identified five overarching themes: (1) limitations of assessment tool; (2) physiological biomarkers; (3) additional factors supporting assessment; (4) organizational responsibility; and (5) implications of AI diagnostic tools. Thus, several factors were reported as crucial but currently missing in burnout assessment - lack of accuracy, multidisciplinary/inter-level dimensions, clinical assessment, and continuous measurement of the disease. Respondents also identified the need for physiological assessment due to the connection of burnout with stress and cognitive workload. Therefore, biomarkers were added such as the variability of weight, the measurement of biofeedback, and the changes of paralinguistic features called vocal biomarkers. Additional factors - non-verbal languages and word semantic analysis - were reported as key elements supporting burnout assessment that needs to be integrated into AI tools. Another theme was related to the organization’s responsibility to assess psychosocial risks and arising financial implications (e.g., questionnaires and wearables). Lastly, respondents shared their concerns about the integration of AI diagnostic tools that refer to practical (e.g., data storage, trust) and ethical (e.g., privacy, missing human interaction) implications. Conclusions Our results highlight the need for more accurate and data-oriented tools supporting the diagnostic processes. However, to make this tool holistic and preventive, additional factors should be integrated - nonverbal cues, physiological parameters, and clinical features.