BACKGROUND As artificial intelligence (AI) becomes deeply embedded in modern workflows, its impact on employee mental health is not sufficiently studied. This paper presents findings from a cross sectional survey of 200 professionals across technical and managerial roles, examining the relationship between AI adoption, burnout, and psychological stress. Using advanced statistical models, including logistic regression interpreted via SHAP (SHapley Additive exPlanations), we identify key predictors of burnout, such as weekly work hours exceeding 50 (odds ratio: 4.87), lack of managerial support, and poor transparency regarding AI enabled evaluation. Notably, fear of AI driven job loss was not a significant predictor (chi-square p = 0.766). To enhance novelty, we introduce a Bayesian updated SHAP framework for dynamic burnout risk assessment, integrating real time user feedback. Building on these insights, we propose an HCI guided intervention model, demonstrating a functional prototype assistant that combines SHAP diagnostics with personalized wellness nudges and gamified elements. The contributions are fourfold: (1) empirical evidence from a larger sample quantifying structural predictors of AI-related burnout, (2) a novel Bayesian-SHAP interpretability model for burnout risk, (3) mathematical formulations for predictor interactions, and (4) a human-centered intervention prototype with pilot evaluation showing 25\% reduction in perceived stress. Our results emphasize that organizational transparency, workload management, and empathetic HCI design are crucial for employee well-being beyond AI-induced job anxiety. OBJECTIVE The primary objective of this study is to investigate how the increasing adoption of artificial intelligence (AI) in workplace environments impacts employee mental health, particularly in terms of burnout and psychological stress. While much of the existing literature emphasizes the economic or productivity aspects of AI, this research aims to shift the focus toward the human experience. Specifically, it explores the structural predictors of AI-related burnoutsuch as overwork, lack of managerial support, and poor transparency in AI-enabled evaluation using explainable machine learning techniques. The study also seeks to introduce a novel, actionable framework that not only predicts burnout risk but explains it in an interpretable, human-centered way, thereby enabling timely interventions. METHODS The study utilized a cross-sectional survey design, collecting responses from 200 professionals across technical and managerial roles through platforms like LinkedIn, Slack, and email forums. The 25-item instrument assessed variables including work hours, AI usage, perceived leadership support, and mental health indicators. Quantitative analyses included logistic regression modeling with SHAP (SHapley Additive Explanations) to identify and interpret key predictors of burnout. To enhance the model's adaptability, a Bayesian update mechanism was integrated into SHAP, allowing real-time risk recalibration based on user input. Qualitative insights were derived from thematic analysis of open-ended responses, which informed the development of user personas. Finally, the study culminated in the design and pilot testing of a functional prototyp,a Streamlit-based Burnout-Aware Assistant, that delivers personalized risk assessments and wellness nudges. RESULTS The results revealed that structural factors—especially working more than 50 hours per week (odds ratio: 4.87), lack of managerial support, and poor transparency in AI evaluations—were far more predictive of burnout than fear of AI-driven job loss (p = 0.766). SHAP analysis highlighted “weekly work hours” and “support” as the top contributors to burnout risk, with an overall model accuracy of 78%. The prototype assistant, when tested in a small pilot, demonstrated a 25% reduction in self-reported burnout scores and a 40% increase in engagement through gamified elements. Bayesian SHAP updates allowed dynamic recalibration of burnout risk, making the system adaptive and transparent. These findings underscore the need for organizations to go beyond AI fear narratives and address deeper systemic and design-level factors affecting employee well-being. CONCLUSIONS In this study, we investigated how AI-integrated work environments contribute to psychological stress and burnout. Our findings show that while fear of AI is often present, it is not a strong predictor of burnout. Instead, structural factors—such as extended work hours and poor managerial support—play a more significant role. To turn these insights into action, we built the Burnout-Aware Assistant: a real-time, interpretable, and human-centered prototype. The assistant explains personalized risk using SHAP values and delivers wellness nudges based on user input. Three illustrative scenarios—low, medium, and high risk—demonstrate how users can both understand and act upon their results. By embedding explainability into wellness design, our assistant shifts burnout monitoring from abstract HR metrics to a personalized, empowering experience. We hope this work inspires the creation of ethical AI tools that support—not just evaluate—the people who use them.
Naik et al. (Fri,) studied this question.