Artificial intelligence in positive mental health: a narrative review

The review introduces artificial intelligence (AI) as a field aiming to create systems that perform human-like cognitive tasks (learning, reasoning, problem solving, language understanding). It traces AI’s evolution from the 1950s to the present “Age of AI,” highlighting its broad definition due to varying conceptions of intelligence. In mental health, AI’s increasing integration underpins digital mental health, enhancing accessibility, personalization, and intervention efficacy. The review’s purpose is to examine how AI contributes to positive mental health through awareness, support, and intervention across conditions and populations, and to critically appraise ethical, reliability, and cultural sensitivity issues.

The narrative review synthesizes historical and contemporary developments of AI, covering milestones from Turing’s foundations and the Dartmouth workshop to modern machine learning and deep learning breakthroughs. It surveys AI’s cross-industry applications (finance, healthcare, manufacturing, transportation) and then details AI components relevant to mental health: machine learning (supervised/unsupervised), deep learning/ANNs, NLP, reinforcement learning, and computer vision. It provides an overview of AI applications in mental health care categorized as awareness (social media sentiment analysis, chatbots for psychoeducation), support (mobile apps for medication reminders, mood tracking, personal sensing/digital phenotyping, online peer support), and intervention (risk prediction, early detection via EHR and behavioral data, NLP chatbots for triage and coping strategies, neurofeedback/BCI, decision support). Advantages are summarized for cognitive domains (automated assessment via SVMs, CNNs and other architectures), intellectual/developmental disorders (AI-assisted screening via EHR, neuroimaging, biomarkers), neurodegenerative disorders (MRI-based ML for early detection), seizure detection (EEG-based ML), and affective/emotional domains (Emotional AI for sensing and regulation via wearables, mindfulness and relaxation guidance). The review also addresses applications in mood disorders (wearable and social rhythm monitoring), ASD (video analysis of gaze/gestures, tablet-based assessments), and schizophrenia (speech/NLP analyses for onset and relapse prediction). Critiques focus on ethics/privacy, reliability/accuracy, and potential biases/lack of cultural sensitivity. Recommendations include diverse training/validation datasets, transparent and accountable AI systems, and strong human oversight and collaboration.

• AI enables scalable awareness through social media sentiment analysis to monitor public discourse and target psychoeducation; AI-driven chatbots provide personalized information, coping strategies, and referrals.
• Support functions include medication adherence reminders, mood tracking, and personal sensing/digital phenotyping (e.g., detecting activity changes indicative of depression) with platforms that can flag concerning posts and foster peer support communities.
• Intervention capabilities span prediction and early detection using EHRs and behavioral data; an AI-based decision support system (DSS) has been developed for detecting and diagnosing multiple mental disorders. NLP chatbots can triage based on mood, stress, energy, and sleep, offering behavioral techniques or prompting clinical escalation when safety concerns arise.
• Cognitive assessment benefits: ML methods (SVM, neural networks; CNN variants such as AlexNet, GoogLeNet, LeNet5) improve diagnostic accuracy and efficiency for cognitive impairments.
• Neurodegenerative disorders: ML analysis of MRI enables early differentiation (e.g., SVM distinguishing Alzheimer’s disease from frontotemporal lobar degeneration and from healthy controls); 3D neural network architectures have also been used for Alzheimer’s detection.
• Seizure detection: EEG-based ML can identify epileptic seizure onset despite pattern variability.
• Emotional AI: Emotion sensing via facial/voice/physiological signals supports real-time regulation (guided relaxation, breathing, mindfulness) and enhances emotional intelligence through longitudinal feedback.
• Evidence in mood disorders: Combined mobile/wearable sensing yields objective markers for depression and bipolar disorder; automated assessment of the Social Rhythm Metric (SRM) via smartphone sensing assists stability monitoring.
• ASD: Early detection via analysis of eye gaze, gestures and tablet-based gameplay; sensor-integrated toys and front-facing camera measures support social-emotional assessment.
• Schizophrenia: Automated speech analysis and NLP track affect-related changes for onset prediction and relapse monitoring.
• Ethical, reliability, and cultural critiques emphasize data privacy/security, consent, algorithmic transparency, dataset bias, contextual understanding limits, and the need for human oversight.
• Recommendations: diversify datasets and labeling practices, implement explainability and documentation, perform regular audits, establish ethics review structures, and integrate clinicians for validation, customization, and oversight.
• Contextual data point: In a 2022 survey of 850 organizations across 18 geographies, 77% prioritized AI regulations as company-wide policies and 80% planned investment in ethical AI, reflecting broader readiness for responsible deployment.

Findings demonstrate that AI contributes to positive mental health across the continuum—raising awareness, enhancing support, and enabling earlier, more personalized interventions. By leveraging ML, NLP, RL, and computer vision, AI augments traditional care with scalable tools for detection, monitoring, and self-regulation, addressing gaps in accessibility and personalization. However, the same technologies introduce concerns about privacy, transparency, reliability, and cultural sensitivity. The review argues for a socio-technical integration: AI should assist clinicians rather than replace them, with robust human oversight to interpret and contextualize outputs. It emphasizes transparent, explainable systems; diverse, representative datasets; and culturally competent design to minimize bias and improve equity. These practices align AI’s strengths with ethical responsibilities, thereby enhancing clinical relevance and trust while advancing positive mental health outcomes.

AI’s evolution and integration into digital mental health reveal substantial promise for improving understanding, diagnosis, and treatment, and for promoting positive mental health through awareness, support, and tailored interventions. The review underscores that responsible deployment requires ongoing collaboration among clinicians, researchers, patients, data and computational scientists, and regulators. It calls for continuous ethical scrutiny, explainable models, diverse datasets, and human-in-the-loop oversight to ensure AI augments care rather than replacing human connection. Future work should expand generalizability across populations and settings, update systems in pace with rapid technological advances, and empirically validate AI interventions through clinical trials and real-world studies.

The review notes limited generalizability due to its narrative scope, which may not capture the full breadth of AI’s impact in mental health. The rapidly evolving AI landscape challenges the currency of findings and necessitates ongoing updates. Methodological details of systematic searching and appraisal are not provided, and many applications remain early-stage, requiring rigorous validation, bias auditing, and cultural adaptation before clinical generalization.