JMIR Publications

ABSTRACT

Background:

The increasing prevalence of mental health issues among adolescents and young adults, coupled with barriers to accessing traditional therapy, has led to growing interest in artificial-intelligence-driven (AI-driven) conversational agents (CAs) as a novel digital mental health intervention. Despite accumulating evidence suggesting the effectiveness of AI-driven CAs for mental health, there is still limited evidence on their effectiveness for different mental health conditions in adolescents and young adults.

Objective:

This study aims to examine the effectiveness of AI-driven Conversational Agents for mental health among young people, and explore the potential moderators of efficacy.

Methods:

Five main databases (PubMed, PsycINFO, EMBASE, Cochrane Library, and Web of Science) were searched systematically dated from the establishment of the database to August 6, 2024. Randomized controlled trials comparing AI-driven CAs with any other type of control condition in improving depressive symptoms, generalized anxiety symptoms, stress, mental well-being, and positive and negative affect were considered eligible when they were conducted in young people aged 12-25 years. The quality of these studies was assessed using the Cochrane Risk of Bias tool. Data were extracted by 2 independent reviewers and checked by a third reviewer. Pooled effect sizes (Hedges’ g) were calculated using random effect models and visually presented in forest plots.

Results:

A total of 15 articles (including 16 trials) were included, involving 1974 participants. The results indicated that, after adjustment for publication bias, AI-driven CAs had a moderate-to-large (g = 0.60, 95% CI 0.36-0.84) effect on depressive symptoms compared to control conditions. However, their adjusted effect sizes for generalized anxiety symptoms (g = 0.06, 95% CI -0.21-0.32), stress (g = 0.002, 95% CI -0.19-0.20), positive affect (g = 0.01, 95% CI -0.24-0.27), negative affect (g = 0.07, 95% CI -0.13-0.27), and mental well-being (g = 0.04, 95% CI -0.21-0.29) were all non-significant. Subgroup analyses revealed that AI-driven CAs were particularly effective in improving depressive symptoms among subclinical populations (g = 0.73, 95% CI 0.51-0.95).

Conclusions:

The findings highlight the potential of AI-driven CAs for early intervention in depression among this population, and underscore the need for further improvements to enhance their efficacy across a broader range of mental health outcomes. Key limitations of the reviewed evidence include heterogeneity in therapeutic orientations of CAs and lack of follow-up measures. Future research should explore the long-term effect of AI-driven CAs on mental-health outcomes.