Researchers at Victoria University have developed an AI platform that can quickly spot mental health issues in online conversations, potentially making mental health care more accessible and personalized. To train the models, scientists analyzed one million health-related threads from Reddit. They built a database of mental health narratives and used expert classifications (linking language to root causes from Healthdirect Australia) to teach two AI models — PaLM 2 and AutoML.
The results are promising: PaLM 2 matched human raters’ judgments in 80% of cases, while AutoML did so in 68%. The researchers believe this approach could support mental health screening, triage, patient onboarding, and the development of digital mental health services. They emphasized that while AI can help identify patterns in large volumes of unstructured text, it’s important to remember that mental health is deeply personal, and further validation is needed before widespread clinical use.
About the Author
