Research shows AI can fast track mental health care

Researchers have analyzed over a million health-related Reddit threads to develop an AI platform aimed at swiftly identifying mental health concerns. Victoria University's Associate Professor Khandakar Ahmed and Saima Rani spent two years constructing a database of mental health narratives from Reddit and trained two AI models—PaLM 2 and AutoML—on this data.

The study, published in JMIR AI, involved manually classifying 800 threads from the larger dataset by a psychiatrist and feeding the results back into the AI models. PaLM 2 matched human judge judgments in 80% of cases, while AutoML achieved this in 68% of instances. Both AI systems were then tasked with evaluating new Reddit threads, which were also assessed by human raters.

While acknowledging potential limitations, researchers believe these AI tools could aid in mental health screening, triage, patient onboarding, and digital mental health services. Mental health is highly individual, and every person's experience is unique. The researchers aimed to determine if AI could identify deeper themes in people's descriptions of their mental health struggles beyond just their symptoms.

If AI models, trained on expert-verified data, can align with human judgment in recognizing the root causes of mental health issues, it could help monitor population-level changes and guide public health responses to mental illness in Australia.