Generative AI NLP Support (GAINS) for Mental Healthcare Behavioral Tagging and Support for Depression and Suicidal Tendencies

Project Description Suicide is a huge challenge for human society. World Health Organization (WHO) has announced that death by suicide is a major
cause of deaths, ahead of deaths by HIV, malaria and breast cancer, depression, extreme anxiety and other psychological,
temperamental and behavioral disorders are commonplace in the United States, particularly among those between the ages of 18 to 25,
as per the NIH. This age group experiences the ‘highest rate of mental health concerns (33.7%)’ and also the ‘highest rate of serious mental
illness (11.4%)’ (NIH). Furthermore, for U.S. children ages 10 to 14, suicide was (2020 statistics) the second leading cause of death. These
tragic scenarios present a challenge that we cannot ignore or evade. Fortunately, given the recent developments in AI, we can now hope to
meaningfully address and improve our support for affected children and people, and this proposal explores a stream of possibilities with
generative AI NLP support (GAINS) for mental healthcare. Generative AI has shown tremendous potential and is transforming the world around us. We propose to use open weights Large Language Models (LLMs) such as Llama 2 and Mistral to help address mental healthcare challenges
in the areas of severe depression and suicidal tendencies. This proposal addresses 2 research challenges: 1) How
can we best identify and flag social media and other text messages of concern using generative AI and NLP
methods and LLMs? And 2) The design for a basic GAINS system.
The current proposal seeks to identify ways to algorithmically tag text messages from human sources as being of
concern in the area of mental healthcare. In the US, the CDC has listed suicide as one of the foremost causes of
death nationally. This project will help us explore ways to Identify and tag potentially susceptible persons leading
to increased chances for timely intervention and also explore GAINS driven encouragement and positive
messaging for the timespan between episodes and human help arriving.