Telstra Foundation supports Gradient Institute research towards safer mental health chatbots

16 August 2023

Young girl holding phone

Telstra Foundation, through its Young & Connected Fund, has provided Gradient Institute with a seed grant to work towards the safe and effective use of AI and Large Language Models (LLM) in mental health chatbots. Gradient, with its technical expertise in the safe and reponsible use of AI, will use the seed grant to work with mental health experts, young people and others to develop the scope for guidelines for the safe use of such technology in digital mental health support.

Gradient Institute has concerns about the use of large language models (LLMs) for mental health chatbots given their uncontrolled nature, their propensity for manipulation and their lack of transparency.

Recent studies show an increase in teenagers experiencing mental health issues — exacerbated by the COVID-19 pandemic — and that young people do not feel well-equipped to manage their mental health. Thus there has been a growth in chatbot applications and services targeted to mental health support.

However, the recent popularisation of more generalist AI-based LLMs like OpenAI’s GPT (as used by ChatGPT) has led to these models being adopted in chatbots marketed as ‘mental health tool’ and existing online mental health services. These services are exposed to young people without the endorsement of mental health professionals and without appropriate guidelines, and the use of LLMs for these purposes is only projected to increase.

Gradient Institute, with a strong aptitude in AI risk identification and mitigation, wants to work with mental health experts, medical health practitioners, groups of school students, and others to develop practical guidelines for the safe and effective use of AI and LLMs in mental health chatbots. The guidelines would be informed by the potential users of such chatbots, youth mental health research and practice and Gradient’s technical knowledge of the capabilities and limitations of AI and LLMs. This seed project will identify the scope and requirements for these guidelines, which would enable a subsequent proposal for a further project to develop and pilot these guidelines.