Youth with migration backgrounds and lower socio-economic positions are at a 2–3x higher risk of mental health problems. Artificial Intelligence (AI) chatbots may serve as personalized support for the development and social participation of young people from diverse backgrounds. However, there is a lack of knowledge on co-creating AI-based mental health interventions for and with young people.  

Can Artificial Intelligence (AI) help youth navigate through life’s challenges, especially when it comes to mental well-being? Our project aims to understand how we can harness the potential of AI, particularly Large Language Models (LLMs), to support the wellbeing of adolescents. 

The objectives of this project were to:  

1. identify what the mental well-being needs of young people from diverse backgrounds partaking in professional youth work are, their AI wishes and experiences. 
2. co-create an LLM prototype assistant that can support adolescents in healthy behaviours and social skills. 

We used the newly developed open-source Erasmian Large Language Model (ELM). ELM safeguards privacy and labour practices, is based on community-wide participation, and offers more educational and research possibilities than commercially available LLMs such as ChatGPT.  

Workshops and expectations 

We did 6 workshops with 10-20 adolescents each. The sessions took place in three youth communities around Delft, Rotterdam and Naaldwijk. We assessed how comfortable they felt with AI (had they used it before, do they trust AI models), what kinds of conversations they want to have with chatbots, and what kinds of functions they want the chatbots to have for health and well being. Based on these insights we developed AI education and a first prototype of an AI chatbot.  

The project team formed strong partnerships with youth organizations and exceeded expectations in recruiting young participants. Significant interest and support were also received from students and community organizations. Throughout this process valuable insights into participatory research were provided and our understanding of effective collaboration and engagement with young people was enhanced. 

Preliminary results:  

• Youth are sceptical about using AI-based chatbots for their mental well-being. They saw these as practical tools, for finding information or learning new skills. Words like annoying, repetitive and lacking emotional intelligence came up when describing them. They had a clear preference for engaging with real people. They saw mental health apps (with) or without chatbots as most needed during social conflict, loneliness, and financial stress, and self-development.  
• If an AI-based chatbot were to be used in a mental wellbeing app, the youth would want the text produced to sound natural. For example, with short texts and a consistent tone. It should give advice, encourage reflection, empower them, and adapt to their personality, although they were wary of sharing private information.  

Conclusions 

Based on these inputs, we can develop design requirements for the co-creation of apps, ensuring to develop a product which works for them. In addition, we are holding interviews with parents, youth workers, and other young people, helping to build a comprehensive understanding of the contexts in which digital wellbeing support is most needed, and best practices to safeguard wellbeing through the LLM. This will support app developers in subsequent stages of the project by aligning system design with lived experience and the specific needs of youth. Their input is crucial, and young people should be active partners in shaping the future of AI development for mental health and well-being.   

Follow-up 

The team is now testing and refining the first prototype of the Digital Health Assistant by collecting conversations between adolescents and AI and conducting interviews with the same adolescents. We are writing several scientific articles, and are seeking funding options to cover additional expenses. 

Caroline Figueroa received the Harkness Fellowships in Health Care Policy and Practice partly based on this research, and will be spending a year at Stanford University to continue research on AI and well-being for young people 

In addition, PhD student Kathleen Guan involved in this project, presented on this topic at various events, including at the Alan Turing Institute in the United Kingdom.