“Wanna talk about it?”: Designing conversational AI interactions to support lonely young adults

Abstract

Loneliness, especially when a chronic issue, can have a large effect on an individual’s mental health. The reasons and contributing factors vary, and dimensions of loneliness, including chronicity, permanence, and intensity, can be different for each person and over time. Research has shown that cognitive behavioral therapy has had a moderate effect size for many populations of lonely people. This research seeks to explore how to give people the most effective support by pairing them with established CBT-based interventions based on their experiences and the causes behind their loneliness, based on psychometric connections.

To do so, interviews with nine non-clinical lonely people between the ages of 18 to 40 were conducted to identify needs and perspectives on loneliness. A set of LLM-based conversational interactions was developed that provided interventions depending on the participant’s connection to personas based on profiles identified in the interviews. A further five participants tested these CA interactions and their opinions were connected to the profile they were assigned based on their psychometric data.

Results suggest that lonely young adults have high expectations for CA interactions and, when reached, appreciate the ability to discuss their situation and reflect on their thought patterns. Preferred conversation patterns are discussed and connected with strategies to implement them through LLM system prompts. The profiles employed in this research were not comprehensive enough to offer meaningful insights, but qualitative differences inform user expectations and preferences.