Conversational Agents for Dementia using Large Language Models

Abstract

Conversational robots are a type of social robot that emphasize verbal communication. One of their main application areas centers on assisting people with dementia and their caregivers, particularly in dealing with problematic behaviors. Recent advances in Large Language Models (LLMs) have increased their potential in developing conversational assistive agents. In this paper, we explore the use of a LLM (ChatGPT) to conduct cognitive stimulation therapies with people with dementia. We compare a conversational robot which includes personalized hand-coded conversational strategies with one using an LLM and prompt engineering. We conclude that the use of LLMs produces engaging conversations and has the potential to facilitate caregiver customization of social robots for dementia.