Konsep Inti
The authors investigate the challenges faced by older adults when interacting with commercial voice assistants, focusing on error handling and conversation breakdowns. They propose leveraging Large Language Models (LLMs) to enhance error prevention and management in voice assistants for older adults.
Abstrak
The study explores the interactions of older adults with voice assistants over a month-long period, highlighting challenges in error handling and conversation breakdowns. By recording audio data "in-the-wild," the research provides insights into user behaviors and dynamics. The findings suggest potential improvements using LLMs to enhance user experiences with voice assistants.
The study addresses the limitations of current voice assistants for older adults, emphasizing the need for improved error management and understanding conversational breakdowns. By deploying ChatGPT-powered Alexa skills, the research explores enhanced interaction quality and identifies opportunities for future VA design considerations.
Key points include:
Challenges faced by older adults with commercial voice assistants.
Exploration of Large Language Models (LLMs) to improve error handling.
Deployment of ChatGPT-powered Alexa skills for enhanced interactions.
Detailed analysis of errors, user behaviors, and strategies for recovery.
Insights into usage patterns, trends over time, and peak usage periods.
Overall, the study provides valuable insights into optimizing voice assistant interactions for older adults through advanced technologies like LLMs.
Statistik
24.76% of Alexa turns resulted in errors during interactions.
Intent recognition errors accounted for 32.3% of all errors.
Only 25.47% of errors were resolved immediately by participants.
Kutipan
"Commercial Voice Assistants are not designed for special populations such as older adults."
"Our work suggests leveraging vocal responses combined with LLMs’ capabilities for enhanced error prevention."