Older adults still faced interaction challenges with voice assistants

What the study found: The study found that older adults still faced interaction-level challenges with commercial voice assistants, especially around conversation breakdowns and error handling. It also found that technical improvements, including the use of vocal and verbal responses and large language models (LLMs, language systems trained on large text data), may help with preventing and managing errors.
Why the authors say this matters: The authors say their approach helps capture older adults’ real interactions with voice assistants more fully than usage logs or post hoc interviews. The study suggests this can better align future voice assistants with older adults’ expectations and lived experiences.
What the researchers tested: The researchers equipped 15 older adults’ homes with smart speakers and custom audio recorders to collect in-the-wild audio interaction data over one month. Midway through the study, they also deployed a ChatGPT-powered voice assistant to examine how LLM integration changed interaction dynamics during errors.
What worked and what didn't: The study suggests that combining vocal and verbal responses with LLMs’ contextual capabilities can improve error prevention and management. At the same time, the findings indicate that some interaction-level problems remained, particularly those unique to older adults.
What to keep in mind: The abstract does not describe detailed limitations beyond the study’s focus on 15 homes and a month-long in-home study. It also does not provide specific results comparing the ChatGPT-powered voice assistant with the original setup.

Key points

• Older adults still experienced conversation breakdowns and error-handling problems with voice assistants.
• The study used smart speakers and custom audio recorders in 15 older adults’ homes for one month.
• A ChatGPT-powered voice assistant was added midway through the study to examine the effect of LLM integration.
• The authors suggest vocal and verbal responses plus LLM context may help with error prevention and management.
• The abstract says interaction-level challenges remained, especially ones unique to older adults.