Some luminaries like former Google CEO Eric Schmidt expect that one day, loquacious AI will provide companionship for the roughly 40% of elderly people who say they regularly experience loneliness. If this vision comes to pass, it’d be no less than transformative from a wellness perspective — loneliness has been found to increase the likelihood of mortality by 26%, and lonely people have a 64% increased chance of developing clinical dementia.
That’s perhaps why researchers at the University of Rochester investigated interactions between older adults and an AI-imbued digital avatar. As they explain in a paper published on the preprint server Arxiv.org (“Discourse Behavior of Older Adults Interacting With a Dialogue Agent Competent in Multiple Topics“), subjects who interacted with the agent regularly over the course of weeks had stronger sentiments about topics concerned with goals rather than routine activities, as well as stronger self-disclosure for more intimate topics.
They say that this suggests avatars could provide “valuable practice” and coaching to help older adults navigate a challenging conversation and improve both their health and quality of life. “We investigate how users’ verbosity, sentiment, and self-disclosure behavior depend on the topic under discussion and the avatar’s tone, and how they evolve with time,” wrote the coauthors. “Our participants were more engaged with the agent when the conversation topics were more emotionally intense and intimate.”
The scientists studied linguistic data automatically transcribed from 80 sessions over three to four weeks. Nine volunteers conversed with the avatar for 15-20 minutes in three subsections containing 4-5 questions on a specific topic (e.g., pets, family, weather, driving, chores, education, and technology), and then they filled out surveys and received feedback on their non-verbal behavior and speech prosody (rhythm and sound).
The topics were suggested by subject-matter experts and categorized into three groups based on their emotional intensity: easy, medium, and hard. “Easy” topics were the kind likely to be broached in making someone’s acquaintance, while harder ones called for more self-disclosure.
During the analysis phase of their research, the researchers looked at verbosity — specifically, differences across different sessions, users, and topic classes — and changes in verbosity over time. They next examined sentiment for sessions, tone change, and self-disclosure cues, in addition to metrics like word count per turn, and positive and negative emotions.
They report that although the avatar’s tone remained almost the same for all classes, people tended to use stronger tones when they talked about medium and hard topics compared with easy ones. They also noted that hard topics contained more words per turn and more negative and positive emotions and drives, but that people used personal pronouns more often in easy topics, like when they introduced themselves or talked about their activities.
“We observed that people tend to talk more when the topics are more intimate, such as life goals and the challenges of getting older, where they also use stronger emotion words … Furthermore, the average response length increases as people progress along the series of interactions,” wrote the team. “These results support the use of dialogue agents with older adults in the context of difficult conversation topics. Our participants were more engaged with the agent when the conversation topics were more emotionally intense and intimate.”
They leave to future work larger studies involving other age and culture groups.