A change in appearance has caused a stir in the industry, but it was recently discovered that artificial intelligence can determine whether a person feels lonely with 94% accuracy.
According to reports, American AI researchers used AI analysis tools such as the IBM Watson supercomputer to test loneliness in elderly individuals.
By analyzing words, phrases, and pauses in silence during interviews, AI assessed symptoms of loneliness in elderly individuals nearly as accurately as when they filled out self-report questionnaires. It also found that lonely individuals tend to take longer to answer direct questions about loneliness and express more sadness in their responses.
The study suggests that most research simply asks, “How often do you feel lonely?” which may lead to biased responses. This study used natural language processing—an objective quantitative measure of expressed emotions—combined with traditional loneliness assessment tools.
What makes this tool interesting is that it doesn’t just rely on specific words indicating fear but examines patterns in language used during responses.
Experts note that the U.S. has seen an “epidemic” of loneliness in recent years, marked by rising suicide rates, opioid use, declining productivity, increased healthcare costs, and higher mortality. A study earlier this year found that 85% of elderly residents in independent living communities experience moderate to severe loneliness.
The COVID-19 pandemic and subsequent lockdowns worsened the situation, increasing loneliness. Researchers wanted to explore how natural language processing and machine learning models could predict loneliness in elderly community residents.
The study focused on 80 independent residents aged 66 to 94, with an average age of 83. Trained researchers conducted semi-structured interviews with participants between April 2018 and August 2019 (before COVID-19).
Participants answered 20 questions from the UCLA Loneliness Scale, using a four-point rating system to respond to statements like: “Do you often feel left out?” and “Do you often feel part of a group of friends?”
These topics were also discussed in private conversations, which were recorded and manually transcribed. Transcripts were analyzed using natural language processing tools, including IBM Watson Natural Language Understanding (WNLU), to quantify emotional expression.
The WNLU system uses deep learning to extract keywords, categories, sentiment, grammar, and metadata. Machine learning allows systematic examination of long interviews across multiple subjects to identify subtle linguistic markers of loneliness.
Human emotional analysis can be inconsistent and requires extensive training for standardization. In contrast, the AI system predicted loneliness with 94% accuracy compared to UCLA Loneliness Scale scores.
AI predicted self-reported loneliness with 94% accuracy and quantified loneliness (based on UCLA scale scores) with 76% accuracy. Lonely individuals took longer to respond in interviews and expressed more sadness when answering direct questions about loneliness.
The study also found gender differences: women were more likely to admit feeling lonely, while men used more words related to fear and joy, suggesting they experience emotions more intensely and may feel freer expressing them.
Subtle gender differences appeared in emotional expression when describing loneliness. The research highlights discrepancies between loneliness assessments and subjective experiences, which AI systems can help identify.
Researchers suggest there may be a “language of loneliness” that could help detect loneliness in elderly individuals, improving clinical and family assessments to better address their needs—especially during isolation.
Current research explores how natural language features of loneliness correlate negatively with wisdom in elderly populations, meaning wiser individuals may feel lonelier. Verbal data could be combined with cognitive, motor, sleep, and mental health assessments to improve understanding of aging.
The study compared AI accuracy with self-reported loneliness. While loneliness scales may not always reflect true emotions, combining AI and self-reports could enhance diagnostic accuracy for psychologists.
Although the UCLA Loneliness Scale remains popular for avoiding the word “lonely” and minimizing gender bias, researchers aim to develop more precise assessment tools.
