Researchers in South Korea have built an artificial intelligence system that can spot signs of daily stress by studying how people use their smartphones. The work comes from the Korea Advanced Institute of Science and Technology, known as KAIST.
KAIST announced on October 15 that a team led by Professor Uichin Lee of the Department of AI Computing developed the technology. It is called “SHIRBT-MMF.”
The system tracks repeated phone behaviors. It looks at how long and how often each behavior happens, along with text-input patterns, to judge a person’s stress level for the day.
How the AI Reads Smartphone Habits
Past studies on phones and stress often looked at screen time, how many times apps were opened, and typing habits. But when apps or their designs change, the data changes too, which can make long-term tracking less reliable.
The KAIST team took a different path. It focused on what a person does, not which app they use.
For example, if a user moves from KakaoTalk to LINE or WhatsApp, the activity is still “messenger chatting.” The analysis is based on the purpose of the action.
The team called these repeated actions “Smartphone Human Interaction Routine Behavior Tasks,” or SHIRBTs. It used its own pattern-finding method to spot screen changes that repeated over several days.
A large language model then gave these patterns simple labels such as “web browsing,” “financial payment,” and “reading/writing email.” The system sorted the same behavior correctly across different apps more than 95% of the time.
To protect privacy, the system did not save any text or message content. It only recorded whether text was typed, how often, and how screens changed, while contact details were hashed so they could not be directly read.
What the Study Found
The team studied real-life data from 26 university students and office workers in their 20s and 30s. Sixteen short-term participants gave about 30.5 days of data on average, while 10 long-term participants gave about 105.1 days.
Researchers found 29 types of repeated behaviors in total. Each person repeated about seven behaviors on average.
Personalized AI models were trained on each person’s patterns and daily self-reported stress. The best model was correct 75.9% of the time for short-term users and 73.8% for long-term users.
Those results were 5.5 and 3.5 percentage points higher than the older method of studying app usage with the same AI technique. Even using “messenger chatting” alone beat the older approach.
“We viewed the smartphone not simply as a device for measuring usage time, but as a space in everyday life where an individual’s communication, information seeking, and financial activities are repeatedly carried out,” Professor Lee said.
The researchers noted that the study shows a link between behavior and stress, not a cause. They plan to test the system on more age groups, jobs, and phone types, since this study only included Android users.
The research was published in December 2025 in the journal Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies. It is being presented at ACM UbiComp/ISWC 2026 in Shanghai, China, which runs from October 13 to 15, 2026.
