Center for Healthy Minds Researchers Use AI and 7.5M Screenshots to Understand Suicide Risk

"We centered this specific part of our work on smartphone use because phones are where so much daily life happens," Jacobucci says.

Researchers not only collected screenshots, but they also combined them with more traditional measures of wellbeing, including having participants report their sleep habits and complete emotion-based check-ins.

Jacobucci, who aims to bridge the critical gap between complex algorithms and practical clinical applications, worked with CHM data engineer Wenpei Shao to use AI to analyze and find patterns in the visual and textual parts of the screenshots using both vision-language models (VLMs) and large language models (LLMs). Crucial to analyzing large amounts of data, VLMs are similar to LLMs but add "sight" to LLMs' linguistic capabilities, making them able to recognize patterns in complex visual data.

"We're building tools right now that I couldn't have imagined five years ago," says Jacobucci. "They are fully automated systems that can learn from how someone uses their phone. These analyses wouldn't have been as sophisticated without these new tools - there's no way."

The team's analyses relied on a system known as "screenomics." The approach helps paint a consistent and frequent picture of study participants' digital habits, allowing researchers to get deeper insights than a simple "screen time" metric

Jacobucci and Ammerman's work supports growing evidence that in an increasingly digitalized world, not all screen time is the same.

This study is the first of its kind to use screenomics to examine adults' nighttime smartphone use to predict risk of suicide, Jacobucci says.

While the findings are a first step, Jacobucci acknowledges that there's still much to learn, noting that the study only included Android users who had recently experienced suicidal thoughts. These findings need to be tested in a broader group, he says, including people who haven't recently had suicidal thoughts, to truly guide prevention efforts.

Jacobucci and Ammerman say future research should also look more closely at why some types of middle-of-the-night phone activity - especially active engagement like typing - might be linked with fewer suicidal thoughts. Understanding which apps or kinds of content make a difference could help shape better support tools.

"In the long run, I hope our work helps save lives and makes life a little more livable for people who are struggling by helping support show up earlier, more compassionately, and in ways that actually fit day-to-day reality," Ammerman says.

By Victoria Vlisides, CHM digital content editor

If you or someone you know is struggling or in crisis, help is available. In the U.S., call or text 988 to reach the Suicide & Crisis Lifeline.