A research project is trying to stop suicides by using AI to analyze at-risk people's smartphone and Fitbit wearable data and identify periods of high danger
drawn from electronic medical records as well as scores of other factors — are used to assign patients a risk score, so that individuals at exceptionally high risk can be provided with extra attention.” https://www.nytimes.com/... Tim Hogan / @timinhonolulu : Can Smartphones Help Predict Suicide? Check out this article from @nytimes. Because I'm a subscriber, you can read it through this gift link without a subscription. https://www.nytimes.com/...
Context & Ripple Effects
Suicide detection has been moving up the stack for a decade: Facebook started by linking troubling posts to trained experts in 2015, then drew expert criticism for its opaque approach to flagging possible threats to police worldwide. Meanwhile, a study of depression and suicide-prevention apps found many skipping best practices, some even listing wrong hotline numbers — so the field's credibility was already shaky before passive sensing arrived.
This project is the next step past both: instead of waiting for a user to post or open an app, AI infers risk continuously from smartphone behavior, Fitbit wearable signals, and electronic medical records, assigning scores so clinicians can concentrate attention on the highest-risk individuals. That makes Fitbit — already being folded into Google accounts and Google Health branding — a medical-adjacent data source, not just a fitness tracker.
First-order effects
- Patients flagged as exceptionally high risk receive extra clinical attention based on a score computed from their phone use, wearable data, and medical records — intervention shifts from self-reported crisis to algorithmically scheduled outreach.
- Fitbit's role changes for its users: the device on their wrist is now feeding a suicide-risk model, which raises the stakes of its ongoing migration into Google's account and health infrastructure.
Second-order effects
- The privacy critique that hit therapy apps like Better Help for notifying third parties when a patient has suicidal thoughts now extends to hardware: wearable makers supplying behavioral data for risk scoring face the same third-party-sharing scrutiny.
- Facebook's experience shows what happens when platforms act on inferred distress — experts questioned whether opaque flagging to police was accurate, effective, or safe — so any deployment of these risk scores outside clinics will inherit that accuracy-and-accountability debate.
Third-order effects
- If passive sensing proves out, mental-health triage structurally shifts from user-initiated help-seeking to continuous ambient monitoring, making consumer devices de facto diagnostic instruments and forcing regulators to decide who may act on a risk score — and who gets told about it.
- The precedent echoes COVID-era smartphone tracking, where researchers tested monitoring apps amid privacy concerns: public-health surveillance built on consumer sensors tends to normalize quickly once one validated use case exists, and suicide prediction could be that case.
The trend: Mental-health intervention is shifting from reactive hotlines and self-reported app check-ins toward continuous AI inference from consumer smartphone and wearable data, with clinicians rather than platforms as the intended actors on the output.