Your smartwatch already knows how much you have moved today. But what if it could look at your recent activity and tell you that you are about to spend too long sitting?
Scientists at the Icahn School of Medicine at Mount Sinai have developed an artificial-intelligence system that can forecast upcoming periods of prolonged sedentary behavior in women living with chronic pelvic pain disorders. The idea is not to simply remind someone to move after they have already been sitting for a long time, but to predict the period in advance and deliver a reminder at a more useful moment. Learning from everyday movement They analyzed wearable-device data from 134 women with chronic pelvic pain disorders, most commonly conditions such as endometriosis. A comparison group of 61 healthy participants was also included. Participants wore Fitbit devices for as long as 90 days. The devices continuously collected information related to physical activity, heart rate and sleep, producing minute-by-minute records of how participants’ activity changed throughout the day.
They then used approximately 10 days of each person’s data to build personalized forecasting models. These models attempted to predict activity levels roughly one hour into the future. The system could then use those predictions to identify potential 15-minute sedentary bouts during waking hours. That window could provide an opportunity for a small movement break, what they describe as an “exercise snack.”
The surprising part: complicated AI wasn’t necessary
Artificial intelligence often brings to mind enormous neural networks and powerful computers. But this study produced an interesting result: relatively simple, interpretable models performed about as well as the more computationally intensive deep-learning approaches tested that matters for wearable technology. A model that does not require substantial computing power could potentially run directly on a phone or wearable rather than constantly sending personal activity information to a remote server. That could reduce computational requirements while also supporting a more privacy-conscious approach to health monitoring. They also found that the forecasting system remained useful when some wearable data were missing, a realistic problem when people remove their devices or forget to synchronize them.
Chronic pelvic pain can make physical activity difficult. Pain, fatigue and other symptoms may contribute to periods of prolonged sitting, making generic advice to simply “move more” difficult to apply. A personalized system could work differently. Instead of sending repeated reminders throughout the day, a future wearable might recognize an individual’s activity pattern and provide a short prompt when a prolonged sedentary period appears likely. But there is an important distinction: this study demonstrated prediction, not a proven health benefit.
They have not yet shown that receiving these AI-generated prompts will actually reduce sedentary time, decrease pelvic pain or improve quality of life. Those questions require prospective clinical trials. The team is now working toward a “just-in-time” intervention that will test whether personalized movement prompts can translate the prediction into a meaningful change in daily behavior. The larger idea is intriguing. Instead of a wearable merely telling us what our bodies have already done, future devices could begin anticipating what we are likely to do next and intervene at precisely the moment when a small change might still be possible.
