The new WBM artificial intelligence learns from your behavioral data—not from heart rate or blood oxygen levels.
A recent study supported by Apple reveals that users’ behavioral data—such as activity levels, sleep, and physical movement—can offer far more accurate insights into human health, even more so than traditional metrics like heart rate or blood oxygen saturation. This research led to the development of a foundational model called WBM, which is trained solely on such behavioral data.
Unlike previous models that relied on raw sensor data, the WBM AI model uses processed information like step count, walking stability, physical fitness level, and other lifestyle-related indicators—all of which are continuously tracked by the Apple Watch.
Researchers state that behavioral data is more suitable than real-time sensor readings for detecting chronic or temporary health conditions such as pregnancy or sleep quality, as these metrics vary over longer periods and reveal more meaningful trends.
Instead of analyzing raw signals, WBM is trained using 27 interpretable behavioral metrics and a modern architecture known as Mamba-2. This approach has led to significantly improved predictions of various health conditions compared to traditional models.
In tests, WBM outperformed other models in most health-related tasks and, when combined with sensor data, achieved a 92% accuracy rate in detecting pregnancy. These findings suggest that integrating AI with behavioral data paves the way for smarter health monitoring.
