Peterson said they can already identify certain clusters of language that do raise flags, for instance, those that surround notions such as loneliness or agitation. And when multiple clusters appear, there’s greater concern. But “this is not some kind of scanning technology,” Peterson told NBC News. “It’s far more complicated.”

In the new phase of the project, the foundation established by the initial 300 patient records will be tested against a much larger population of volunteers, with data coming from a variety of apps that will continuously upload the subjects’ social media and mobile phone interactions. While there’s no minimum number of participants, the project hopes to enlist up to 100,000.

The accumulated social information — safeguarded by HIPPA standards of medical privacy and stored in the Geisel School of Medicine at Dartmouth’s secure onsite database — is analyzed by computer programs developed during the earlier phase. None of the data is shared with third parties, and the information is objectively safer than any stored on Facebook, OKCupid or Edward Snowden’s flash drive. Similar to many medical tests, this stage is only for observation. There will be no diagnosis or intervention, though participants are welcome to drop out of the study at any time.