An AI Used Facebook Data to Predict Mental Illness

Sharath Guntuku, an assistant professor of computer science at the University of Pennsylvania who was not involved in the research, cautions that, even if these algorithms achieve impressive results, they are nowhere near replacing the role of clinicians in diagnosing patients. “I don’t think there’ll be a time, at least in my lifetime, where just social media data is used to diagnose a person. It’s just not going to happen,” Guntuku says. But algorithms like the one designed by Birnbaum and his team could still play a crucial role in mental health care. “What we are increasingly looking at is using these as a complimentary data source to flag people at risk and to see if they need additional care or additional contact from the clinician,” Guntuku says.

Schwartz notes that diagnosing mental illness is an inexact science, one that could be improved with the addition of more data sources. “The idea is, you’re triangulating mental health,” he says. “Assessing mental health is an exercise that can’t just rely on one single tool.” And since social media provides a continuous record of a person’s thoughts and actions across a substantial period of time, it could effectively complement the hour-long clinical interviews that are typically used to make diagnoses. In such an interview, says Schwartz, “you’re still relying on a patient to recollect everything, to recollect things about themselves. The clinician has to determine when they’re being influenced by desirability biases”—that is, the patient telling their clinician what they think they want to hear. Perhaps, then, social media data could provide a less skewed impression of a patient’s mental state.

Munmun de Choudhury, a professor of interactive computing at Georgia Tech who has previously worked with Birnbaum but was not involved in this particular study, envisions an opt-in social media plugin that could warn users when they may be at risk of mental illness. But such a plugin immediately raises privacy concerns—data about an individual’s psychiatric state, if leaked, could be misused by insurance companies or employers, or force an individual to reveal their mental illness status before they are ready to do so. To work at all, de Choudhury says, the makers of the plugin would have to be entirely transparent about how it handles and secures user data. But, if such an algorithm could detect symptoms of mental illness a year and a half before a patient would typically be diagnosed, it could make an enormous difference in people’s lives. “If we catch these symptoms much earlier on, there could be other mechanisms to alleviate these concerns that don’t necessarily need a trip to the doctor,” she says.

There is already precedent for using social media to prevent mental health crises. “Facebook and Google, they’re already doing this at some level,” Guntuku says. If a user searches for suicide-related terms on Google, the National Suicide Prevention Lifeline number appears before all other results; Facebook uses artificial intelligence to detect posts that may indicate suicide risk and sends them to human moderators for review. If the moderators agree that the post indicates a real risk, Facebook can send suicide prevention resources to the user or even contact law enforcement. But suicide presents a clear and imminent danger, whereas the mere act of receiving a mental health diagnosis often does not—social media users may be willing to sacrifice more privacy to prevent suicide than to catch the onset of schizophrenia a bit earlier. “Any sort of public, large-scale mental health detection, at the level of individuals, is very tricky and very ethically risky,” Guntuku says.

For his own part, Birnbaum sees a less grand, but nevertheless impactful, use case for this research. A clinician himself, he thinks that social media data could not only help therapists triangulate diagnoses but also aid them in monitoring patients as they progress through long-term treatment. “Thoughts, feelings, actions—they’re dynamic, and they change all the time. Unfortunately, in psychiatry, we get a snapshot once a month, at best,” he says. “Incorporating this type of information really allows us to get a more comprehensive, more contextual understanding of somebody’s life.”

Researchers still have a long way to go in designing these algorithms and figuring out how to implement them ethically. But Birnbaum is hopeful that, in the next five to 10 years, social media data could become a normal part of psychiatric practice. “One day, digital data and mental health will really combine,” he says. “And this will be our X-ray into somebody’s mind. This will be our blood test to help support the diagnoses and the interventions that we recommend.”

More Great WIRED Stories

Leave a Reply