Researchers from Dartmouth can now use AI to analyze a person’s social media history to determine the likelihood of having a mental illness.
All of our insightful, angry, funny and/or pointless social media comments are public information, easily obtained, and in the hands of artificial intelligence, can paint a sorely needed picture of our mental health. 3 researchers from Dartmouth College developed a new AI model that can aggregate, label, analyze, and find trends in our post history. The researchers believe their model can discern the emotion of a post and then use this to chart how our emotions fluctuate over time. The unique pattern that emerges can then be compared to the well-known patterns for bipolar, major depressive, and various anxiety disorders, providing a new tool for a world in desperate need of mental health solutions.
A World in Need
Almost a billion people globally suffer from some form of mental illness. According to research from Our World in Data, the most prevalent illnesses are anxiety disorders (284 million people), depression (264 million people), alcohol abuse (107 million people), drug abuse (71 million people), bipolar disorder (46 million people), schizophrenia (20 million people), and eating disorders (16 million people).
Not only do these people suffer, their lives are shortened as well. A study from the journal JAMA Psychiatry, calculated that their median life expectancy is shortened by over 10 years and that mental illness is responsible for over 14% of global deaths. This comes out to about 8 million deaths per year, an unfortunate figure that could be heavily reduced with proper screening and care, although many areas of the world lack the necessary access and education. The WHO believes “Many mental health conditions can be effectively treated at relatively low cost, yet the gap between people needing care and those with access to care remains substantial. Effective treatment coverage remains extremely low.”
For a world that is increasingly spending their lives online, the new AI model from Dartmouth might help bridge this treatment gap. According to Our World in Data, of the 7.7 billion in the world, 3.5 billion use some form of social media, and all of their comments might be a window into their mental health.
Mental Health Stigma
The gap in treating mental illness is due to prejudice and discrimination. Many people, in both developed and undeveloped countries, don’t understand the causes of mental illness, see it as a sign of weakness, don’t understand its symptoms, believe it’s less severe than it actually is, or believe they or their loved ones don’t need professional care. Unfortunately, these attitudes prevent those in need from seeking and getting help, leaving millions globally to suffer needlessly and die prematurely.
A good explanation of this was published in the journal JAMA Network Open. The researchers interviewed 4129 US participants from 1996 to 2006 and found that only 11.8% believed schizophrenia had scientific attributions, such as genetics. Similarly, the data regarding depression was 13%, and alcohol abuse was 10.9%. Instead of attributing these mental disorders to scientific reasons, the respondents explained them with God’s will, bad character, poor upbringing, random emotional fluctuations, etc. This means that the vast majority of people believed these mental disorders had no scientific origin but were caused by the sufferers’ choices, their parents’ choices, the supernatural, or the uncontrollable.
The same survey was given from 2006 to 2018, and the results were slightly better but still disheartening. Some stigmas improved, some remained stagnant, and others regressed. This prompted the researchers to call for new methods beyond what we can currently provide. “With indications that the level of stigma may be reducing, strategies to identify factors associated with the decrease in stigma for depression, to address stagnation or regression in other disorders, and to reach beyond current scientific limits are essential to confront mental illness’s contribution to the global burden of disease and improve population health.”
If the Dartmouth AI model is as good as the creators hope it is, then it might be the solution we’re looking for, as it can provide a discreet, quick diagnosis away from judging eyes.
The COVID-19 Pandemic
A solution like Dartmouth’s new AI model is needed now more than ever, thanks to the COVID-19 pandemic. Mental health professionals all over the world are reporting a sharp increase in patients seeking help. Valentine Raiteri, a New York based psychiatrist said “I can’t refer people to other people because everybody is full. Nobody’s taking new patients… So I’ve never been as busy in my life, during the pandemic, and ever in my career.” He also noted that many former patients returned looking for his help.
One of the main causes of mental illness since the pandemic began is isolation, due to working from home, lockdowns, quarantines, etc. Raiteri said that many of his patients had feelings of malaise, being lost, and disconnection. Other causes include trauma, stress, depression, and anxiety from losing loved ones, missing important events, losing jobs, etc.
A study from the Lancet calculated just how bad the effects of the pandemic are. By looking at the data from 204 territories and countries, they estimated that 193 million people suffered from major depressive disorder, and the pandemic added an extra 53 million people, an increase of 27.6%, with women experiencing it slightly more than men.
Dartmouth’s AI model is unique because it focuses on emotional transitions or the lack thereof, as “atypical (and problematic) patterns of emotional reactivity and regulation are the main symptoms of emotional disorders.” It’s natural for people to transition between emotions, but those with mental illness demonstrate telltale patterns. People with bipolar disorder have rapid mood swings. People with major depressive disorder change moods very little, as they’re persistently negative or sad. And those with anxiety disorders display excessive fear and anxieties. Despite the topic of the post, all the researchers can discern the emotion of a post and use it to complete somebody’s online emotional profile.
To test their model they used Reddit, a social media platform with nearly half a billion users. They searched the collective history of the platform, which is public information and relatively easy to get a hold of, for phrases like “I have been diagnosed with an emotional disorder,” “I was just diagnosed,” or any possible variations. By doing so they found the users that had self-reported a mental illness, and from here they further subdivided them into 4 groups. “The sample size of bipolar is 686,359, of major depressive is 914,082, of anxiety is 686,369, and of control group is 2,516,696.” By analyzing the posts of these users and comparing them to the control group, the researchers were excited by the model’s accuracy to discern mental illness.
The researchers also noted that their model gets around the problem of “information leakage.” Similar AI models are known to inaccurately label posts with an emotion and a user with a mental illness because the model cannot differentiate between the reasons for the post. For example, if a doctor posts regularly about medications for depression, the AI model might conclude the user is depressed. However, the model from Dartmouth gets around this problem by focusing on transitions between emotions and removing misleading language, such as words “closely related with emotional disorders as well as the names of drugs used to treat emotional disorders.” By doing so, their model out-performed other models.
The researchers are confident AI models like theirs will be the future of mental health treatment, especially in a world torn by pandemics, war, isolation, addiction, and ever-increasing social media use.