The Emerging Science of Computational Psychiatry
Machine learning, data mining, and artificial intelligence are revolutionizing the study and understanding of mental illness.
Psychiatry, the study and prevention of mental disorders, is currently undergoing a quiet revolution. For decades, even centuries, this discipline has been based largely on subjective observation. Large-scale studies have been hampered by the difficulty of objectively assessing human behavior and comparing it with a well-established norm. Just as tricky, there are few well-founded models of neural circuitry or brain biochemistry, and it is difficult to link this science with real-world behavior.
That has begun to change thanks to the emerging discipline of computational psychiatry, which uses powerful data analysis, machine learning, and artificial intelligence to tease apart the underlying factors behind extreme and unusual behaviors.
Computational psychiatry has suddenly made it possible to mine data from long-standing observations and link it to mathematical theories of cognition. It’s also become possible to develop computer-based experiments that carefully control environments so that specific behaviors can be studied in detail.
How is this new-fangled science influencing researchers’ understanding of mental illness? Today we get an answer of sorts, thanks to the work of Sarah Fineberg and colleagues at Yale University in New Haven.
Fineberg and co review the impact that computational psychiatry is having on the study of borderline personality disorder, a condition that affects almost 2 percent of the population at any time. They show that the field is profoundly influencing the way mental-health professionals study and diagnose this affliction.
Borderline personality disorder is characterized by an inability to form stable relationships, an unstable sense of self, and unstable emotions. People with this diagnosis are significantly more likely to harm themselves, and some 10 percent commit suicide.
The cause of borderline personality disorder is not known. But a wide range of genetic, environmental, and social factors seem to play a role. As a result, characterizing the condition is still a challenge. But computational approaches are beginning to help.
A good example is the computer game Cyberball, which measures social rejection. The game involves three computerized players passing a ball back and forth on a screen. The subject controls one of the players, thinking that other people are controlling the other two. In reality, the other players are computer-controlled.
A key feature of the game is that unbeknownst to the subject, researchers can control how often the subject receives the ball. “By varying the percentage of the time the ball is passed to the participant, feelings of social rejection can be evoked,” say Fineberg and co.
In the most extreme case, the subject passes the ball to another of the players, and they then pass it between themselves for the rest of the game. “This experience elicits sadness and anger in as few as six rounds of play,” say Fineberg and co. That allows researchers to study how these feelings differ between people with and without borderline personality disorder.
It turns out that both groups experience similar feelings, but those with the disorder experience it with much more intensity. More interesting is that people with borderline personality disorder feel excluded when they receive the ball a fair number of times, even when they accurately assess that number. “Negative emotions are reduced, but not fully eliminated, when subjects with BPD receive the ball more times than any other player,” say Fineberg and co.
Virtual reality offers another realm in which behavior can be studied under carefully controlled conditions. In this work, the subject controls an avatar in an immersive virtual environment while interacting with another avatar. This allows researchers to study interpersonal behaviors such as distance regulation, gaze direction, and posture.
Recommended for You
When asked, subjects infer various details about the other avatar’s behavior, sometimes with unusual results. “One group claimed during an experiment that the second avatar was being controlled by the subject’s romantic partner,” say Fineberg and co.
One area where machine learning is having a profound impact is linguistics, and the insights there are beginning to feed into psychiatry. There has long been anecdotal evidence that people with borderline personality disorder use language in certain unusual ways, but quantifying this has been hard. Natural-language processing offers a way.
“We and others have identified language features that mark psychological states and traits,” say Fineberg and co. That is becoming a powerful tool. “Computational models based on word-use patterns can predict which writers have psychosis or will progress to psychosis,” they say.
Beyond this, other computational approaches have the potential to provide a much clearer view of the range of acceptable and unacceptable behaviors.
The hard thing about learning to understand normal ranges of behavior is recruiting a large number of subjects to study. But it has recently become much easier thanks to crowdsourcing services such as Amazon’s Mechanical Turk. In the U.S., Turkers are more diverse than college students, the workhorses for many behavioral studies, although they are not entirely representative of the population as a whole. They can also be re-contacted for follow-up studies.
Because of the vast numbers that can be reached in this way at relatively low cost, such crowdsourcing could change the understanding of mental disorders. “Mechanical Turk may be a good venue for testing research hypotheses in people with borderline personality disorder and [associated] features, especially those that do not present for clinical attention,” say Fineberg and co.
The overall picture they paint is of psychiatry as a discipline in transition thanks to the transformative impact of processing power.
That has significant implications. Whenever revolutions in science occur, there is usually low-hanging fruit to be had. That makes computational psychiatry an interesting place to work and one that ought to be attracting a brightest and the best. Expect to see important new insights as a result.
Ref: arxiv.org/abs/1707.03354 : Computational Psychiatry in Borderline Personality Disorder