The Washington PostDemocracy Dies in Darkness

Can a video game tell if you’re depressed?

An AI-driven start-up uses games to detect danger signs of mental illness. It’s a novel approach, but not everyone is sold.

A screenshot of the Thymia game being played by Emilia Molimpakis, co-founder of the start-up. The business says it has found a fresh way to help diagnose depression. (Thymia)

A few years ago, Emilia Molimpakis had a destabilizing experience when a friend who regularly saw a psychiatrist attempted suicide.

“What I couldn’t grasp was she had seen her psychiatrist just two days before,” recalled Molimpakis, who found her friend. “Why did he not see this coming?”

Rather than rail against the health system, Molimpakis took another route. A postdoctoral neuroscience researcher at the University College London, she decided to leave academia and launch a company. Her goal was to use entrepreneurship, and artificial intelligence, to fill in gaps in mental health treatment. She reasoned that since a limited supply of doctors didn’t have the time or, possibly, even the raw processing power to solve a growing depression epidemic, perhaps technology could step in.

And in the most unusual twist, she decided to use video games for the task.

The result is Thymia, a simple set of mobile video games that Molimpakis and her co-founder say might, with the help of AI, pick up on depression signals as well as any office examination. When it begins clinical trials later this spring, Thymia will try to improve and even save lives as it alerts doctors to warning signs they might otherwise miss.

According to the founders, the tool — which comes as part of a larger movement to deploy AI to address mental health — could revolutionize how society cares for depression. The start-up joins a host of related start-ups using digital tools to broaden access to mental health care, including wearables company Fitbit, which recently filed for a patent on a new mental health detection system, and TalkSpace, which primarily uses a data-driven approach to match consumers with therapists.

But some experts who were asked about Thymia worry that machines are an inadequate and potentially risky replacement for a profession that requires high levels of human interpretation, echoing broader concerns of doctors that the “app-ification” of mental health could sacrifice quality for the sake of cost and easy access.

Psychiatric evaluations pose an innate paradox: The only way to really know how someone is feeling is to ask the person directly, but they can also be the person least likely to hold a clear view on the subject.

Traditional psychiatry has tried to crack this by combining a subjective firsthand report with ostensibly objective questions. The long-running standard for depression diagnosis is the “PHQ-9,” a set of nine questions about how a patient is feeling. Do they have “little interest or pleasure in doing things?” “Poor appetite or overeating?” The patient ranks themselves on a scale of “not at all” to “nearly every day.” If five of the nine answers are positive, they are clinically diagnosed with depression.

The creators of Thymia say there’s a better way.

“There’s a rich body of mathematical tools that we haven’t been using,” said Stefano Goria, who co-founded the company with Molimpakis and serves as its chief technology officer. “What we can do is extract information that we may not understand naturally so we can put objective numbers on it and make a more informed decision. We can get a lot more information.”

There is an urgent need for fresh approaches: Americans’ mental health has worsened in the past several years as technology, seeking to aid humans with communication, has increased their atomization. The isolation brought on by the pandemic has only deepened the problem. The Boston University School of Public Health recently found that rates of depression among Americans — at 8.5 percent before the pandemic — now stand at nearly quadruple that, at 32.8 percent.

Video games are an unlikely solution. While many gamers have no noticeable positive or negative psychological effects from playing, a landmark Brigham Young University study revealed that “a significant minority” of gamers “can become truly addicted to video games and as a result can suffer mentally, socially and behaviorally.”

How friendships thrived in video games during the pandemic

Thymia execs, however, say that the game works well because it creates a natural setting for a subject to spin off data; a person immersed in a video game is less likely even to think they are taking a test. (Thymia is, in this regard, a distant cousin to EndeavorRX, the prescription video game used to treat attention-deficit/hyperactivity disorder in children.) Such a tool also won’t have trouble attracting users — 68 percent of American adults now play video games, according to the trade group the Entertainment Software Association, up 6 percent over the course of the pandemic.

The Thymia games, available free for clinicians from its site, are minimalist. They involve simple tasks in whimsical natural settings; in one, a player tries to track bees buzzing around sets of flowers. But the machine is gathering critical information.

As a player swipes in the game, or moves their eyes, or changes their expression or speaks, the AI is recording the data. It is measuring for markers like agitation, anxiety, attention and working-memory. (Agitation might seem a difficult metric to gauge in the stressful confines of a video game, but can be done, Molimpakis says, via jumps or dips in areas like psychomotor speed.)

The algorithm processes this data and compares it both to baselines for a given user and the expected reactions for a user’s demographic. If, say, it perceives working-memory degrading over time, or eye movements too rapid for a person’s age group, this is factored in. It calculates all it tracks and produces a score for each symptom, which it sends to a clinician — at the moment only a patient’s mental health specialist, not a primary care doctor. (Patients choose the data to be shared via a pop-up opt-out every time recording will be used; they can allow voice but not eye-tracking, for instance, or vice versa.)

The doctor then makes treatment decisions — even, in emergency situations, sending people to the hospital. Data is not shared with the patient.

In doing things this way, Thymia execs say alarms are sounded that would have remained silent in a traditional office-visit structure.

“It’s increasingly hard to get in to see a psychiatrist because it’s so expensive,” Molimpakis said. (In the United States, the average psychiatrist makes $105 an hour, according to the Bureau of Labor Statistics, although that number is usually much higher for private consultations.) “Even then you may only go in to see them once every four weeks or six weeks. And who knows what’s really happening to you in that time?” she said.

A novel MTV program is trying to make programming more mentally healthy

Some doctors familiar with depression and technology, however, ask if the service will really provide all the information that a clinician needs.

“There’s no doubt that generally speaking AI and technology hold promise for mental health and are really the future of it,” said Murali Doraiswamy, a professor of psychiatry and professor in medicine at Duke University School of Medicine.

“But psychiatric disorders like depression are very personalized,” he said. “Symptoms are extremely variable. … I’m not sure that I’ve seen the evidence that a computer can be programmed to understand all this the way a person can.”

Mental health advocates also worry generally that cheaper app-based approaches to mental health could deter insurance companies from paying for human doctors. Thymia founders say that this is one of the reasons patients will not be allowed to use the service themselves and instead must go through their clinician. Molimpakis says that Thymia does not share data with third parties, including insurers.

Thymia has 12 employees but aims for 17 by summer; it has raised $2.6 million in early-stage funding, Molimpakis says, and will conduct a seed round later this spring. About 2,000 people have used the product in preclinical trials, with all-important clinical trials expected to begin at a number of U.K. medical institutions in the coming months.

Executives have already begun working with British regulators to have Thymia games approved by the end of the summer, hoping to classify them as a medical device that would allow formal usage in doctors’ offices. Europe and the United States, Molimpakis said, should follow soon after.

Backers say they see a big social upside.

“As investors we asked ‘is there a massive problem to be solved and is this the right team to be taking it on?’ ” said Patrick Newton, a principal at the London venture-capital firm Form Ventures, which has invested an undisclosed sum in Thymia. “We answered yes to both. The idea of giving a clinician the ability to remotely and regularly check in and understand how someone’s condition is performing could be huge.”

But some say they worry about the pitfalls in the move from a first-person to third-person — or, third-machine — approach to diagnosis.

“Objective markers can provide better scores than self-reported scores in a lot of cases, there’s no question about that,” said Liam Kaufman, co-founder of Winterlight Labs, a Canadian start-up that uses AI to detect voice changes and aid in Alzheimer’s diagnosis. “The problem is there are a lot of ways you don’t know that it’s really objective. For instance, when you play games over and over, you get better at them. So are you measuring symptoms or are you measuring gameplay?

“We can’t even treat many cancers right now,” he said. “And the brain is arguably much more complicated.”

It’s unclear whether Thymia execs ultimately intend for the tech to become a diagnostic tool all on its own. While Molimpakis and Goria take pains to emphasize it is meant as a supplement — “all we’re doing is extracting all this information that would come naturally to a clinician and providing it back to them,” Goria said — the founding story involving Molimpakis’s friend suggests that the company is premised on the idea that the tech sometimes just may be able to do the job better.

Thymia executives foresee a future point when results could be shared with primary care physicians and possibly even patients. Such decentralization makes physicians like Doraiswamy wary. The panic over Googling a physical condition is bad enough, they say; mental health information for patients could be that much more dangerous.

“If you were trying to treat someone with a phobia, would you really want an app that spotted 40 different ones going off at all hours of the day?” he asked.

The battle over Thymia and related tech is part of a larger conflict between an industry that says it resists technology not out of stodginess but human skillfulness, and a group of disrupters who believe the machines in some cases may simply be more effective.

“Many psychiatrists have the strong feeling they should trust their own instincts,” Molimpakis said. “I understand that. But I think objective measurements can be very important. We should be fighting to make them better.”