San Francisco Chronicle - (Sunday)
AI app created to process shooting trauma
Kai Koerber leaned back in his booth as he glanced around Caffe Strada, a popular tree-shaded hangout just off UC Berkeley’s campus.
“This has always been the spot,” he said.
Less than a year ago, Koerber wasn’t so different from those college students sipping iced lattes as they discussed their postgrad plans. A data-science major who grew up in New Jersey and attended high school in Florida, he wanted to emulate his entrepreneurial idols, create the next great app and change the world.
What made him unique now was that he is already doing it. Just 12 months removed from his UC Berkeley graduation, Koerber, 23, is traversing the globe to hobnob with fellow CEOs and speak at tech symposiums about his latest feat: the Joy AI Wellness Platform, which uses artificial intelligence to help people improve their mood.
The app gauges how users feel from the mere sound of their voice. In addition to suggesting quick mindfulness activities based on people’s perceived mental state, it charts their mood throughout a given week.
Like many other AI apps, Joy’s readings become more accurate the more users speak into it. Koerber should know. Over the past couple years, he has used Joy’s emotion-recognition technology to help him process lingering trauma from the worst day of his life.
As a junior at Marjory Stoneman Douglas High School in Parkland, Fla., Koerber was in band class on Valentine’s Day 2018 when he asked his teacher to use the restroom. But one of his classmates was out, so he was told to wait until she returned.
Koerber would later credit that stroke of circumstance for helping to save his life. Had his teacher let him leave when he asked, he would have been in the same hallway as former student Nikolas Cruz when Cruz began firing his semiautomatic rifle indiscriminately at students and teachers.
Cruz killed 17 people that day, making it the deadliest high school shooting in United States history. Asked how the experience has shaped him, Koerber paused for a moment, shook his head and wiped a bead of sweat from his brow.
“I think that’s why I’m so steadfast in what I believe is my own purpose now,” he said. “When you survive a tragedy, it’s kind of hard to not think, ‘Hey, I should live the rest of my life exactly the way I want to.’ ”
“For me, it just happens to be making positive change and making sure people can deal with the mental trauma that comes from being a part of such horrific events.”
It was a sunny Tuesday evening in late April. Koerber had just flown in from the Fort Lauderdale, Fla., area, where he lives with his mother, Alana, while managing Joy’s 14-person staff over Zoom.
Almost every employee is a UC Berkeley undergrad. Over the next few days, Koerber would attend a research symposium and lead team-building activities before jetting to Europe for more meetings.
This is all part of what he calls his “Steve Jobs era.” Nearly half a century after Jobs famously worked late nights building Apple in his parents’ Los Altos garage, Koerber often crashes at his family home in Coral Springs, Fla., around 3 a.m., exhausted from 14-hour days trying to build something seminal.
When Henry Cen — a UC Berkeley sophomore — told Koerber last month that he was overwhelmed juggling his work on Joy with a rigorous course load, Koerber gently reminded Cen: That was me last year. Throughout college, Koerber had retreated to his bedroom after class, slogged through schoolwork until around midnight, then stayed up until 5 a.m. reading AI research papers.
“When you talk to Kai, you can just see his passion,” Cen said. “He makes us all believe we can do something special.”
Koerber’s goal was ambitious by any standard: to create the world’s first wellness app that can meet the emotional needs of its users simply by deciphering the emotion in their voices. But unlike many other forwardthinking undergrads, Koerber had the luxury of learning from one of his chosen field’s foremost experts.
Dacher Keltner is a UC Berkeley psychology professor whose award-winning research on the science behind emotion has made him a sought-after advisor for tech companies trying to break into the mental health and wellness area. During his first sit-down with Koerber five years ago, Keltner was struck by the thenfreshman’s clarity of purpose.
“Kai is a great example of this new generation of college students,” Keltner said. “He came in wanting to learn about emotion, happiness and AI. But more than anything, he wanted to change the world.”
In the immediate aftermath of the shooting at his high school, Koerber had done numerous interviews — including one with Trevor Noah on “The Daily Show” — about the need for better gun-control laws in America. Often lost in those conversations was the need for better mental-health support for young people.
This seemed odd to Koerber, who had once shared a first-period class with Cruz. Even then, Koerber could tell his schoolmate was unwell. There was how Cruz talked to himself and rarely interacted with peers; dragged his feet while walking between classes; carried his books in a plastic bag because school officials had banned him from wearing a backpack.
“There were warning signs,” Koerber wrote in an opinion piece for CNN.com four months after the shooting. “Collectively we either missed them or chose to ignore them.”
That’s when Koerber started to become a prominent mental-health advocate. He wrote more opinion pieces for nationally renowned publications, developed a 400-page mental health program, and even founded a nonprofit to get that curriculum into schools.
By the time he moved cross-country for UC Berkeley’s top-ranked computer science program, Koerber was reconsidering his career plan. Long set on following his uncle into rocket science, he now dreamed of building an app that could help people like himself see better days.
“He was always getting different software sent to him around AI and emotional intelligence around AI,” said Mohammed Zareef-Mustafa, Koerber’s freshman-year roommate. “Other kids our age were all stressed about the smallest things, and he was over here trying to find a solution.”
For more than two years, Koerber worked with other UC Berkeley data-science majors to design an algorithm that could pinpoint emotion just from bioacoustic signals and vocal expressions. Sad people, for example, tend to speak in a softer, slower and more monotonic voice. That’s different from anxious people, who often talk faster and with more labored breathing.
But as hundreds of digital mental-health tools flooded the marketplace in recent years, Koerber saw the danger of diagnosing disorders through an app. Artificial intelligence remains a work in progress. In the case of something like Joy, developers had no way to fully eliminate the potential for misinterpretation.
What if people are only speaking quietly into the app because they’re in a loud place? Would the algorithm misread those soft voices as signs of sadness?
The more a person uses the app, the better chance it has of picking up such nuances. It also helps that Joy’s stakes are relatively low. Rather than trying to diagnose depression, anxiety or other mental-health issues, it provides benign suggestions — like cleaning the house or going for a walk — to help improve mood. And if users disagree with the mood the app assigns them, they can change it.
Koerber has used meditation, Joy’s weekly mood charts and many other resources to help him cope with post-traumatic stress disorder. Though he called that “an ongoing process,” he believes he is in a much better place than he was a few years ago.
“The scope of what a technology is taking on does matter a lot,” said Colin Walsh, an associate professor of biomedical-informatics at Vanderbilt University who has studied the use of AI in mental-health and wellness apps. “Joy is meant to be an additional resource to help individuals who ideally also have a human provider who can help them through things. It’s not a cure-all; it’s a resource.”
It might also be the beginning. Though Koerber is pleased with Joy, he views it as his entry point into what he hopes is a long, successful career in AI. His company, Koer AI, has been working on another app — the details of which remain under wraps — that he hopes will be even more groundbreaking.
For Koerber to really make this his Steve Jobs era, he has to keep toiling until early in the morning. At least now he doesn’t have to worry about schoolwork.
After finishing the interview at Caffe Strada, Koerber got up from his booth, checked his watch and smiled.
“Right on schedule,” he said. “It’s time for my 6 p.m. lunch.”