Hundreds of millions of people around the world were suffering from common mental health issues including anxiety and depression before Covid-19, and the scale of the health-care crisis has escalated as a result of the pandemic. But demand for mental health services is far outstripping the available supply of trained professionals. Machines are rising to the challenge as a first point of contact for struggling individuals, but just how far can the robot brain go in treating the mind of the human individual?
The research is still in the early days, but as artificial intelligence technology including natural language processing experiences a period of rapid advances, experts confront the delicate issue of how to properly use technology for mental health treatment. One factor is becoming undeniable, though: many people prefer to reveal their mental health struggle to a non-human confidante: a robot.
A recent survey from Workplace Intelligence and Oracle found that across more than 12,000 workers around the globe only 18% prefer humans over robots to support their mental health. Sixty-eight percent prefer to talk to a robot over their manager about stress and anxiety at work, and 80% indicated they were open to having a robot as a therapist or counselor.
"Unbiased information is what people want," said Dan Schawbel, founder and managing partner of Workplace Intelligence.
There are a few primary reasons people are turning to technology for this sensitive conversation: accessibility of getting help 24-7, and getting that help without having to admit to a struggle.
"There really is a stigma behind mental health globally. Talking about stress, or anxiety and depression with managers, employees will hold back. People don't seek help from humans because they don't want to be judged," Schawbel said.
Technology does have the potential to provide mental health support at scale, as well as unbiased information, non-judgmental responses, and a blindness to rank in the workplace context— a machine doesn't discern if the employee seeking help is a CEO or lower down the company hierarchy.
"We're not going to have a billion therapists in the world so we need technology," Schawbel said. "But there is no AI replacement for one of the greatest values that therapists provide: human empathy. Robots can't do that yet."
The use of chatbots in mental health is backed, at least in a general sense, by research already decades-old: people were more likely to be honest using telephone voice-response systems than when talking to a live human.
"Chatbot are OK for basic things," said Bruce Rollman, director of the Center for Behavioral Health and Smart Technology at the University of Pittsburgh School of Medicine, who has researched online treatments for mood and anxiety disorders. "But when we start getting into mental health, it's a game of chess and we're not always playing with Deep Blue," he said, referencing the IBM AI that beat world champions.
"I'd be skeptical if was just computer algorithm, but it might be fine before you talk to a therapist, and AI that does a questionnaire," Rollman said.
Employers expect a wave of mental health challenges for the labor force in the prolonged remote work period, and they realize a chatbot might be preferable for reasons that go beyond the basic stigmatization of mental health, but because employees worry about risking promotions or raises, and job safety. Managers are not always adept at providing the right answers, either.
"Humans are not adequately trained on mental health issues. When people tell you they are stressed or depressed, we often give the wrong answers, and technology is a great way to scale some sequence of questions and best practices," said Emily He, senior vice president of the human capital management in Oracle's cloud business group.
Technology can help guide an employee through a mental health journey in a manner similar to it already does for the onboarding process as a new hire. Conversational AIs, or chatbots, can interface on a daily basis and track answers to questions, in some cases monitoring voice tone as well, and identify and predict someone who needs more advanced treatment.
"The end goal is to enable humans to do what they are best at, which is managing relationships, but there are some baseline questions and great ways to leverage technology," He said.
Record levels of venture capital money are flowing into the sector. According to digital healthcare-focused venture fund Rock Health's proprietary database, $9.4 billion was invested in overall digital health this year, and $4 billion of that was in Q3 alone. Investment in U.S.-based, AI-powered digital health start-ups it has tracked since 2011 are above $10 billion, with investment in mental/behavioral health AI reaching over $230 million across close to 20 deals. And the numbers are getting bigger: in 2020, there was $72 million invested across two sizable mental health AI transactions. The sums recently invested into mental health start-ups, including those not focused specifically on AI, are far higher.
"As an investor in a handful of mental and behavioral health start-ups, we know first hand that our portfolio companies have experienced strong and rising demand throughout the pandemic," said Rock Health CEO Bill Evans. "Like never before, automation and thoughtful use of technologies like AI is absolutely critical to delivering a human touch to those of us in greatest need."
"Eighty percent of the U.S. population owns a phone, and phones can tell you if there's been a change in your behavior," Pitt's Rollman said. He added that while it may sound creepy, predictive analytics are the future across many aspects of our lives, from Spotify knowing what music we prefer to listen to, to maybe mental health. The big gap right now in mental health is an AI that can make the right suggestion at the right time, especially if it is a high-risk person, a person with substance abuse or suicidal tendencies.
Among the AI pioneers pushing the boundaries of what the technology can do is Alison Darcy, founder and president of Woebot Health, a start-up that has engineered a conversational agent (chatbot) to provide digital mental health counseling. Darcy, a research psychologist and former software engineer, worked earlier in her career with Coursera co-founder Andrew Ng at his Stanford University Health Innovation Lab in Computer Science (Ng is the chairman of Woebot).
Darcy said the technology has come a long way in a short number of years and it can help to push people past the primary reason they don't reach out for help: the stigma and fear of being judged.
"Therapists have to spend so much time building rapport. A robot can see me on my worst day and it's just a robot, it's not judging me. They have no judgment," Darcy said. "It's just software."
Woebot refers to itself as a robot in communication with users, and the company chose to not create a human avatar for the interface, though these design decisions also point to the challenge. "There is no human connection, no deep relationship, and that is the limit of the technology," Darcy said.
The limitations, combined with the recent increased investment, worry some mental health experts.
"There is a gold rush in the space and we believe in it, but also believe in science," said Catherine Serio, a clinical psychologist and associate vice president of digital behavior solutions at the University of Pittsburgh Medical Center. "There is lots of money to be made out there."
The University of Pittsburgh Medical Center's Insurance Services Division recently launched a behavioral health app called RxWell which aims for a middle ground between reliance on technology and the need to offer increased access to individuals enabling them to take the first step in seeking help for depression and anxiety.
Serio, who has treated patients with depression and anxiety for years, acknowledged the issue with stigma and scaling in behavioral health, but believes a hybrid model is the only responsible route. "We believe the AI needs to mature more. A chatbot is basically a series of business rules to respond to people and we have not seen a completely coded response that is going to work for people with depression and anxiety," she said.
Building a relationship, a therapeutic bond, has been cited as the reason digital therapy cannot be effective. But Woebot has shown in a trial of young adults the ability to reduce mental health symptoms and deliver cognitive behavioral therapy. "That is not to say it's replacing therapy. It's really not, but it is allowing for full potency of therapy to be unlocked," Darcy said. "Our data shows it can be a useful first step. It is incredibly easy and destigmatizing."
Woebot can also detect crisis language and in those cases it is programmed to tell the user they require a human therapist. Or in other words, the robot's programming is designed to identify its own limitations. "The robot would rather ask you about your mood than detect it. The best person to tell us how you are doing is you," Darcy said.
She said research shows users are also turning to the robot further along in recovery as follow up care. "We see people talking to robots over a long period, maybe for three months when it is a difficult time, and then checking in again nine months later when in another difficult patch. And that is a longer-term perspective than what we normally think about," the Woebot founder said. "Humans respond well to the needs of a patient in the moment. A chatbot can do it too. It is responsive to where a person is at."
The Covid-19 pandemic has resulted in a boom in technology-based health care, and in psychiatry departments, a rare financial feat: an area of care that usually loses money for health institutions becoming a source of profits, according to Soo Jeong Youn, a research psychologist at Harvard Medical School and the Massachusetts General Hospital. She says technology should be used more to combat stigma in mental health, and to provide greater access to care across more cultures and populations, but she added that the research is still preliminary.
"We're not there yet. We're not close to what we see in Sci-Fi movies and responses catered to each person, but the AI has gotten really good," she said.
For example, if a person says they are feeling anxious, the AI can provide resources tailored to anxiety. "Even just searching on 'I'm anxious. What do I do?' There is something to having more information through an app," she said.
The need for help is great, and getting greater, as more Americans have face issues including job loss and food insecurity. Among the client base that UPMC works with to provide health care, the population saying their mental health was negatively impacted jumped from 32% in March to 58% in August. "That one statistic alone is a massive amount of individuals," said Jim Kinville, senior director of the LifeSolutions group at UPMC.
Wellness and prevention apps offering access to skills, and cognitive behavioral therapy techniques, to help people manage stress and anxiety are now widespread. Some apps like Talkspace provide a way to connect with a human therapist online, or combine digital tools with live support, like UMPC's RxWell.
"There is lots of unmet need out there and people we call the 'walking wounded' ... functioning OK, but could benefit more from digital tools," Kinville said. "These tech tools can start the process and get people engaged."
The recent VC deals in the space show that the bets on mental health business models using AI are not limited to the creation of compelling chatbots. Ginger raised $50 million August, including funds from major insurer VC arms at Cigna and Kaiser Permanente, for its on-demand behavioral health platform offering access to coaching, video therapy and self-guided activities. In early 2020, Spring Health raised $22 million for what it describes as "precision mental health care" which uses a proprietary machine learning approach to diagnose conditions and identify the best therapy options for individuals.
While core AI technology including natural language processing underlying chatbots has advanced in the past few years, research shows algorithms continue to analyze the same data sets and come to different results, predicting different outcomes.
That makes Harvard's Youn cautious beyond what she is comfortable saying a chatbot can do today: the equivalent of a first session with a therapist, in which the goal is an understanding of what an individual is going through.
"Hopefully with the push of the pandemic we will get there much faster, and there is huge room and space and need for these chat-based apps to deliver help and relieve distress through the power of AI," the Harvard psychologist said.
For some experts working at the intersection of technology and human performance, choosing a side in the battle between human mental-health professionals and machines risks missing how serious the battle has become and the fact that we need to throw all we have at our disposal at it.
The recent increases in serious mental health struggles, especially among younger adults who say they have felt suicidal, speak to the importance and poignancy of improving access to mental health care, especially for people of color and lower incomes.
"The demands on mental health are massively increasing," said Bill Duane, former Google wellness and performance executive who now runs his own consulting firm. "Existential fear, financial insecurity, nebulous boundaries between work and home ... fear of job security causing people to try and push through and work harder, which only works for the short term. I'm heartbroken at everything going on. But I am extraordinarily optimistic about ways AI can be involved."
Woebot is experiencing increased usage during the pandemic, "huge increases," Darcy said, and it has tracked more need for support among younger users, which the recent research shows to be at elevated levels of risk. At the time of its August deal, Ginger noted "skyrocketing demand" for anxiety and depression care among U.S. workers.
More employees are willing to ask for help because there is a greater shared sense of going through a difficult experience as a community, Oracle's He said, and tech-based support for mental health is a logical extension of how people already interact with machines — fitness apps support better physical health and have edged into other areas of wellness like sleep patterns.
But UPMC's Serio said that once an initial assessment has been done, there is no AI replacement for the "nuance and cues, and all those things human beings do. ... What people need is empathy. Anyone who says that will be fully automated one day, I don't know what reality they are grounded in."
Duane thinks people should not understate the value of how far the technology already has come — not needing to make an appointment or deal with a doctor as a first step, eliminating feelings of shame or discomfort. It is a stigma workaround, a Band-aid on the larger problem of getting more individuals to seek help, but he said it also speaks to the fact that chatbots already are part of the solution.
"To everyone feeling the weight of 2020, it's a reasonable response to what's going on and the massive increase in demand. Access and timeliness are really important when we look at the quantity of people who need mental health care. .... Can AI and chatbots be helpful? Absolutely and partially."