Psychological well-being impacts humans’ ability to enjoy their lives, but most people can’t recognize mental health warning signs unless they consult a professional psychologist. A Silicon Valley-based startup has now put the process into an AI-powered chatbot.
Woebot is designed to identify, evaluate and treat mental health conditions such as depression, anxiety, and stress disorders based on Cognitive Behavioral Therapy (CBT), one of the most widely used mental health treatment frameworks. Woebot’s recently released IOS app features an animated robot character that communicates with users using natural language processing (NLP) technology.
Woebot Labs was founded by Stanford School of Medicine Psychologist Alison Darcy with the goal of making mental health evaluation accessible and even entertaining. Woebot is the first AI system to compete for a share of the mental health market, which in 2015 totaled $196 billion in the US alone.
“We now understand that you need to look after your physical fitness everyday. I think people are reaching the understanding that actually mental health is something that you should also look after and actively seek self-care opportunities,” says Dr. Darcy.
While Woebot Labs is an early-stage startup, the company has garnered much attention in the AI community since AI guru and Coursera co-founder Andrew Ng joined Woebot’s board as Chairman, assisting Dr. Darcy with growing the team, and advising on data and machine learning strategies.
“Andrew was telling me that he doesn’t get involved in things unless he thinks they can be potentially the biggest contribution of his life. That’s a huge bonus for us, and a very nice thing that he thinks we can be that impact,” says Dr. Darcy.
Woebot functions like a mental health mentor. It prompts users to check in daily, then asks about their temper, activities for the day, motivations and frustrations and so on. The questions are inspired by CBT, whose methodology is to treat mental illness by changing unhelpful patterns in behavior, emotional self-regulation, and thoughts.
The chatbot also encourages users to try new activities such as meditation, which is a common CBT therapeutic strategy. Nick Panchyshyn, co-Founder of smart life management and AI Assistant startup LifeMap, gave Synced a positive review of his Woebot experience. “I enjoyed a lot of conversations that were really engaging. Woebot was also offering me different activities in a very nice manner.”
To enable automated responses that are both accurate and natural, Woebot uses a combination of structured dialog along with natural language understanding (NLU). Woebot Head of Engineering Joe Doyle says the team uses both deep learning and NLU to understand users’ moods and activities. Based on their mood, Woebot can guide users to appropriate content.
“We utilize the FastText algorithm for some classification as well as applying some internally developed solutions based on popular machine learning frameworks such as TensorFlow. We also rely on many different techniques ranging from the use of simple Regular Expressions all the way to complex, deep neural networks,” says Doyle.
Woebot is now engaged in more than two million conversation per week, and Dr. Darcy has discovered interesting behavior patterns, for example, in how users express distorted thoughts, such as “I’m not smart enough” or “I shouldn’t have said that.”
Woebot includes quizzes designed to educate users about cognitive distortions and distorted thoughts, and help them identify negative words and correct cognitions. “That is why NLP is going to be so fundamental to transform mental health care,” says Dr. Darcy.
Last June, Dr. Darcy and her colleague Kathleen Kara Fitzpatrick published a randomized trial of Woebot in the Journal of Medical Internet Research. They recruited 70 people who self-reported depression or anxiety, and had them either use Woebot or read a mental health e-book for two weeks. Test subjects using Woebot experienced a higher reduction in depression then the book group.
This certainly does not mean Woebot is effective enough to replace professional therapists. John Torus, Chair of the American Psychiatric Association’s Smartphone App Evaluation Work Group, said in a Washington Post interview that “these things can work well on a superficial level with superficial conversations. Are they effective tools, do they change outcomes and do they deliver more efficient care? It’s still early.”
Dr. Darcy does not want to exaggerate Woebot’s capabilities. “Woebot is a guide and that is a legitimate role in the field of mental health. Woebot is not going to be able to deliver you therapy but he can certainly ask you the right questions to help people figure out things on their own.”
Woebot still needs improvement: it remains only an IOS application, has relatively poor conversational capabilities and a narrow spectrum of clinical skills. But Dr. Darcy is dreaming big: “I would like Woebot to become a household name, known as a reliable effective helper that you can reach out to at any time in daily life.”
Journalist: Tony Peng | Editor: Michael Sarazen