Dr. Schueller is the Executive Director of PsyberGuide and an Assistant Professor at Northwestern University’s Feinberg School of Medicine. He is a faculty member of Northwestern’s Center for Behavioral Intervention Technologies (CBITs) and his work focuses on increasing the accessibility and availability of mental health resources through technology.
I’ll admit my bias upfront, I am completely bullish on the potential of artificial intelligence to transform mental health. This is not because I believe that your therapist will be replaced by a machine anytime soon. The automated therapist, ELIZA, developed in the 1960s showed us the limitations of machines to produce meaningful therapy-like conversations. Despite all of the advances in technology since then we still see “conversational artificial intelligences” making mistakes that are wholly unacceptable in the context of mental health. Failures to identify acute emergencies as such, responding to sensitive topics with nonsensical replies, and an inability to understand and adapt to emotional cues. The reason, however, that I’m excited about the potential of artificial intelligence is that I believe it will offer a new type of intelligence into mental health treatments that can supplement the strengths of human care. I also think artificial intelligence, especially conversational versions that can respond in plain speech can accelerate the development of digital mental health interventions as they will no longer be confined by visual heavy user interfaces (like websites and apps are) and can instead rely on text- or speech-based interactions that better mimic the processes of therapy.
Adam Miner along with other colleagues from Stanford University tackled this issue in a thoughtful piece recently published in JAMA.1 Miner and colleagues note that the gap is closing between what technology is capable of and how we imagine a human might respond in a similar situation. They also point out that closing this gap might not be necessary, as realism – defined as mistaking the conversational artificial agent as a real human – may be overrated. Again, if the notion is that such conversation artificial intelligences should be able to replace therapists we would want them to be able to act like therapists and do everything that therapists can currently do. However, if we think that artificial intelligences could benefit mental health services in a unique way then our question should be one of worth and value rather than the human-like quality of the interaction.
Two recent products incorporate conversational artificial intelligences to create digital mental health interventions: Woebot and Wysa. Both products are “chatbots” drawing from principles of cognitive-behavioral therapy. Woebot is accessible through Facebook Messenger and Wysa is available through Facebook Messenger, Android, and iOS (and reviewed on PsyberGuide!). Each product avoids direct comparisons with a therapist through the use of a non-human persona. Woebot plays the part of a robot developed to provide emotional support. Wysa is a “pocket penguin” design to compassionately support behavioral health. Cognitive-behavioral therapy is an efficacious treatment delivered face-to-face and in various Internet formats including websites and mobile apps. As such, both products can be considered “evidence-informed.” Woebot, additionally, has been subjected to a randomized controlled trial in 70 college students comparing the Woebot “chatbot” to a mental health ebook.2 Participants were given two weeks to use their intervention and assessed again on depression and anxiety. Participants who used Woebot showed significant reductions in depressive symptoms over the two weeks and were about 2.5 points lower on the PHQ-9 at the end of the study. It is worth noting that this is neither a clinically important difference (which would be 5 points) nor did people on average obtain remission. Although that should be weighed against the fact that this as a two-week intervention where a typically dose of face-to-face cognitive behavioral therapy would be 16 sessions and Internet cognitive behavioral therapy would be about 8 weeks. This was a small study with several other limitations but nevertheless it shows that such resources can have value within mental health especially if they can be scaled to reach people who might not otherwise receive treatment.
Another reason I’m excited about conversational artificial intelligences specifically is that they can help foster diversity in digital mental health interventions. I’m hoping we see a range of different products developed, guided by science and improved over time that can truly combine best practices in mental health and technology to create engaging and impactful products. One day we might all have our own personal robots, or penguins, or any of a sort of these class of new conversational artificial intelligences to help support our own emotional and behavioral health.
- Miner, A. S., Milstein, A., & Hancock, J. T. (2017). Talking to Machines About Personal Mental Health Problems. Jama, 318(13), 1217-1218.
- Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Mental Health, 4(2), e19.