Can a chatbot be your therapist?

In this talk, I will present the results of a study we conducted at Stanford about the behavior of chatbots during mental health conversations. With a team of therapists, we examined how various types of AI chatbots respond to messages during sensitive conversations and compared them to responses that human therapists gave. While chatbots can be empathetic and reassuring, they tend to have a directive tone without asking for enough contextual information. Concerningly, most chatbots did not respond adequately to crisis situations and did not refer users to helpful crisis resources. When looking at the real-world use of AI companions like Replika or Character AI, there is a real danger of people, especially adolescents, getting attached to and overreliant on these chatbots. With this talk, I want to encourage developers of chatbots, legislators, and users of chatbots to reflect on how we can use chatbots in positive ways to improve our mental health while minimizing their potential negative impact.