Should AI chatbots give mental health advice?

Should AI chatbots give mental health advice?

More cases of problems with the use of AI chatbots as ‘therapists’ are being reported as people look for immediate sources of mental health advice and feelings of being listened to and supported. People are becoming more used to ‘chatting’ with AI software, and see chatbots as a practical alternative to long waiting lists for NHS talking therapies. The NHS itself provides access to patients with lower level mental health concerns through the Wysa AI software.

But there are now more examples of where conversations have gone wrong. In the US, a man committed suicide after an AI chatbot persuaded him it would be a way of helping to ‘save the Earth’ from man-made environmental impacts. One chatbot, Character.ai is currently the subject of a legal case with the parents of a 14-year-old boy who was allegedly encouraged to take his own life by the software. The National Eating Disorder Association had to suspend use of its chatbot after it started to recommend limiting calorie intakes.

The ‘Yes’ service

Academics warn of flaws in the way AI chatbots are ‘trained’ on the basis of large language models. They ‘learn’ from general models of what mental health is, what ‘good’ mental health looks like, and its opposite, and imitate patterns of typical conversations.

Professor Hamed Haddadi at Imperial College London describes AI chatbots as being like inexperienced therapists — and a particular issue is how they are designed to keep people chatting, and so will be relentlessly supportive: “so even if you say harmful content, it will probably cooperate with you”.

At the same time, the use of chatbots is being seen to provide instant reassurance and positive impacts for some. A Dartmouth College study showed a 51% reduction in symptoms of depression among those with anxiety, depression or an eating disorder compared with a control group.

Hard limitations

EAPA UK chair, Karl Bennett said: “Not everyone has a straightforward option when it comes to sharing worries and thoughts, or doesn’t feel able to share personal issues with friends or even family. AI chatbots are an easy answer. But the limitations are obvious — they can’t provide human reassurance, a human presence or the same kinds of human accountability. And we know it: in a recent YouGov survey, only 12% of people thought an AI chatbot would make a “good therapist”.

“Ai chatbots can be trained and programmed to take on specific roles, and that can include becoming more sensitive to mental health conditions and the nature of mental health conversations. And most importantly, ‘understand’ its limitations and the need to step back from advice and signpost to genuine and professional sources of mental health support. “The use of chatbots only highlights the huge importance of EAP services and access to human contact, human professionals and their experience.”