Skip to main content

We use necessary cookies that allow our site to work. We also set optional cookies that help us improve our website. For more information about the types of cookies we use, visit our Cookies policy.

Cookie settings

Having conversations about mental health can be extremely challenging. We very quickly recognised that Shout Volunteers often found it difficult to build their confidence. To address this, we built an LLM (large language model)-based, generative AI chatbot trained on the anonymised Shout dataset, which can simulate Shout conversations, for volunteers to use during their training. More than 1,000 people have now used this generative AI solution to help them build their skills.

Many of the concerns raised about LLMs and generative AI are of less significance in this scenario. For example, what happens if it generates nonsensical text? In this case the volunteer can simply refresh the message. The key challenge, not trivial, is for the typical performance of the chatbot to remain good enough for the volunteers, and in turn the texters, to benefit.

"This is a powerful example of an effective use-case for generative AI. We can see widespread opportunities, beyond Shout, to use generative AI to train people to hold better conversations about mental health."

Dr Mark Ungless.png

Dr Mark Ungless, Director of Data Insights