To follow

Abstract

Building AI chatbots for addiction counselling: lessons from smoking cessation

The rapid proliferation of generative AI chatbots has made machine-delivered, highly-tailored addiction counselling a reality. In this talk, Felix will describe the learning gained from developing Nova, a chatbot that delivers routine smoking cessation support to pregnant people who smoke as part of the NIHR-funded eSupport programme. Nova is the central feature of a pregnancy-specific version of the Smoke Free app, where we have translated the NHS-adopted Standard Treatment Programme into a digital stop smoking service by developing a set of new pregnancy-specific features. Our experience has shown that building a chatbot that is engaging, adherent to guidance, clinically competent and safe is challenging. Furthermore, configuring a chatbot to deliver structured, multi-sessional counselling, and going beyond the more typical question-and-answer style of interaction, introduces further complexity.

Felix will discuss key lessons the team have learnt whilst co-developing Nova with experts and people with lived experience. These include: integrating rule-based and generative chatbot capabilities to balance support delivery fidelity with conversational flexibility; addressing misalignment between pre-trained model knowledge and UK clinical guidance through retrieval-augmented generation; developing and testing safeguards; and linking the chatbot with other app features to enable coordinated support delivery and enhance user engagement.