Want to quit smoking? Try QuitBot, an AI-powered companion

QuitBot features an avatar named Ellen, who checks in with the user a few times a day to offer praise, encouragement, and support. Ellen also shares progress charts and, like a skilled coach, teaches users strategies to stay motivated and manage cravings effectively.

“This is the first conversational chatbot for quitting smoking that uses large language models (LLMs). There are many quit smoking tools that use menu-based chatbots or text messaging but their interactivity and engagement with people are quite limited. These are like one-way conversations. QuitBot offers users the opportunity to have two-way conversations,” Dr Jonathan Bricker, the creator of QuitBot, told The Indian Express.

Bricker is a professor and director of the Health and Behavioral Innovations in Technology (HABIT) Lab at the Cancer Prevention Program of the Division of Public Health Sciences at the Fred Hutchinson Cancer Center in Seattle.

“Advances in LLMs inspired the development of QuitBot. We are reaching an inflection point in digital conversational abilities which are making it possible for people to connect with tools that can provide useful and scientifically based advice for improving their health behaviours,” Bricker said.

“The AI of QuitBot is the LLMs we have fine tuned for quitting smoking. We have a large library of knowledge that trained the QuitBot to understand a person’s questions and provide clinically useful answers,” Bricker said, pointing out that by design, QuitBot does not collect any personal data.

Developed by Fred Hutch’s Health and Behavioral Innovations in Technology research group, which has more than 20 years of experience in creating and analysing innovative smoking cessation and behavioural health tools, QuitBot took six years of iteration, collaboration and testing.

Initial funding to develop the tool was provided by the National Cancer Institute. The Fred Hutch researchers then collaborated with Microsoft AI for Good Lab, using their AI expertise to refine the conversational chat function.

Offering step-by-step guidance

QuitBot utilises a form of AI known as LLM natural language processing cloud computing, combined with the principles of effective coaching, to build a personalised connection with its users.

“What makes the virtual coach effective is the step-by-step guidance it provides people, walking them through the process of quitting smoking through a series of brief (2-3) daily conversations. Being available 24/7, always encouraging the users, and being highly knowledgeable contribute to its effectiveness,” Bricker added.

The key behavioural science principles that the bot employs are ones that teach people how to be aware to be open to their cravings to smoke, rather than trying to fight them, he said.

QuitBot reminds people why they are quitting, what triggers them to smoke, and offers help for what gets in their way. When people have setbacks, QuitBot provides help for dealing with urges and inspirations to quit that are tailored to the user’s unique needs.

Clinical outcomes and success rate

QuitBot is a comprehensive 42-day quit smoking program of 3- to 5-minute focused conversations over two phases: 14 days pre-quit date and 28 days post-quit date, covering topics ranging from motivations to quit, setting a quit date, choosing approved medications, identifying and coping with a wide range of triggers, and recovering from lapses or relapses.

QuitBot demonstrated high user engagement and promising cessation rates compared to the National Cancer Institute’s SmokefreeTXT text messaging programme, a mobile text messaging service designed for adults in the United States who wish to quit smoking, particularly among those who viewed all 42 days of program content.

When language bias is a challenge

“Sometimes the QuitBot has trouble understanding someone’s question about quitting smoking. When this happens, the QuitBot requests that the user ask the question in a different way. This rephrasing of a question is usually effective. And we continue to improve on the LLM of QuitBot so that it gets better at understanding what someone is asking,” Bricker said.

Talking about the biases, if any, for the app, he said, “The way we addressed the problem of language bias is that we trained the bot on data provided by diverse groups of people who are trying to quit smoking, so it is designed to reflect the way that people ask questions and kinds of questions they are concerned about.”

“I think a key enhancement is to develop conversational chatbots that are tailored to specific languages and cultures, and the way people speak within those language communities. We are starting to do this now with our newest project which will be tailoring QuitBot to indigenous peoples,” Bricker said.

A tool to change addictive behaviour

Asked about AI-powered bots being leveraged for other forms of addiction, Bricker said, “QuitBot is an overall platform for changing an addictive behaviour (in this case, cigarette smoking) that can be adapted for other addictive behaviours such as alcohol or substance use. The structure is the same as setting a goal, identifying triggers, learning skills for coping with triggers, identifying motivations and remaining motivated, preventing and recovering from lapses. Transfer would involve switching the knowledge content area to the specific substance.”

AI-driven platforms and behavioural healthcare

“I think AI-driven platforms have the potential to vastly increase people’s access to behavioural healthcare. They are low-cost options that can be used to overcome the stigma of seeking behavioural treatment. I could see AI being a low-barrier first point of entry into treatment. Those who do not respond to AI-based treatment can then be stepped up to treatment by a healthcare provider. This approach recognises two factors, which are one, there will never be enough trained providers to offer treatment to those who need it, and two, for many people an AI treatment will be sufficient. Taking this view allows healthcare systems to focus on providing more costly and resource-intensive treatment to those who need it most,” Bricker explained.

However, he added, “AI will never be able to replace the skill and compassion of a well-trained health provider. The future is about changing the contexts in which clinicians provide healthcare.”