AI coach helps smokers quit via Reinforcement Learning
Research shows how personalizing support increases the effectiveness of AI-based eHealth applications.
Published on March 1, 2025

Team IO+ selects and features the most important news stories on innovation and technology, carefully curated by our editors.
Nele Albers yesterday defended her dissertation at TU Delft on the use of AI coaches in smoking and vaping cessation. She developed an AI coach that uses psychologically informed Reinforcement Learning (RL), a form of machine learning in which a model learns through reward, similar to how humans learn behavior. Albers based her research on insights from behavior change theories and data from three large-scale studies with more than 500 participants.
In addition to the technical side, Albers examined ethical, economic, and psychological aspects and analyzed how different factors contribute to effective behavior change: how to convince quitting smokers, what they are asked to do, and by whom to be supported. Her research shows that AI coaches who take psychological principles into account have great potential to provide adequate support to people who want to quit smoking.
Albers: “My research shows how personalizing support - by considering both a person's current and future state - increases the effectiveness of AI-based eHealth applications. This offers many opportunities for behavior change.”
Personalized AI coaching
AI coaches in eHealth applications could effectively guide people through behavior change while saving costs. Yet these applications are not yet widely deployed. Challenges remain, for example, dropout rates and lack of engagement. Better-tailored support - which considers knowledge, motivation, vitality, thinking patterns - can systematically increase use. To improve the effectiveness of quit-smoking support, Albers examined both the algorithmic side and the interaction between smokers and AI coaches. For example, her model can determine when involving a human coach and the AI coach is useful. She also looked at the tension between smokers' preferences and the advice of health experts. Health experts determine what is right, but smokers often look at this differently. Albers' algorithm tries to find a balance between the two perspectives.
Research findings
A key finding from her research is that support is more effective when it adapts to an individual's situation, such as when suggesting different activities. For example, the AI coach can encourage someone to think about stimuli that trigger the desire to smoke or the person they want to be in the future. Consider a motivation such as, “I want to be a better parent to my child by setting a good example.” In addition, context plays a role in the effectiveness of messages. Depending on the situation, a wording such as “Other smokers find...” work better than “Doctors recommend.” Based on her analysis and findings, she developed models that increase smokers' engagement and better support them in building quitting skills.