Co-designing AI with young people to create tailored vaping awareness messages

Youth vaping is one of Australia's most pressing public health challenges. Recent data from the 2022-2023 National Drug Strategy Household Survey indicate that approximately 17.9% of Australians aged 15–24 are currently vaping[1], which is a significant increase from 4.5% in 2019 [1]. While it’s true that vaping generally exposes users to fewer toxins than combustible cigarettes, recreational vaping still carries significant health risks for young people [2].

Public health has responded by implementing mass media awareness campaigns, which can effectively reduce vaping prevalence by shifting young people’s knowledge, attitudes, and social norms. Similar to the success of anti-smoking campaigns, meta-analyses show that exposure to vaping prevention messages leads to higher vaping risk perceptions, increased vaping knowledge and lower intentions to vape [3].

Despite their overall effectiveness, the messaging in campaigns doesn’t resonate equally across diverse groups of young people. This is because most campaigns adopt a "one size fits all" approach, typically only focusing on the health consequences of vaping without accounting for variations in young people’s beliefs, interests, and motivations.

So, could we achieve a greater impact by tailoring vaping awareness messages to match individual characteristics, and could artificial intelligence help us do this at scale?

Tailoring vaping prevention and cessation messages

Tailoring health communication is based on the premise that individuals pay greater attention to, and are more influenced by messaging that speaks directly to their unique circumstances, motivations, and psychological makeup [4]. For example, a systematic review found that intensity-matched culturally tailored smoking cessation interventions increased quit success compared to non-tailored interventions [5]. Likewise, personality-matched tailored interventions, such as anxiety-focused interventions for highly anxious students, demonstrated significantly greater effectiveness in reducing alcohol and tobacco use than standardised interventions [6-8].

Yet, the promise of tailoring has remained largely unfulfilled in public health messaging due to the considerable expertise, time, and resources required to develop tailored messages that appeal to specific individual characteristics [9].

Could AI overcome barriers to tailoring?

Generative artificial intelligence (AI) may offer a way to overcome this barrier because of its ability to rapidly produce large volumes of diverse vaping awareness materials. Recent studies indicate that AI-generated communications can match or even surpass human-created content in terms of perceived effectiveness, clarity, and quality [10,11].

However, due to the inherent biases in the datasets used to build AI algorithms, the generated messages can be highly variable, sometimes producing factually incorrect information or messages that fails to effectively engage certain groups [12]. This is why we must co-design with diverse youth groups and maintain human-expert oversight to ensure public health communication is accurate, relevant, and inclusive.

What are we doing?

With funding from the Medical Research Future Fund (#MRF2031246), our team has been working directly with diverse groups of young people to co-design AI-generated vaping awareness materials [13]. In a recent randomised experimental study, we compared how young people perceived the effectiveness of these co-designed materials against materials by official public health agencies. To build on these results, we will implement a randomised controlled trial to test whether exposure to tailored multi-themed materials is more effective at deterring youth vaping than one-size-fits-all materials.

By leveraging the potential of AI and youth co-design, we may overcome the limitations of traditional one-size-fits-all approaches, extensive resources associated with tailoring, and biases associated with AI systems to develop a scalable and tailored campaign with materials that target multiple themes for a sustained period to achieve maximal effects in deterring youth vaping. We also hope that this project can serve as a model for using AI effectively across a variety of public health campaigns targeting other critical health behaviours and reaching underserved groups.

Dr Tianze Sun presented the invited NDARC Webinar ‘Can we use AI to solve a problem like youth vaping?’ in October 2024.

[1] Currently vaping includes people who reported vaping daily, weekly, monthly or less than monthly.

In this issue