Social media is rapidly transforming the way we talk about public health. The steady increase of social media content and users makes it easier to disseminate health information, but it also speeds up the spread of misinformation — undermining the expertise of healthcare professionals.
That’s where Sijia Yang comes in. He’s an associate professor with the School of Journalism & Mass Communication (SJMC) who’s pursuing pioneering research on the use of AI to improve public health communication, especially in underserved communities lacking resources to create effective campaigns.
“Social media and the rise of AI also provide opportunities for people to care about the quality of health-related information,” Yang says. “It’s easier for citizens now to be involved in health-related discussions or discourses.”
Yang’s research group — Computational Approaches and Message Effects Research (CAMER) — has been leading initiatives to identify effective messaging strategies to promote public health. For example, Yang has researched the rising online marketing promoting recreational cannabis use, particularly edibles, among younger demographics as well as potential cost-effective educational interventions such as cannabis warning labels. He found that many individuals are drawn to campaigns comparing marijuana to candy, such as chocolate or gummies. That said, cannabis warning labels enhanced with compelling visuals and specific health risk statements can help counter such youth-targeted marketing.
Currently, the research team is collecting and analyzing data on AI to improve the ease and quality of informative vaccine messaging in rural communities.
During the COVID-19 pandemic, a group of researchers affiliated with the UW-Madison Prevention Research Center and the UW Institute for Clinical and Translational Research, including Yang, received funding from the Centers for Disease Control and Prevention to investigate health messaging and its effectiveness for vaccine promotion in rural communities throughout Wisconsin.
“The COVID-19 pandemic reminds us of the urgent need for health messages tailored to local communities. Part of the problem is that many of these local clinics don’t have the resources and money to hire a professional communication liaison, so it’s forced back on staff members who are already overwhelmed and overburdened,” Yang says.
While the staff at these clinics wanted to invest resources into creating persuasive healthcare messages to promote vaccines in the local communities, they did not always have the means to do so, Yang says. With AI, it is possible for clinics to analyze thousands of campaign messages, create persuasive social media posts, and identify how to maximize social media reach without the need for extensive human work.
“Now you can leverage those tools and combine that with profound knowledge about community needs, so it’s essentially much easier to create potentially persuasive and effective campaign messages,” Yang says.
From the study, Yang and associates found that vaccine messaging was most successful when there was a personal testimonial connected to the campaign, and when the message came from a clinician versus a community leader.
Once this information is obtained, CAMER uses Meta to determine which version of AI-generated campaign messages is most likely to maximize social media reach. Meta’s algorithms use data on demographics, interests and social media trends to determine what audience is most likely to click the link at the right time. To double check the validity of the most successful visual messages identified by Meta, CAMER tests the messages in a national survey experiment before releasing it on social media platforms.
These social media posts are seen by people in rural communities throughout Wisconsin, providing citizens with accurate information on vaccines while reducing human labor.
“A lot of our group’s work is applied in nature, and we think about how to do it in a way that’s engaging the relevant communities and providing useful results back to the communities,” Yang said. “We want the results from the research project to benefit the community.”
To test AI’s power in helping researchers and health campaign managers identify effective message design strategies, Yang’s group has been implementing AI to sift through thousands of creative tobacco control messages to identify effective visual persuasive appeals. AI can then pinpoint which message features and design principles are the most effective. For example, as proof of concept, AI was able to automatically uncover effective features such as using testimonials and statistical evidence without human supervision.
Yang was recently awarded a research grant as a co-investigator from UW–Madison’s Office of the Vice Chancellor for Research which is supported by the Wisconsin Alumni Research Foundation (WARF) as part of the Research Forward initiative to foster innovative and collaborative projects. With the grant, Yang will collaborate with SJMC’s Dhavan Shah , the Jack McLeod Professor of Communication Research, Maier-Bascom Chair, director for the Mass Communication Research Center and research director for the Center for Communication and Civic Renewal, as well as principal investigator Professor Timothy Rogers from the Department of Psychology and co-investigator Jerry Zhu, a professor of computer science.
The purpose of this grant is to build an AI Terrarium, which will simulate human opinions and responses to messages and mimic real small-group human interactions. The team was awarded this Research Forward grant to investigate how to program and deploy AI personas to build a useful simulation system. Once developed, one of the AI Terrarium’s potential applications includes screening a large number of candidate public health campaign messages before testing with real human participants, which can be costly to implement.
Since this project is considered high-risk and high-reward, funding would be hard to obtain without the Research Forward grant. However, with this grant, researchers can simulate human behavior and identify possible applications of the AI Terrarium to improve public interest communication, among other potential use cases.
“By understanding and improving the fidelity of AI-powered ‘digital twins’, we aim to create a network of interacting AI agents to simulate human opinion dynamics and message expression and reception processes,” Yang says.