Blog by Lorien Abroms, Associate Dean for PhD & MS Programs; Professor of Prevention and Community Health, GW Milken Institute School of Public Health
This blog post reflects on a semester of teaching a new course, Generative AI (GenAI) Applications in Health Promotion, at the George Washington University Milken Institute School of Public Health. As one of the first courses of its kind in a school of public health, it explored both the promise of GenAI for health behavior change such as supporting smoking cessation, improving diet, or promoting sleep and the risks, including bias, safety concerns, and ethical challenges.
The central lesson that emerged is this: public health cannot afford to either fully trust or fully ignore generative AI. Instead, we must adopt a model of responsible experimentation, moving forward while rigorously evaluating, refining, and safeguarding these tools.
Learning by Building
In a class of 25 MPH students, we drew on my work with colleagues at GW’s School of Engineering to develop and test BeFreeBot, a GenAI chatbot designed to function as a smoking cessation coach. Students followed a four-step process to build their own chatbots: conceptualizing the intervention, developing instructions, assembling a knowledge base, and programming using an off-the-shelf large language model like ChatGPT Plus or Gemini 3. Once developed, students also learned to validate and test their own chatbots, including how to ensure that chatbots adhered to their instructions for content and how to surface safety issues and bias before release.
By the end of the semester, all students had created functioning chatbots, some ready for preliminary release. The process revealed that while building a chatbot is now relatively easy, building one that is safe, accurate, and effective is not. Students learned to validate outputs, identify bias, and test adherence to intended instructions, skills that will be essential for the next generation of public health practitioners.
From Skepticism to Qualified Trust
First, trying something (while evaluating!) is better than sitting it out and watching from the sidelines. As I started on the research, I knew that GenAI was being hyped as innovative and a game changer, but I did not trust it as a tool for health behavior change. However, from my research, over the course of two years of testing and revising, I was surprised to learn that chatbots can be reliable. After putting the chatbot through multiple rounds of rigorous testing, and revising and revising it, I can say that my views have changed.
Though I still worry about extreme edge cases that might lead to hallucinations, having seen BeFreeBot improved over time and extensively tested, I trust its ability to function as a reliable quit smoking coach. I have also seen first hand how users of the technology appreciate it and come to trust it with repeated use over time.
Expanding Creativity and Capacity in Public Health
One of the most striking outcomes of the course was how GenAI expanded what students could create without coding expertise. MPH-trained professionals were able to design sophisticated, interactive interventions that would have previously required technical teams.
Students envisioned chatbots to support soldiers in basic training, improve sleep, and help individuals navigate health insurance systems. These tools have the potential to extend reach in resource-constrained public health settings by delivering tailored, responsive support at scale.
At the same time, students confronted a familiar challenge in digital health: engagement. Initial enthusiasm is often high, but sustained use is difficult to maintain. This forced them to think more deeply about behavior change mechanisms such as how to prompt engagement over time and how engagement must be intentionally designed from the beginning.
Adoption Before Evidence?
These developments in Generative AI chatbots raise pressing questions for public health systems. At meetings such as the North American Quitline Consortium, leaders are actively debating whether to adopt GenAI tools now or wait for stronger evidence.
Historically, public health has prioritized gold-standard evidence before widespread adoption. However, some consumer-facing health industries such as weight loss have already moved ahead, deploying GenAI tools at scale and responding to millions of user queries. Quitlines and other public health systems will likely face pressure to follow.
The key question is not whether adoption will happen, but how. Early adoption may increase reach and accessibility, but it also introduces risks related to fidelity, safety, and quality. This reinforces the need for ongoing evaluation and adaptive oversight rather than one-time validation.
Rethinking the Role of Education
At a recent GW Board of Trustees meeting, I was asked what GenAI means for the future of education. The question underscored a fundamental shift: when knowledge and content generation are instantly accessible, the value of education must evolve.
My answer is that in addition to increasing our technology investment, universities should double down on what AI cannot replicate: human relationships, mentorship, and real-world experience. For GW, this means investing in opportunities that leverage its unique location and networks: internships, micro-internships, co-ops, and collaborative projects. These experiences require intentional effort and institutional support, and access to such experiences will increasingly define the value of higher education.
In this context, investment in people—not just technology—becomes essential. Faculty time spent building partnerships and mentoring students is no longer peripheral; it is central to the mission.
In addition, GW’s place as a leader in policy, offers the opportunity for training in the development of policies that can guide the responsible governance of such new technologies, as well as policies that can address the ramifications of the new technologies for jobs and shifts in the labor market that are likely to come.
Moving Forward with Responsible Experimentation and Stewardship
Generative AI is already reshaping public health practice, education, and research. The experience of this semester reinforced a central tension: we must move forward with both urgency and caution. Waiting for perfect evidence risks irrelevance, while uncritical adoption risks harm.
A middle path grounded in responsible experimentation offers a way forward. This means building, testing, evaluating, and refining in continuous cycles, while maintaining strong ethical guardrails. This also means responsible stewardship. While these technologies are privately owned, they affect the public health. Students will need to be trained to see themselves as not only adopters of these new tools, but stewards in the safe adoption of these technologies.