A Risk Taxonomy and Reflection Tool for Large Language Model Adoption in Public Health

Area of Science:

• Public Health
• Artificial Intelligence
• Health Communication

Background:

• Large language models (LLMs) offer potential but raise concerns for public health information dissemination.
• Existing risk assessment frameworks for LLMs are insufficient for the high-stakes public health domain.

Purpose of the Study:

• To explore concerns regarding LLM adoption in public health.
• To develop a risk taxonomy for evaluating LLM use in critical public health areas.

Main Methods:

• Conducted focus groups with public health professionals and individuals with lived experience.
• Examined concerns across infectious disease prevention, opioid use disorder, and intimate partner violence.

Main Results:

• Developed a risk taxonomy with four dimensions: individual, human-centered care, information ecosystems, and technology accountability.
• Identified specific risks and provided reflection questions for practitioners.

Conclusions:

• LLM adoption in public health requires a risk-reflexive approach.
• A shared vocabulary and reflection tool are needed for collaborative risk assessment and harm mitigation.
• Emphasizes the importance of lived experience and domain expertise in evaluating LLM applications.