The Practical, Robust Implementation and Sustainability (PRISM)-capabilities model for use of Artificial Intelligence in community-engaged implementation science research.

Background: Community-engaged research (CER) leverages knowledge, insights, and expertise of researchers and communities to address complex public health challenges and improve community well-being. CER fosters collaboration throughout all research phases, from problem identification and implementation to evaluation. Artificial Intelligence (AI) could enhance the collaborative process by improving data collection, analysis, insight, and engagement, while preserving research ethics. By integrating AI into CER, researchers could enhance their capacity to work collaboratively with communities, making research more efficient, inclusive, and impactful. However, careful consideration must be given to the ethical and social implications of AI to ensure that it supports the goals of CER. This paper introduces the PRISM-Capabilities model for AI to promote a human-centered approach that emphasizes collaboration, transparency, and inclusivity when using AI within CER.

Methods: The PRISM-Capabilities model for AI includes six components to ensure that ethical concerns are addressed, trust and transparency are maintained, and communities are equipped to use and understand AI technology. This conceptual model is specifically tailored for community-engaged implementation science research, facilitating close collaboration between researchers and community partners to guide the use of AI throughout. This paper also proposes next steps to validate the model using the HEALing Communities Study (HCS), the largest community-engaged research study to date, which aimed to reduce fatal overdose deaths in 67 highly impacted communities in the United States.

Case Study: The PRISM-Capabilities model consists of six components: Optimizing engagement of implementers, settings, and recipients; characteristics of intervention implementers, settings, and recipients; equity assessment and risk management; implementation and sustainability infrastructure; external environment; and ethical assessment and evaluation. Although AI was not initially used during the HCS, we highlight how AI will be leveraged to complete post-hoc analyses of each of the six components and validate the PRISM-Capabilities model.

Conclusion: The application of AI to CER relies on human-centered principles that prioritize human-AI collaboration, allowing for the operationalization of responsible AI practices. The PRISM-Capabilities model provides a framework to account for the complexities of real-world social science problems and explicitly positions AI tools at bottlenecks experienced with conventional approaches.