RAISE Health Seed Grants

Advancements in AI have the potential to enhance primary care by streamlining medical tasks and improving diagnostic accuracy. However, these benefits cannot be fully realized if patients, particularly from marginalized communities of color, distrust AI-based healthcare tools. People with limited health literacy and/or English proficiency often exhibit lower trust in both AI and healthcare systems. Failing to consider how people from diverse cultural and linguistic backgrounds perceive the benefits and harms of AI in healthcare risks further eroding trust in healthcare providers (Lee et al., 2021). This project will address the urgent need to understand how people from marginalized, multilingual communities perceive and respond to the introduction of AI-tools in primary care. We will first conduct focus groups and interviews with 120 Black, Latino/a/e, and Asian American participants to explore the barriers to trusting AI in healthcare. We’ll collaborate closely with community organizations to ensure culturally relevant insights. Second, we will develop and test a multilingual psychometric scale to measure trust in AI in healthcare, using responses from three large groups of Spanish and Mandarin-speaking individuals (1,300 people per group).We will leverage our team’s expertise on human-centered AI, primary care among marginalized communities, and developing culturally responsive interventions to develop empirically-based pathways to build trust in AI in primary care so that Americans of all backgrounds can make informed decisions about their health, including regarding new AI tools that can enhance providers’ ability to provide timely, high-quality care.  This project will help healthcare providers build patient trust in AI and support people from marginalized communities in navigating these new technologies.

Ensuring that artificial intelligence (AI) tools in healthcare operate safely and effectively requires robust evaluation within realistic clinical contexts. Traditional evaluation methods often rely on standardized tests that fail to capture the full complexity of patient care, while manually curated benchmark datasets can be both time-consuming and limiting. CuraBench introduces a configurable benchmark generation system designed to create customized synthetic datasets tailored to specific clinical needs, patient populations, and use cases.

The system's innovative approach enables the generation of diverse evaluation scenarios—from assessing how AI systems interpret longitudinal patient histories and assist in diagnosis, to evaluating their ability to accurately summarize medical records. By leveraging real-world healthcare data, CuraBench produces realistic yet synthetic scenarios that can be configured to match the requirements of various medical settings, specialties, and demographics.

By streamlining the creation of comprehensive benchmark datasets, CuraBench makes it both easier and more cost-effective to thoroughly evaluate AI systems and verify the presumed benefits of using AI tools in healthcare. This work represents a significant step toward responsible AI deployment, ensuring that each model is rigorously tested in environments that mirror its intended clinical use.

Without prompt action, the lack of proper AI training among clinicians will risk patient safety and deepen existing health disparities. To address this critical educational gap, we will design and implement a modular curriculum to empower clinicians to use AI effectively and responsibly in patient care. The Improving Medical Practice AI Competency Training (IMPACT) curriculum will be the first program to provide comprehensive, interdisciplinary, competency-based AI training across all stages of clinical education, from preclinical students to practicing clinicians. The curriculum focuses on core AI competencies identified in preliminary research, including AI literacy, ethical and legal awareness, patient communication, and model evaluation for clinical relevance and safety.