CAREER: Toward Improving Scalability for Effective Health LLMs

The rapid adoption of large language models (LLMs), which use deep learning to process natural language, in aspects of healthcare has the potential to revolutionize critical processes, including clinical decision support and administrative tasks. However, these applications pose significant challenges in ensuring scalability and effectiveness of use. Addressing these challenges is essential to prevent modern AI tools from perpetuating or exacerbating ongoing issues. This project seeks to improve the effectiveness of healthcare-focused LLMs by developing scalable methods for evaluating and mitigating incomplete data. The research will empower healthcare providers and decision-makers with more robust AI systems, fostering trust and improving outcomes for all. Additionally, the project integrates educational initiatives to promote responsible AI principles among students, professionals, and the public, contributing to the national interest by advancing healthcare delivery and technological progress. This award focuses on two complementary research objectives: evaluating effectiveness in LLMs and mitigating identified data issues The evaluation framework will incorporate evidence-based datasets and novel techniques to address factual and faithfulness hallucinations, ensuring truthful and transparent LLM outputs. On the mitigation front, the project will introduce innovative reinforcement learning methods to align LLM outputs with effective principles. The work includes developing preference optimization techniques and flexible inference-time approaches, providing practical tools for responsible AI deployment in high-stakes environments. Furthermore, the project will implement a comprehensive educational strategy, offering an interdisciplinary course on human-centered AI, engaging a wide range of students, and creating accessible public educational resources on AI's societal impact. By bridging research and education, this project advances the responsible use of LLMs in healthcare and beyond, addressing critical fairness issues at the intersection of technology and society. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.