John C. Malone Professor Computer Science Mark Dredze has received an Optum Research Award in the area of foundational methods development.

The goal of the Optum Research Awards program is to advance health care by deploying artificial intelligence and machine learning, advanced computing, digital signals, and behavioral and engagement research in the areas of precision health care, health system transformation, and the development of foundational methods for AI in health care. Award recipients will receive up to $100,000 in funding over one year with the possibility of an extension of one to two additional years of funding.

This award supports Dredze’s project on using large language models, or LLMs, to build clinical question-answering systems that preserve patient privacy. With electronic medical records growing in size, it can be difficult for doctors to find the information they need when treating patients. While new AI systems may be able to scan medical records to find the answers doctors are looking for, they must first be trained on real medical records—which are confidential by law. Dredze and his team propose using LLMs like ChatGPT to create pretend examples of questions and answers from medical records to build effective AI-powered question-answering systems that still maintain the confidentiality of patients’ data.