Posted by wanz on Friday, June 21, 2024 in News.

I was invited to present as a rising star in the Rising Stars Presentation Session on AI in medicine in the AI for Reliable and Equitable Real-world Evidence Generation in Medicine Workshop (in Conjunction with AIME2024) hosted by the University of Utah on July 9th, 2024 in Salt Lake City, Utah, USA. See the program details here.

Title: Harnessing AI Agents for Ethical and Responsive Real World Data Sharing in Medicine

Abstract: In the latest surge of AI advancements, effective deep learning and large language models that depend on vast amounts of training data are revolutionizing healthcare and medical research. However, insufficient real-world data can result in unreliable and biased machine-learning models. Additionally, sharing health data raises significant privacy concerns. Years ago, I began working on optimizing methods for the responsive and responsible sharing of real-world medical data, emphasizing privacy and fairness. Utilizing a game-theoretic perspective, I identified the best strategies for sharing genomic data, electronic health records (EHR), and voice recordings from patients to third parties or the public. In numerous experiments, I analyzed privacy risks and data utility using real-world EHR data, supplemented with external resources such as voter registration lists and recreational genomic databases. This multi-agent-based approach is fundamentally different as it considers adversarial behavior and capabilities, tailoring protections to anticipated recipients with reasonable resources rather than adversaries with unlimited means. Results indicate that most data can be shared effectively and equitably with minimal privacy risk within a game-theoretic framework. Additionally, I used these risk assessment frameworks to guide the development of data protection methods, such as synthetic data generation, which creates artificial health records based on real-world data.