AI evolution in biomedical and health informatics Abstract: I will review the progression of AI systems in medicine, from early knowledge-based systems focused on narrow domains to highly supervised data-driven models, to the current evolving state where generative AI systems are expected to be accurate even for problems they were not trained to solve.
Adoption of AI in medical centers is high, but post-marketing surveillance systems to continuously evaluate their performance are lacking. Understanding the privacy risks of medical LLMs is still an evolving research topic.
The goal of this presentation is to initiate discussions on the responsible use of generative AI in medicine. Biography: Lucila Ohno-Machado, MD, PhD, MBA, is the Deputy Dean for Biomedical Informatics and the Chair of Biomedical Informatics and Data Science.
As Deputy Dean for Biomedical Informatics, Ohno-Machado oversees the infrastructure related to biomedical informatics research across the academic health system. Biomedical Informatics and Data Science serves as the hub for biomedical collaboration at Yale.
It brings informatics to the clinic and the bedside; innovates new approaches to the analysis of big data across the biomedical research spectrum from basic genetic, proteomic, cellular, and systems biology to the understanding of the social determinants of health; and works in concert with colleagues in data science.
Ohno-Machado was health sciences associate dean for informatics and technology, founding chief of the Division of Biomedical Informatics in the Department of Medicine, and distinguished professor of medicine at the University of California San Diego (UCSD).
She also was founding chair of the UCSD Health Department of Biomedical Informatics and founding faculty of the UCSD Halicioğlu Data Science Institute in La Jolla, California. She received her medical degree from the University of São Paulo, Brazil; her MBA from the Escola de Administração de São Paulo, Fundação Getúlio Vargas, Brazil; and her PhD in medical information sciences and computer science at Stanford University.
She has led informatics centers that were funded by various NIH initiatives and by agencies such as AHRQ, PCORI, and NSF. She organized the first large-scale initiative to share clinical data across five UC medical systems and later extended it to various institutions in California and around the country.
Prior to joining UCSD, she was distinguished chair in biomedical informatics at Brigham and Women’s Hospital, and faculty at Harvard Medical School and at MIT’s Health Sciences and Technology Division.
She is an elected member of the National Academy of Medicine, the American Society for Clinical Investigation, the American Institute for Medical and Biological Engineering, the American College of Medical Informatics, and the International Academy of Health Sciences Informatics.
She is a recipient of the American Medical Informatics Association leadership award, as well as the William W. Stead Award for Thought Leadership in Informatics. Long fascinated by the combination of life science and computer science, Ohno-Machado has conducted research in predictive models and data sharing since the start of her career.
Her doctoral thesis work involved neural network models for survival analysis, and she subsequently focused on new methods to evaluate predictive performance of models based on clinical and molecular data.
Since AI models require large amounts of data, and institutions prefer to keep the data locally, she worked on innovative algorithms to distribute the computation so that data could stay local, but multivariate models could be built and evaluated in a federated manner.