The HealthyML has demonstrated that naive application of state-of-the-art techniques like differentially private machine learning cause minority groups to lose predictive influence in health tasks. Such asymmetries in the latent space must be corrected methodologically with methods that distill multi-level knowledge, or deliberately targeted to decorrelate sensitive information from the prediction setting. Models can also be optimized so that explicit fairness constraints are enforced for practical health deployment settings.
Reproducible and ethical machine learning in health are important, along with improved understanding of the bias in that may be present in models learned with medical images, clinical notes, or through processes and devices. Evaluating how clinical experts use the systems in practice is an important part of this effort. Using explainability methods can worsen model performance on minorities in these settings. More work should be done to establish how advice from biased AI can be mitigated by delivery method, for instance by presenting it descriptively rather than prescriptively.
--------------------
Dr. Marzyeh Ghassemi is an Associate Professor at MIT in Electrical Engineering and Computer Science (EECS) and Institute for Medical Engineering & Science (IMES), and a Vector Institute faculty member holding a Canadian CIFAR AI Chair and Canada Research Chair. She holds MIT affiliations with the Jameel Clinic and CSAIL.
Professor Ghassemi holds a Herman L. F. von Helmholtz Career Development Professorship, and was named a CIFAR Azrieli Global Scholar and one of MIT Tech Review’s 35 Innovators Under 35. Previously, she was a Visiting Researcher with Alphabet’s Verily and an Assistant Professor at University of Toronto. Prior to her PhD in Computer Science at MIT, she received an MSc. degree in biomedical engineering from Oxford University as a Marshall Scholar, and B.S. degrees in computer science and electrical engineering as a Goldwater Scholar at New Mexico State University.
Professor Ghassemi has previously served as a NeurIPS Workshop Co-Chair and General Chair for the ACM Conference on Health, Inference and Learning (CHIL). She also founded the non-profit Association for Health Learning and Inference. Professor Ghassemi has published across computer science and clinical venues, including NeurIPS, KDD, AAAI, MLHC, JAMIA, JMIR, JMLR, AMIA-CRI, Nature Medicine, Nature Translational Psychiatry, and Critical Care. Her work has been featured in popular press such as Fortune, MIT News, NVIDIA, and The Huffington Post.
