Dataset Bias in Diagnostic AI systems: Guidelines for Dataset Collection and Usage
Published in ACM Conference in Health, Inference, and Learning, Workshop Spotlight, 2020
Recommended citation: J. Vaughn, A. Baral, M. Vadari "Analyzing the Dangers of Dataset Bias in Diagnostic AI systems: Setting Guidelines for Dataset Collection and Usage", ACM Conference on Health, Inference and Learning, 2020 Workshop http://juliev42.github.io/files/CHIL_paper_bias.pdf
In fall of 2019, I helped develop a paper describing a policy framework for addressing bias in medical diagnostic system. The paper combines some of the latest research from CSAIL on algorithmic bias (Suresh et al., 2019), and a deep dive into current FDA practices in regulated artificial intelligence. The paper outlines different kinds of interventions to prevent different varieties of bias that may arise during the data collection process. I am passionate about creating less biased and robust healthcare AI systems, and hope that through sharing this work I can spark conversations about the issue of bias in diagnostic AI. I am continuing to learn more about this area alongside my current Master’s research.