This is an excellent conceptual paper looking at the ML pipeline, the potential risks at each step for introducing bias/inequity/injustice. They make an interesting argument - and provide an great discussion. There are some implicit assumptions - e.g. that communal benefit outweighs concerns regarding individual data privacy concerns - that should be addressed. However, this is one of hte best papers I’ve read setting out a framework for evaluation of healthcare ML from an ethical lens. 2009.10576.pdf (704 KB)