EthicAmanuensis: supporting machine learning practitioners making and recording ethical decisions

Abstract

Ethics should be a practice, not a checkbox. Data scientists want to answer questions about individuals and society using the vast torrent of data that flows around us. Machine learning practitioners want to develop and connect complex models of the world and use them safely in critical situations. Ethical issues can be seen as getting in the way of the core idea and form pain points around managing, using and learning from data, as well as designing human-centric and ethical systems. This is because there is a design gap around ethics in data science and machine learning: the tools that we use do not support ethical data use, which means that data scientists and machine learning practitioners, already engaged in technically complex, multidisciplinary work, must add another dimension to their thinking. This work proposes and outlines an infrastructure and framework that can support in-the-moment ethical decision making and recording, as well as post-hoc audits and ethical model deployment.