Tori Ellison
Abstract
Recent studies question the effectiveness of traditional case study methods in teaching data science ethics, particularly in developing moral sensitivity for everyday ethical decisions made by data scientists. These case studies, often in which CEOs or policymakers are the decision-makers, may not resonate with students' experiences. I introduce an interactive microcosm of an innovative, applied, algorithmic value alignment-based approach to teaching data science ethics to undergraduates. This course teaches students how to both embed human values and evaluate the extent to which they are met in predictive models using an optimization-based approach. We introduce cutting-edge algorithms that incorporate human values, such as fairness, explainability, and accuracy, and apply them to real datasets using the AIF360 and AIX360 packages (Python and R). Participants will learn how to code two such algorithms with the AIF360 package and compare the ethical pros and cons of these two algorithms applied to a real dataset.