Grant / January 2020

Malpractice, Malice, and Accountability in Machine Learning

As data-driven systems, data science, and machine learning become more prevalent and important, the imperative to govern such technologies grows, too. However, there is currently no accepted definition of best practice, malpractice, or negligence for such systems. This project will study malpractice in data analysis and work to identify the nature of malice in data science, separating it from malpractice and negligence; develop a taxonomy of failure modes for such systems; produce a continuously updated catalog of data-driven system failures; and explore the space of possible attacks and mitigations available to the designers of data-driven systems.