Examining Axiological Assumptions in Machine Learning Publications

Abstract

This paper presents a study of the values embedded within machine learning research papers. A novel annotation scheme is developed to analyze how values are represented in scholarly documents, focusing on the rationales for research projects, the emphasized attributes of those projects, and the discussion or neglect of potential negative impacts. The methodology is applied to a corpus of influential papers from toptier machine learning conferences. The analysis explores the relationship between these encoded values and factors such as institutional affiliations and funding sources, aiming to contribute to a more nuanced understanding of the ethical dimensions of machine learning research.